1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155 7156 7157 7158 7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280 7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297 7298 7299 7300 7301 7302 7303 7304 7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399 7400 7401 7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472 7473 7474 7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505 7506 7507 7508 7509 7510 7511 7512 7513 7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591 7592 7593 7594 7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682 7683 7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724 7725 7726 7727 7728 7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917 7918 7919 7920 7921 7922 7923 7924 7925 7926 7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962 7963 7964 7965 7966 7967 7968 7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027 8028 8029 8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141 8142 8143 8144 8145 8146 8147 8148 8149 8150 8151 8152 8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184 8185 8186 8187 8188 8189 8190 8191 8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226 8227 8228 8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240 8241 8242 8243 8244 8245 8246 8247 8248 8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278 8279 8280 8281 8282 8283 8284 8285 8286 8287 8288 8289 8290 8291 8292 8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341 8342 8343 8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417 8418 8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496 8497 8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508 8509 8510 8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538 8539 8540 8541 8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554 8555 8556 8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594 8595 8596 8597 8598 8599 8600 8601 8602 8603 8604 8605 8606 8607 8608 8609 8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699 8700 8701 8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715 8716 8717 8718 8719 8720 8721 8722 8723 8724 8725 8726 8727 8728 8729 8730 8731 8732 8733 8734 8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759 8760 8761 8762 8763 8764 8765 8766 8767 8768 8769 8770 8771 8772 8773 8774 8775 8776 8777 8778 8779 8780 8781 8782 8783 8784 8785 8786 8787 8788 8789 8790 8791 8792 8793 8794 8795 8796 8797 8798 8799 8800 8801 8802 8803 8804 8805 8806 8807 8808 8809 8810 8811 8812 8813 8814 8815 8816 8817 8818 8819 8820 8821 8822 8823 8824 8825 8826 8827 8828 8829 8830 8831 8832 8833 8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868 8869 8870 8871 8872 8873 8874 8875 8876 8877 8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891 8892 8893 8894 8895 8896 8897 8898 8899 8900 8901 8902 8903 8904 8905 8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919 8920 8921 8922 8923 8924 8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935 8936 8937 8938 8939 8940 8941 8942 8943 8944 8945 8946 8947 8948 8949 8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972 8973 8974 8975 8976 8977 8978 8979 8980 8981 8982 8983 8984 8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044 9045 9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074 9075 9076 9077 9078 9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095 9096 9097 9098 9099 9100 9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119 9120 9121 9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133 9134 9135 9136 9137 9138 9139 9140 9141 9142 9143 9144 9145 9146 9147 9148 9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160 9161 9162 9163 9164 9165 9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181 9182 9183 9184 9185 9186 9187 9188 9189 9190 9191 9192 9193 9194 9195 9196 9197 9198 9199 9200 9201 9202 9203 9204 9205 9206 9207 9208 9209 9210 9211 9212 9213 9214 9215 9216 9217 9218 9219 9220 9221 9222 9223 9224 9225 9226 9227 9228 9229 9230 9231 9232 9233 9234 9235 9236 9237 9238 9239 9240 9241 9242 9243 9244 9245 9246 9247 9248 9249 9250 9251 9252 9253 9254 9255 9256 9257 9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293 9294 9295 9296 9297 9298 9299 9300 9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322 9323 9324 9325 9326 9327 9328 9329 9330 9331 9332 9333 9334 9335 9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356 9357 9358 9359 9360 9361 9362 9363 9364 9365 9366 9367 9368 9369 9370 9371 9372 9373 9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391 9392 9393 9394 9395 9396 9397 9398 9399 9400 9401 9402 9403 9404 9405 9406 9407 9408 9409 9410 9411 9412 9413 9414 9415 9416 9417 9418 9419 9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430 9431 9432 9433 9434 9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453 9454 9455 9456 9457 9458 9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470 9471 9472 9473 9474 9475 9476 9477 9478 9479 9480 9481 9482 9483 9484 9485 9486 9487 9488 9489 9490 9491 9492 9493 9494 9495 9496 9497 9498 9499 9500 9501 9502 9503 9504 9505 9506 9507 9508 9509 9510 9511 9512 9513 9514 9515 9516 9517 9518 9519 9520 9521 9522 9523 9524 9525 9526 9527 9528 9529 9530 9531 9532 9533 9534 9535 9536 9537 9538 9539 9540 9541 9542 9543 9544 9545 9546 9547 9548 9549 9550 9551 9552 9553 9554 9555 9556 9557 9558 9559 9560 9561 9562 9563 9564 9565 9566 9567 9568 9569 9570 9571 9572 9573 9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587 9588 9589 9590 9591 9592 9593 9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609 9610 9611 9612 9613 9614 9615 9616 9617 9618 9619 9620 9621 9622 9623 9624 9625 9626 9627 9628 9629 9630 9631 9632 9633 9634 9635 9636 9637 9638 9639 9640 9641 9642 9643 9644 9645 9646 9647 9648 9649 9650 9651 9652 9653 9654 9655 9656 9657 9658 9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672 9673 9674 9675 9676 9677 9678 9679 9680 9681 9682 9683 9684 9685 9686 9687 9688 9689 9690 9691 9692 9693 9694 9695 9696 9697 9698 9699 9700 9701 9702 9703 9704 9705 9706 9707 9708 9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728 9729 9730 9731 9732 9733 9734 9735 9736 9737 9738 9739 9740 9741 9742 9743 9744 9745 9746 9747 9748 9749 9750 9751 9752 9753 9754 9755 9756 9757 9758 9759 9760 9761 9762 9763 9764 9765 9766 9767 9768 9769 9770 9771 9772 9773 9774 9775 9776 9777 9778 9779 9780 9781 9782 9783 9784 9785 9786 9787 9788 9789 9790 9791 9792 9793 9794 9795 9796 9797 9798 9799 9800 9801 9802 9803 9804 9805 9806 9807 9808 9809 9810 9811 9812 9813 9814 9815 9816 9817 9818 9819 9820 9821 9822 9823 9824 9825 9826 9827 9828 9829 9830 9831 9832 9833 9834 9835 9836 9837 9838 9839 9840 9841 9842 9843 9844 9845 9846 9847 9848 9849 9850 9851 9852 9853 9854 9855 9856 9857 9858 9859 9860 9861 9862 9863 9864 9865 9866 9867 9868 9869 9870 9871 9872 9873 9874 9875 9876 9877 9878 9879 9880 9881 9882 9883 9884 9885 9886 9887 9888 9889 9890 9891 9892 9893 9894 9895 9896 9897 9898 9899 9900 9901 9902 9903 9904 9905 9906 9907 9908 9909 9910 9911 9912 9913 9914 9915 9916 9917 9918 9919 9920 9921 9922 9923 9924 9925 9926 9927 9928 9929 9930 9931 9932 9933 9934 9935 9936 9937 9938 9939 9940 9941 9942 9943 9944 9945 9946 9947 9948 9949 9950 9951 9952 9953 9954 9955 9956 9957 9958 9959 9960 9961 9962 9963 9964 9965 9966 9967 9968 9969 9970 9971 9972 9973 9974 9975 9976 9977 9978 9979 9980 9981 9982 9983 9984 9985 9986 9987 9988 9989 9990 9991 9992 9993 9994 9995 9996 9997 9998 9999 10000 10001 10002 10003 10004 10005 10006 10007 10008 10009 10010 10011 10012 10013 10014 10015 10016 10017 10018 10019 10020 10021 10022 10023 10024 10025 10026 10027 10028 10029 10030 10031 10032 10033 10034 10035 10036 10037 10038 10039 10040 10041 10042 10043 10044 10045 10046 10047 10048 10049 10050 10051 10052 10053 10054 10055 10056 10057 10058 10059 10060 10061 10062 10063 10064 10065 10066 10067 10068 10069 10070 10071 10072 10073 10074 10075 10076 10077 10078 10079 10080 10081 10082 10083 10084 10085 10086 10087 10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098 10099 10100 10101 10102 10103 10104 10105 10106 10107 10108 10109 10110 10111 10112 10113 10114 10115 10116 10117 10118 10119 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 10130 10131 10132 10133 10134 10135 10136 10137 10138 10139 10140 10141 10142 10143 10144 10145 10146 10147 10148 10149 10150 10151 10152 10153 10154 10155 10156 10157 10158 10159 10160 10161 10162 10163 10164 10165 10166 10167 10168 10169 10170 10171 10172 10173 10174 10175 10176 10177 10178 10179 10180 10181 10182 10183 10184 10185 10186 10187 10188 10189 10190 10191 10192 10193 10194 10195 10196 10197 10198 10199 10200 10201 10202 10203 10204 10205 10206 10207 10208 10209 10210 10211 10212 10213 10214 10215 10216 10217 10218 10219 10220 10221 10222 10223 10224 10225 10226 10227 10228 10229 10230 10231 10232 10233 10234 10235 10236 10237 10238 10239 10240 10241 10242 10243 10244 10245 10246 10247 10248 10249 10250 10251 10252 10253 10254 10255 10256 10257 10258 10259 10260 10261 10262 10263 10264 10265 10266 10267 10268 10269 10270 10271 10272 10273 10274 10275 10276 10277 10278 10279 10280 10281 10282 10283 10284 10285 10286 10287 10288 10289 10290 10291 10292 10293 10294 10295 10296 10297 10298 10299 10300 10301 10302 10303 10304 10305 10306 10307 10308 10309 10310 10311 10312 10313 10314 10315 10316 10317 10318 10319 10320 10321 10322 10323 10324 10325 10326 10327 10328 10329 10330 10331 10332 10333 10334 10335 10336 10337 10338 10339 10340 10341 10342 10343 10344 10345 10346 10347 10348 10349 10350 10351 10352 10353 10354 10355 10356 10357 10358 10359 10360 10361 10362 10363 10364 10365 10366 10367 10368 10369 10370 10371 10372 10373 10374 10375 10376 10377 10378 10379 10380 10381 10382 10383 10384 10385 10386 10387 10388 10389 10390 10391 10392 10393 10394 10395 10396 10397 10398 10399 10400 10401 10402 10403 10404 10405 10406 10407 10408 10409 10410 10411 10412 10413 10414 10415 10416 10417 10418 10419 10420 10421 10422 10423 10424 10425 10426 10427 10428 10429 10430 10431 10432 10433 10434 10435 10436 10437 10438 10439 10440 10441 10442 10443 10444 10445 10446 10447 10448 10449 10450 10451 10452 10453 10454 10455 10456 10457 10458 10459 10460 10461 10462 10463 10464 10465 10466 10467 10468 10469 10470 10471 10472 10473 10474 10475 10476 10477 10478 10479 10480 10481 10482 10483 10484 10485 10486 10487 10488 10489 10490 10491 10492 10493 10494 10495 10496 10497 10498 10499 10500 10501 10502 10503 10504 10505 10506 10507 10508 10509 10510 10511 10512 10513 10514 10515 10516 10517 10518 10519 10520 10521 10522 10523 10524 10525 10526 10527 10528 10529 10530 10531 10532 10533 10534 10535 10536 10537 10538 10539 10540 10541 10542 10543 10544 10545 10546 10547 10548 10549 10550 10551 10552 10553 10554 10555 10556 10557 10558 10559 10560 10561 10562 10563 10564 10565 10566 10567 10568 10569 10570 10571 10572 10573 10574 10575 10576 10577 10578 10579 10580 10581 10582 10583 10584 10585 10586 10587 10588 10589 10590 10591 10592 10593 10594 10595 10596 10597 10598 10599 10600 10601 10602 10603 10604 10605 10606 10607 10608 10609 10610 10611 10612 10613 10614 10615 10616 10617 10618 10619 10620 10621 10622 10623 10624 10625 10626 10627 10628 10629 10630 10631 10632 10633 10634 10635 10636 10637 10638 10639 10640 10641 10642 10643 10644 10645 10646 10647 10648 10649 10650 10651 10652 10653 10654 10655 10656 10657 10658 10659 10660 10661 10662 10663 10664 10665 10666 10667 10668 10669 10670 10671 10672 10673 10674 10675 10676 10677 10678 10679 10680 10681 10682 10683 10684 10685 10686 10687 10688 10689 10690 10691 10692 10693 10694 10695 10696 10697 10698 10699 10700 10701 10702 10703 10704 10705 10706 10707 10708 10709 10710 10711 10712 10713 10714 10715 10716 10717 10718 10719 10720 10721 10722 10723 10724 10725 10726 10727 10728 10729 10730 10731 10732 10733 10734 10735 10736 10737 10738 10739 10740 10741 10742 10743 10744 10745 10746 10747 10748 10749 10750 10751 10752 10753 10754 10755 10756 10757 10758 10759 10760 10761 10762 10763 10764 10765 10766 10767 10768 10769 10770 10771 10772 10773 10774 10775 10776 10777 10778 10779 10780 10781 10782 10783 10784 10785 10786 10787 10788 10789 10790 10791 10792 10793 10794 10795 10796 10797 10798 10799 10800 10801 10802 10803 10804 10805 10806 10807 10808 10809 10810 10811 10812 10813 10814 10815 10816 10817 10818 10819 10820 10821 10822 10823 10824 10825 10826 10827 10828 10829 10830 10831 10832 10833 10834 10835 10836 10837 10838 10839 10840 10841 10842 10843 10844 10845 10846 10847 10848 10849 10850 10851 10852 10853 10854 10855 10856 10857 10858 10859 10860 10861 10862 10863 10864 10865 10866 10867 10868 10869 10870 10871 10872 10873 10874 10875 10876 10877 10878 10879 10880 10881 10882 10883 10884 10885 10886 10887 10888 10889 10890 10891 10892 10893 10894 10895 10896 10897 10898 10899 10900 10901 10902 10903 10904 10905 10906 10907 10908 10909 10910 10911 10912 10913 10914 10915 10916 10917 10918 10919 10920 10921 10922 10923 10924 10925 10926 10927 10928 10929 10930 10931 10932 10933 10934 10935 10936 10937 10938 10939 10940 10941 10942 10943 10944 10945 10946 10947 10948 10949 10950 10951 10952 10953 10954 10955 10956 10957 10958 10959 10960 10961 10962 10963 10964 10965 10966 10967 10968 10969 10970 10971 10972 10973 10974 10975 10976 10977 10978 10979 10980 10981 10982 10983 10984 10985 10986 10987 10988 10989 10990 10991 10992 10993 10994 10995 10996 10997 10998 10999 11000 11001 11002 11003 11004 11005 11006 11007 11008 11009 11010 11011 11012 11013 11014 11015 11016 11017 11018 11019 11020 11021 11022 11023 11024 11025 11026 11027 11028 11029 11030 11031 11032 11033 11034 11035 11036 11037 11038 11039 11040 11041 11042 11043 11044 11045 11046 11047 11048 11049 11050 11051 11052 11053 11054 11055 11056 11057 11058 11059 11060 11061 11062 11063 11064 11065 11066 11067 11068 11069 11070 11071 11072 11073 11074 11075 11076 11077 11078 11079 11080 11081 11082 11083 11084 11085 11086 11087 11088 11089 11090 11091 11092 11093 11094 11095 11096 11097 11098 11099 11100 11101 11102 11103 11104 11105 11106 11107 11108 11109 11110 11111 11112 11113 11114 11115 11116 11117 11118 11119 11120 11121 11122 11123 11124 11125 11126 11127 11128 11129 11130 11131 11132 11133 11134 11135 11136 11137 11138 11139 11140 11141 11142 11143 11144 11145 11146 11147 11148 11149 11150 11151 11152 11153 11154 11155 11156 11157 11158 11159 11160 11161 11162 11163 11164 11165 11166 11167 11168 11169 11170 11171 11172 11173 11174 11175 11176 11177 11178 11179 11180 11181 11182 11183 11184 11185 11186 11187 11188 11189 11190 11191 11192 11193 11194 11195 11196 11197 11198 11199 11200 11201 11202 11203 11204 11205 11206 11207 11208 11209 11210 11211 11212 11213 11214 11215 11216 11217 11218 11219 11220 11221 11222 11223 11224 11225 11226 11227 11228 11229 11230 11231 11232 11233 11234 11235 11236 11237 11238 11239 11240 11241 11242 11243 11244 11245 11246 11247 11248 11249 11250 11251 11252 11253 11254 11255 11256 11257 11258 11259 11260 11261 11262 11263 11264 11265 11266 11267 11268 11269 11270 11271 11272 11273 11274 11275 11276 11277 11278 11279 11280 11281 11282 11283 11284 11285 11286 11287 11288 11289 11290 11291 11292 11293 11294 11295 11296 11297 11298 11299 11300 11301 11302 11303 11304 11305 11306 11307 11308 11309 11310 11311 11312 11313 11314 11315 11316 11317 11318 11319 11320 11321 11322 11323 11324 11325 11326 11327 11328 11329 11330 11331 11332 11333 11334 11335 11336 11337 11338 11339 11340 11341 11342 11343 11344 11345 11346 11347 11348 11349 11350 11351 11352 11353 11354 11355 11356 11357 11358 11359 11360 11361 11362 11363 11364 11365 11366 11367 11368 11369 11370 11371 11372 11373 11374 11375 11376 11377 11378 11379 11380 11381 11382 11383 11384 11385 11386 11387 11388 11389 11390 11391 11392 11393 11394 11395 11396 11397 11398 11399 11400 11401 11402 11403 11404 11405 11406 11407 11408 11409 11410 11411 11412 11413 11414 11415 11416 11417 11418 11419 11420 11421 11422 11423 11424 11425 11426 11427 11428 11429 11430 11431 11432 11433 11434 11435 11436 11437 11438 11439 11440 11441 11442 11443 11444 11445 11446 11447 11448 11449 11450 11451 11452 11453 11454 11455 11456 11457 11458 11459 11460 11461 11462 11463 11464 11465 11466 11467 11468 11469 11470 11471 11472 11473 11474 11475 11476 11477 11478 11479 11480 11481 11482 11483 11484 11485 11486 11487 11488 11489 11490 11491 11492 11493 11494 11495 11496 11497 11498 11499 11500 11501 11502 11503 11504 11505 11506 11507 11508 11509 11510 11511 11512 11513 11514 11515 11516 11517 11518 11519 11520 11521 11522 11523 11524 11525 11526 11527 11528 11529 11530 11531 11532 11533 11534 11535 11536 11537 11538 11539 11540 11541 11542 11543 11544 11545 11546 11547 11548 11549 11550 11551 11552 11553 11554 11555 11556 11557 11558 11559 11560 11561 11562 11563 11564 11565 11566 11567 11568 11569 11570 11571 11572 11573 11574 11575 11576 11577 11578 11579 11580 11581 11582 11583 11584 11585 11586 11587 11588 11589 11590 11591 11592 11593 11594 11595 11596 11597 11598 11599 11600 11601 11602 11603 11604 11605 11606 11607 11608 11609 11610 11611 11612 11613 11614 11615 11616 11617 11618 11619 11620 11621 11622 11623 11624 11625 11626 11627 11628 11629 11630 11631 11632 11633 11634 11635 11636 11637 11638 11639 11640 11641 11642 11643 11644 11645 11646 11647 11648 11649 11650 11651 11652 11653 11654 11655 11656 11657 11658 11659 11660 11661 11662 11663 11664 11665 11666 11667 11668 11669 11670 11671 11672 11673 11674 11675 11676 11677 11678 11679 11680 11681 11682 11683 11684 11685 11686 11687 11688 11689 11690 11691 11692 11693 11694 11695 11696 11697 11698 11699 11700 11701 11702 11703 11704 11705 11706 11707 11708 11709 11710 11711 11712 11713 11714 11715 11716 11717 11718 11719 11720 11721 11722 11723 11724 11725 11726 11727 11728 11729 11730 11731 11732 11733 11734 11735 11736 11737 11738 11739 11740 11741 11742 11743 11744 11745 11746 11747 11748 11749 11750 11751 11752 11753 11754 11755 11756 11757 11758 11759 11760 11761 11762 11763 11764 11765 11766 11767 11768 11769 11770 11771 11772 11773 11774 11775 11776 11777 11778 11779 11780 11781 11782 11783 11784 11785 11786 11787 11788 11789 11790 11791 11792 11793 11794 11795 11796 11797 11798 11799 11800 11801 11802 11803 11804 11805 11806 11807 11808 11809 11810 11811 11812 11813 11814 11815 11816 11817 11818 11819 11820 11821 11822 11823 11824 11825 11826 11827 11828 11829 11830 11831 11832 11833 11834 11835 11836 11837 11838 11839 11840 11841 11842 11843 11844 11845 11846 11847 11848 11849 11850 11851 11852 11853 11854 11855 11856 11857 11858 11859 11860 11861 11862 11863 11864 11865 11866 11867 11868 11869 11870 11871 11872 11873 11874 11875 11876 11877 11878 11879 11880 11881 11882 11883 11884 11885 11886 11887 11888 11889 11890 11891 11892 11893 11894 11895 11896 11897 11898 11899 11900 11901 11902 11903 11904 11905 11906 11907 11908 11909 11910 11911 11912 11913 11914 11915 11916 11917 11918 11919 11920 11921 11922 11923 11924 11925 11926 11927 11928 11929 11930 11931 11932 11933 11934 11935 11936 11937 11938 11939 11940 11941 11942 11943 11944 11945 11946 11947 11948 11949 11950 11951 11952 11953 11954 11955 11956 11957 11958 11959 11960 11961 11962 11963 11964 11965 11966 11967 11968 11969 11970 11971 11972 11973 11974 11975 11976 11977 11978 11979 11980 11981 11982 11983 11984 11985 11986 11987 11988 11989 11990 11991 11992 11993 11994 11995 11996 11997 11998 11999 12000 12001 12002 12003 12004 12005 12006 12007 12008 12009 12010 12011 12012 12013 12014 12015 12016 12017 12018 12019 12020 12021 12022 12023 12024 12025 12026 12027 12028 12029 12030 12031 12032 12033 12034 12035 12036 12037 12038 12039 12040 12041 12042 12043 12044 12045 12046 12047 12048 12049 12050 12051 12052 12053 12054 12055 12056 12057 12058 12059 12060 12061 12062 12063 12064 12065 12066 12067 12068 12069 12070 12071 12072 12073 12074 12075 12076 12077 12078 12079 12080 12081 12082 12083 12084 12085 12086 12087 12088 12089 12090 12091 12092 12093 12094 12095 12096 12097 12098 12099 12100 12101 12102 12103 12104 12105 12106 12107 12108 12109 12110 12111 12112 12113 12114 12115 12116 12117 12118 12119 12120 12121 12122 12123 12124 12125 12126 12127 12128 12129 12130 12131 12132 12133 12134 12135 12136 12137 12138 12139 12140 12141 12142 12143 12144 12145 12146 12147 12148 12149 12150 12151 12152 12153 12154 12155 12156 12157 12158 12159 12160 12161 12162 12163 12164 12165 12166 12167 12168 12169 12170 12171 12172 12173 12174 12175 12176 12177 12178 12179 12180 12181 12182 12183 12184 12185 12186 12187 12188 12189 12190 12191 12192 12193 12194 12195 12196 12197 12198 12199 12200 12201 12202 12203 12204 12205 12206 12207 12208 12209 12210 12211 12212 12213 12214 12215 12216 12217 12218 12219 12220 12221 12222 12223 12224 12225 12226 12227 12228 12229 12230 12231 12232 12233 12234 12235 12236 12237 12238 12239 12240 12241 12242 12243 12244 12245 12246 12247 12248 12249 12250 12251 12252 12253 12254 12255 12256 12257 12258 12259 12260 12261 12262 12263 12264 12265 12266 12267 12268 12269 12270 12271 12272 12273 12274 12275 12276 12277 12278 12279 12280 12281 12282 12283 12284 12285 12286 12287 12288 12289 12290 12291 12292 12293 12294 12295 12296 12297 12298 12299 12300 12301 12302 12303 12304 12305 12306 12307 12308 12309 12310 12311 12312 12313 12314 12315 12316 12317 12318 12319 12320 12321 12322 12323 12324 12325 12326 12327 12328 12329 12330 12331 12332 12333 12334 12335 12336 12337 12338 12339 12340 12341 12342 12343 12344 12345 12346 12347 12348 12349 12350 12351 12352 12353 12354 12355 12356 12357 12358 12359 12360 12361 12362 12363 12364 12365 12366 12367 12368 12369 12370 12371 12372 12373 12374 12375 12376 12377 12378 12379 12380 12381 12382 12383 12384 12385 12386 12387 12388 12389 12390 12391 12392 12393 12394 12395 12396 12397 12398 12399 12400 12401 12402 12403 12404 12405 12406 12407 12408 12409 12410 12411 12412 12413 12414 12415 12416 12417 12418 12419 12420 12421 12422 12423 12424 12425 12426 12427 12428 12429 12430 12431 12432 12433 12434 12435 12436 12437 12438 12439 12440 12441 12442 12443 12444 12445 12446 12447 12448 12449 12450 12451 12452 12453 12454 12455 12456 12457 12458 12459 12460 12461 12462 12463 12464 12465 12466 12467 12468 12469 12470 12471 12472 12473 12474 12475 12476 12477 12478 12479 12480 12481 12482 12483 12484 12485 12486 12487 12488 12489 12490 12491 12492 12493 12494 12495 12496 12497 12498 12499 12500 12501 12502 12503 12504 12505 12506 12507 12508 12509 12510 12511 12512 12513 12514 12515 12516 12517 12518 12519 12520 12521 12522 12523 12524 12525 12526 12527 12528 12529 12530 12531 12532 12533 12534 12535 12536 12537 12538 12539 12540 12541 12542 12543 12544 12545 12546 12547 12548 12549 12550 12551 12552 12553 12554 12555 12556 12557 12558 12559 12560 12561 12562 12563 12564 12565 12566 12567 12568 12569 12570 12571 12572 12573 12574 12575 12576 12577 12578 12579 12580 12581 12582 12583 12584 12585 12586 12587 12588 12589 12590 12591 12592 12593 12594 12595 12596 12597 12598 12599 12600 12601 12602 12603 12604 12605 12606 12607 12608 12609 12610 12611 12612 12613 12614 12615 12616 12617 12618 12619 12620 12621 12622 12623 12624 12625 12626 12627 12628 12629 12630 12631 12632 12633 12634 12635 12636 12637 12638 12639 12640 12641 12642 12643 12644 12645 12646 12647 12648 12649 12650 12651 12652 12653 12654 12655 12656 12657 12658 12659 12660 12661 12662 12663 12664 12665 12666 12667 12668 12669 12670 12671 12672 12673 12674 12675 12676 12677 12678 12679 12680 12681 12682 12683 12684 12685 12686 12687 12688 12689 12690 12691 12692 12693 12694 12695 12696 12697 12698 12699 12700 12701 12702 12703 12704 12705 12706 12707 12708 12709 12710 12711 12712 12713 12714 12715 12716 12717 12718 12719 12720 12721 12722 12723 12724 12725 12726 12727 12728 12729 12730 12731 12732 12733 12734 12735 12736 12737 12738 12739 12740 12741 12742 12743 12744 12745 12746 12747 12748 12749 12750 12751 12752 12753 12754 12755 12756 12757 12758 12759 12760 12761 12762 12763 12764 12765 12766 12767 12768 12769 12770 12771 12772 12773 12774 12775 12776 12777 12778 12779 12780 12781 12782 12783 12784 12785 12786 12787 12788 12789 12790 12791 12792 12793 12794 12795 12796 12797 12798 12799 12800 12801 12802 12803 12804 12805 12806 12807 12808 12809 12810 12811 12812 12813 12814 12815 12816 12817 12818 12819 12820 12821 12822 12823 12824 12825 12826 12827 12828 12829 12830 12831 12832 12833 12834 12835 12836 12837 12838 12839 12840 12841 12842 12843 12844 12845 12846 12847 12848 12849 12850 12851 12852 12853 12854 12855 12856 12857 12858 12859 12860 12861 12862 12863 12864 12865 12866 12867 12868 12869 12870 12871 12872 12873 12874 12875 12876 12877 12878 12879 12880 12881 12882 12883 12884 12885 12886 12887 12888 12889 12890 12891 12892 12893 12894 12895 12896 12897 12898 12899 12900 12901 12902 12903 12904 12905 12906 12907 12908 12909 12910 12911 12912 12913 12914 12915 12916 12917 12918 12919 12920 12921 12922 12923 12924 12925 12926 12927 12928 12929 12930 12931 12932 12933 12934 12935 12936 12937 12938 12939 12940 12941 12942 12943 12944 12945 12946 12947 12948 12949 12950 12951 12952 12953 12954 12955 12956 12957 12958 12959 12960 12961 12962 12963 12964 12965 12966 12967 12968 12969 12970 12971 12972 12973 12974 12975 12976 12977 12978 12979 12980 12981 12982 12983 12984 12985 12986 12987 12988 12989 12990 12991 12992 12993 12994 12995 12996 12997 12998 12999 13000 13001 13002 13003 13004 13005 13006 13007 13008 13009 13010 13011 13012 13013 13014 13015 13016 13017 13018 13019 13020 13021 13022 13023 13024 13025 13026 13027 13028 13029 13030 13031 13032 13033 13034 13035 13036 13037 13038 13039 13040 13041 13042 13043 13044 13045 13046 13047 13048 13049 13050 13051 13052 13053 13054 13055 13056 13057 13058 13059 13060 13061 13062 13063 13064 13065 13066 13067 13068 13069 13070 13071 13072 13073 13074 13075 13076 13077 13078 13079 13080 13081 13082 13083 13084 13085 13086 13087 13088 13089 13090 13091 13092 13093 13094 13095 13096 13097 13098 13099 13100 13101 13102 13103 13104 13105 13106 13107 13108 13109 13110 13111 13112 13113 13114 13115 13116 13117 13118 13119 13120 13121 13122 13123 13124 13125 13126 13127 13128 13129 13130 13131 13132 13133 13134 13135 13136 13137 13138 13139 13140 13141 13142 13143 13144 13145 13146 13147 13148 13149 13150 13151 13152 13153 13154 13155 13156 13157 13158 13159 13160 13161 13162 13163 13164 13165 13166 13167 13168 13169 13170 13171 13172 13173 13174 13175 13176 13177 13178 13179 13180 13181 13182 13183 13184 13185 13186 13187 13188 13189 13190 13191 13192 13193 13194 13195 13196 13197 13198 13199 13200 13201 13202 13203 13204 13205 13206 13207 13208 13209 13210 13211 13212 13213 13214 13215 13216 13217 13218 13219 13220 13221 13222 13223 13224 13225 13226 13227 13228 13229 13230 13231 13232 13233 13234 13235 13236 13237 13238 13239 13240 13241 13242 13243 13244 13245 13246 13247 13248 13249 13250 13251 13252 13253 13254 13255 13256 13257 13258 13259 13260 13261 13262 13263 13264 13265 13266 13267 13268 13269 13270 13271 13272 13273 13274 13275 13276 13277 13278 13279 13280 13281 13282 13283 13284 13285 13286 13287 13288 13289 13290 13291 13292 13293 13294 13295 13296 13297 13298 13299 13300 13301 13302 13303 13304 13305 13306 13307 13308 13309 13310 13311 13312 13313 13314 13315 13316 13317 13318 13319 13320 13321 13322 13323 13324 13325 13326 13327 13328 13329 13330 13331 13332 13333 13334 13335 13336 13337 13338 13339 13340 13341 13342 13343 13344 13345 13346 13347 13348 13349 13350 13351 13352 13353 13354 13355 13356 13357 13358 13359 13360 13361 13362 13363 13364 13365 13366 13367 13368 13369 13370 13371 13372 13373 13374 13375 13376 13377 13378 13379 13380 13381 13382 13383 13384 13385 13386 13387 13388 13389 13390 13391 13392 13393 13394 13395 13396 13397 13398 13399 13400 13401 13402 13403 13404 13405 13406 13407 13408 13409 13410 13411 13412 13413 13414 13415 13416 13417 13418 13419 13420 13421 13422 13423 13424 13425 13426 13427 13428 13429 13430 13431 13432 13433 13434 13435 13436 13437 13438 13439 13440 13441 13442 13443 13444 13445 13446 13447 13448 13449 13450 13451 13452 13453 13454 13455 13456 13457 13458 13459 13460 13461 13462 13463 13464 13465 13466 13467 13468 13469 13470 13471 13472 13473 13474 13475 13476 13477 13478 13479 13480 13481 13482 13483 13484 13485 13486 13487 13488 13489 13490 13491 13492 13493 13494 13495 13496 13497 13498 13499 13500 13501 13502 13503 13504 13505 13506 13507 13508 13509 13510 13511 13512 13513 13514 13515 13516 13517 13518 13519 13520 13521 13522 13523 13524 13525 13526 13527 13528 13529 13530 13531 13532 13533 13534 13535 13536 13537 13538 13539 13540 13541 13542 13543 13544 13545 13546 13547 13548 13549 13550 13551 13552 13553 13554 13555 13556 13557 13558 13559 13560 13561 13562 13563 13564 13565 13566 13567 13568 13569 13570 13571 13572 13573 13574 13575 13576 13577 13578 13579 13580 13581 13582 13583 13584 13585 13586 13587 13588 13589 13590 13591 13592 13593 13594 13595 13596 13597 13598 13599 13600 13601 13602 13603 13604 13605 13606 13607 13608 13609 13610 13611 13612 13613 13614 13615 13616 13617 13618 13619 13620 13621 13622 13623 13624 13625 13626 13627 13628 13629 13630 13631 13632 13633 13634 13635 13636 13637 13638 13639 13640 13641 13642 13643 13644 13645 13646 13647 13648 13649 13650 13651 13652 13653 13654 13655 13656 13657 13658 13659 13660 13661 13662 13663 13664 13665 13666 13667 13668 13669 13670 13671 13672 13673 13674 13675 13676 13677 13678 13679 13680 13681 13682 13683 13684 13685 13686 13687 13688 13689 13690 13691 13692 13693 13694 13695 13696 13697 13698 13699 13700 13701 13702 13703 13704 13705 13706 13707 13708 13709 13710 13711 13712 13713 13714 13715 13716 13717 13718 13719 13720 13721 13722 13723 13724 13725 13726 13727 13728 13729 13730 13731 13732 13733 13734 13735 13736 13737 13738 13739 13740 13741 13742 13743 13744 13745 13746 13747 13748 13749 13750 13751 13752 13753 13754 13755 13756 13757 13758 13759 13760 13761 13762 13763 13764 13765 13766 13767 13768 13769 13770 13771 13772 13773 13774 13775 13776 13777 13778 13779 13780 13781 13782 13783 13784 13785 13786 13787 13788 13789 13790 13791 13792 13793 13794 13795 13796 13797 13798 13799 13800 13801 13802 13803 13804 13805 13806 13807 13808 13809 13810 13811 13812 13813 13814 13815 13816 13817 13818 13819 13820 13821 13822 13823 13824 13825 13826 13827 13828 13829 13830 13831 13832 13833 13834 13835 13836 13837 13838 13839 13840 13841 13842 13843 13844 13845 13846 13847 13848 13849 13850 13851 13852 13853 13854 13855 13856 13857 13858 13859 13860 13861 13862 13863 13864 13865 13866 13867 13868 13869 13870 13871 13872 13873 13874 13875 13876 13877 13878 13879 13880 13881 13882 13883 13884 13885 13886 13887 13888 13889 13890 13891 13892 13893 13894 13895 13896 13897 13898 13899 13900 13901 13902 13903 13904 13905 13906 13907 13908 13909 13910 13911 13912 13913 13914 13915 13916 13917 13918 13919 13920 13921 13922 13923 13924 13925 13926 13927 13928 13929 13930 13931 13932 13933 13934 13935 13936 13937 13938 13939 13940 13941 13942 13943 13944 13945 13946 13947 13948 13949 13950 13951 13952 13953 13954 13955 13956 13957 13958 13959 13960 13961 13962 13963 13964 13965 13966 13967 13968 13969 13970 13971 13972 13973 13974 13975 13976 13977 13978 13979 13980 13981 13982 13983 13984 13985 13986 13987 13988 13989 13990 13991 13992 13993 13994 13995 13996 13997 13998 13999 14000 14001 14002 14003 14004 14005 14006 14007 14008 14009 14010 14011 14012 14013 14014 14015 14016 14017 14018 14019 14020 14021 14022 14023 14024 14025 14026 14027 14028 14029 14030 14031 14032 14033 14034 14035 14036 14037 14038 14039 14040 14041 14042 14043 14044 14045 14046 14047 14048 14049 14050 14051 14052 14053 14054 14055 14056 14057 14058 14059 14060 14061 14062 14063 14064 14065 14066 14067 14068 14069 14070 14071 14072 14073 14074 14075 14076 14077 14078 14079 14080 14081 14082 14083 14084 14085 14086 14087 14088 14089 14090 14091 14092 14093 14094 14095 14096 14097 14098 14099 14100 14101 14102 14103 14104 14105 14106 14107 14108 14109 14110 14111 14112 14113 14114 14115 14116 14117 14118 14119 14120 14121 14122 14123 14124 14125 14126 14127 14128 14129 14130 14131 14132 14133 14134 14135 14136 14137 14138 14139 14140 14141 14142 14143 14144 14145 14146 14147 14148 14149 14150 14151 14152 14153 14154 14155 14156 14157 14158 14159 14160 14161 14162 14163 14164 14165 14166 14167 14168 14169 14170 14171 14172 14173 14174 14175 14176 14177 14178 14179 14180 14181 14182 14183 14184 14185 14186 14187 14188 14189 14190 14191 14192 14193 14194 14195 14196 14197 14198 14199 14200 14201 14202 14203 14204 14205 14206 14207 14208 14209 14210 14211 14212 14213 14214 14215 14216 14217 14218 14219 14220 14221 14222 14223 14224 14225 14226 14227 14228 14229 14230 14231 14232 14233 14234 14235 14236 14237 14238 14239 14240 14241 14242 14243 14244 14245 14246 14247 14248 14249 14250 14251 14252 14253 14254 14255 14256 14257 14258 14259 14260 14261 14262 14263 14264 14265 14266 14267 14268 14269 14270 14271 14272 14273 14274 14275 14276 14277 14278 14279 14280 14281 14282 14283 14284 14285 14286 14287 14288 14289 14290 14291 14292 14293 14294 14295 14296 14297 14298 14299 14300 14301 14302 14303 14304 14305 14306 14307 14308 14309 14310 14311 14312 14313 14314 14315 14316 14317 14318 14319 14320 14321 14322 14323 14324 14325 14326 14327 14328 14329 14330 14331 14332 14333 14334 14335 14336 14337 14338 14339 14340 14341 14342 14343 14344 14345 14346 14347 14348 14349 14350 14351 14352 14353 14354 14355 14356 14357 14358 14359 14360 14361 14362 14363 14364 14365 14366 14367 14368 14369 14370 14371 14372 14373 14374 14375 14376 14377 14378 14379 14380 14381 14382 14383 14384 14385 14386 14387 14388 14389 14390 14391 14392 14393 14394 14395 14396 14397 14398 14399 14400 14401 14402 14403 14404 14405 14406 14407 14408 14409 14410 14411 14412 14413 14414 14415 14416 14417 14418 14419 14420 14421 14422 14423 14424 14425 14426 14427 14428 14429 14430 14431 14432 14433 14434 14435 14436 14437 14438 14439 14440 14441 14442 14443 14444 14445 14446 14447 14448 14449 14450 14451 14452 14453 14454 14455 14456 14457 14458 14459 14460 14461 14462 14463 14464 14465 14466 14467 14468 14469 14470 14471 14472 14473 14474 14475 14476 14477 14478 14479 14480 14481 14482 14483 14484 14485 14486 14487 14488 14489 14490 14491 14492 14493 14494 14495 14496 14497 14498 14499 14500 14501 14502 14503 14504 14505 14506 14507 14508 14509 14510 14511 14512 14513 14514 14515 14516 14517 14518 14519 14520 14521 14522 14523 14524 14525 14526 14527 14528 14529 14530 14531 14532 14533 14534 14535 14536 14537 14538 14539 14540 14541 14542 14543 14544 14545 14546 14547 14548 14549 14550 14551 14552 14553 14554 14555 14556 14557 14558 14559 14560 14561 14562 14563 14564 14565 14566 14567 14568 14569 14570 14571 14572 14573 14574 14575 14576 14577 14578 14579 14580 14581 14582 14583 14584 14585 14586 14587 14588 14589 14590 14591 14592 14593 14594 14595 14596 14597 14598 14599 14600 14601 14602 14603 14604 14605 14606 14607 14608 14609 14610 14611 14612 14613 14614 14615 14616 14617 14618 14619 14620 14621 14622 14623 14624 14625 14626 14627 14628 14629 14630 14631 14632 14633 14634 14635 14636 14637 14638 14639 14640 14641 14642 14643 14644 14645 14646 14647 14648 14649 14650 14651 14652 14653 14654 14655 14656 14657 14658 14659 14660 14661 14662 14663 14664 14665 14666 14667 14668 14669 14670 14671 14672 14673 14674 14675 14676 14677 14678 14679 14680 14681 14682 14683 14684 14685 14686 14687 14688 14689 14690 14691 14692 14693 14694 14695 14696 14697 14698 14699 14700 14701 14702 14703 14704 14705 14706 14707 14708 14709 14710 14711 14712 14713 14714 14715 14716 14717 14718 14719 14720 14721 14722 14723 14724 14725 14726 14727 14728 14729 14730 14731 14732 14733 14734 14735 14736 14737 14738 14739 14740 14741 14742 14743 14744 14745 14746 14747 14748 14749 14750 14751 14752 14753 14754 14755 14756 14757 14758 14759 14760 14761 14762 14763 14764 14765 14766 14767 14768 14769 14770 14771 14772 14773 14774 14775 14776 14777 14778 14779 14780 14781 14782 14783 14784 14785 14786 14787 14788 14789 14790 14791 14792 14793 14794 14795 14796 14797 14798 14799 14800 14801 14802 14803 14804 14805 14806 14807 14808 14809 14810 14811 14812 14813 14814 14815 14816 14817 14818 14819 14820 14821 14822 14823 14824 14825 14826 14827 14828 14829 14830 14831 14832 14833 14834 14835 14836 14837 14838 14839 14840 14841 14842 14843 14844 14845 14846 14847 14848 14849 14850 14851 14852 14853 14854 14855 14856 14857 14858 14859 14860 14861 14862 14863 14864 14865 14866 14867 14868 14869 14870 14871 14872 14873 14874 14875 14876 14877 14878 14879 14880 14881 14882 14883 14884 14885 14886 14887 14888 14889 14890 14891 14892 14893 14894 14895 14896 14897 14898 14899 14900 14901 14902 14903 14904 14905 14906 14907 14908 14909 14910 14911 14912 14913 14914 14915 14916 14917 14918 14919 14920 14921 14922 14923 14924 14925 14926 14927 14928 14929 14930 14931 14932 14933 14934 14935 14936 14937 14938 14939 14940 14941 14942 14943 14944 14945 14946 14947 14948 14949 14950 14951 14952 14953 14954 14955 14956 14957 14958 14959 14960 14961 14962 14963 14964 14965 14966 14967 14968 14969 14970 14971 14972 14973 14974 14975 14976 14977 14978 14979 14980 14981 14982 14983 14984 14985 14986 14987 14988 14989 14990 14991 14992 14993 14994 14995 14996 14997 14998 14999 15000 15001 15002 15003 15004 15005 15006 15007 15008 15009 15010 15011 15012 15013 15014 15015 15016 15017 15018 15019 15020 15021 15022 15023 15024 15025 15026 15027 15028 15029 15030 15031 15032 15033 15034 15035 15036 15037 15038 15039 15040 15041 15042 15043 15044 15045 15046 15047 15048 15049 15050 15051 15052 15053 15054 15055 15056 15057 15058 15059 15060 15061 15062 15063 15064 15065 15066 15067 15068 15069 15070 15071 15072 15073 15074 15075 15076 15077 15078 15079 15080 15081 15082 15083 15084 15085 15086 15087 15088 15089 15090 15091 15092 15093 15094 15095 15096 15097 15098 15099 15100 15101 15102 15103 15104 15105 15106 15107 15108 15109 15110 15111 15112 15113 15114 15115 15116 15117 15118 15119 15120 15121 15122 15123 15124 15125 15126 15127 15128 15129 15130 15131 15132 15133 15134 15135 15136 15137 15138 15139 15140 15141 15142 15143 15144 15145 15146 15147 15148 15149 15150 15151 15152 15153 15154 15155 15156 15157 15158 15159 15160 15161 15162 15163 15164 15165 15166 15167 15168 15169 15170 15171 15172 15173 15174 15175 15176 15177 15178 15179 15180 15181 15182 15183 15184 15185 15186 15187 15188 15189 15190 15191 15192 15193 15194 15195 15196 15197 15198 15199 15200 15201 15202 15203 15204 15205 15206 15207 15208 15209 15210 15211 15212 15213 15214 15215 15216 15217 15218 15219 15220 15221 15222 15223 15224 15225 15226 15227 15228 15229 15230 15231 15232 15233 15234 15235 15236 15237 15238 15239 15240 15241 15242 15243 15244 15245 15246 15247 15248
|
/* Copyright (c) 2000, 2014, Oracle and/or its affiliates.
Copyright (c) 2008, 2014, Monty Program Ab.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */
/*
TODO:
Fix that MAYBE_KEY are stored in the tree so that we can detect use
of full hash keys for queries like:
select s.id, kws.keyword_id from sites as s,kws where s.id=kws.site_id and kws.keyword_id in (204,205);
*/
/*
This file contains:
RangeAnalysisModule
A module that accepts a condition, index (or partitioning) description,
and builds lists of intervals (in index/partitioning space), such that
all possible records that match the condition are contained within the
intervals.
The entry point for the range analysis module is get_mm_tree() function.
The lists are returned in form of complicated structure of interlinked
SEL_TREE/SEL_IMERGE/SEL_ARG objects.
See quick_range_seq_next, find_used_partitions for examples of how to walk
this structure.
All direct "users" of this module are located within this file, too.
PartitionPruningModule
A module that accepts a partitioned table, condition, and finds which
partitions we will need to use in query execution. Search down for
"PartitionPruningModule" for description.
The module has single entry point - prune_partitions() function.
Range/index_merge/groupby-minmax optimizer module
A module that accepts a table, condition, and returns
- a QUICK_*_SELECT object that can be used to retrieve rows that match
the specified condition, or a "no records will match the condition"
statement.
The module entry points are
test_quick_select()
get_quick_select_for_ref()
Record retrieval code for range/index_merge/groupby-min-max.
Implementations of QUICK_*_SELECT classes.
KeyTupleFormat
~~~~~~~~~~~~~~
The code in this file (and elsewhere) makes operations on key value tuples.
Those tuples are stored in the following format:
The tuple is a sequence of key part values. The length of key part value
depends only on its type (and not depends on the what value is stored)
KeyTuple: keypart1-data, keypart2-data, ...
The value of each keypart is stored in the following format:
keypart_data: [isnull_byte] keypart-value-bytes
If a keypart may have a NULL value (key_part->field->real_maybe_null() can
be used to check this), then the first byte is a NULL indicator with the
following valid values:
1 - keypart has NULL value.
0 - keypart has non-NULL value.
<questionable-statement> If isnull_byte==1 (NULL value), then the following
keypart->length bytes must be 0.
</questionable-statement>
keypart-value-bytes holds the value. Its format depends on the field type.
The length of keypart-value-bytes may or may not depend on the value being
stored. The default is that length is static and equal to
KEY_PART_INFO::length.
Key parts with (key_part_flag & HA_BLOB_PART) have length depending of the
value:
keypart-value-bytes: value_length value_bytes
The value_length part itself occupies HA_KEY_BLOB_LENGTH=2 bytes.
See key_copy() and key_restore() for code to move data between index tuple
and table record
CAUTION: the above description is only sergefp's understanding of the
subject and may omit some details.
*/
#ifdef USE_PRAGMA_IMPLEMENTATION
#pragma implementation // gcc: Class implementation
#endif
#include <my_global.h>
#include "sql_priv.h"
#include "key.h" // is_key_used, key_copy, key_cmp, key_restore
#include "sql_parse.h" // check_stack_overrun
#include "sql_partition.h" // get_part_id_func, PARTITION_ITERATOR,
// struct partition_info, NOT_A_PARTITION_ID
#include "sql_base.h" // free_io_cache
#include "records.h" // init_read_record, end_read_record
#include <m_ctype.h>
#include "sql_select.h"
#include "sql_statistics.h"
#include "filesort.h" // filesort_free_buffers
#ifndef EXTRA_DEBUG
#define test_rb_tree(A,B) {}
#define test_use_count(A) {}
#endif
/*
Convert double value to #rows. Currently this does floor(), and we
might consider using round() instead.
*/
#define double2rows(x) ((ha_rows)(x))
static int sel_cmp(Field *f,uchar *a,uchar *b,uint8 a_flag,uint8 b_flag);
/*
this should be long enough so that any memcmp with a string that
starts from '\0' won't cross is_null_string boundaries, even
if the memcmp is optimized to compare 4- 8- or 16- bytes at once
*/
static uchar is_null_string[20]= {1,0};
class RANGE_OPT_PARAM;
/*
A construction block of the SEL_ARG-graph.
The following description only covers graphs of SEL_ARG objects with
sel_arg->type==KEY_RANGE:
One SEL_ARG object represents an "elementary interval" in form
min_value <=? table.keypartX <=? max_value
The interval is a non-empty interval of any kind: with[out] minimum/maximum
bound, [half]open/closed, single-point interval, etc.
1. SEL_ARG GRAPH STRUCTURE
SEL_ARG objects are linked together in a graph. The meaning of the graph
is better demostrated by an example:
tree->keys[i]
|
| $ $
| part=1 $ part=2 $ part=3
| $ $
| +-------+ $ +-------+ $ +--------+
| | kp1<1 |--$-->| kp2=5 |--$-->| kp3=10 |
| +-------+ $ +-------+ $ +--------+
| | $ $ |
| | $ $ +--------+
| | $ $ | kp3=12 |
| | $ $ +--------+
| +-------+ $ $
\->| kp1=2 |--$--------------$-+
+-------+ $ $ | +--------+
| $ $ ==>| kp3=11 |
+-------+ $ $ | +--------+
| kp1=3 |--$--------------$-+ |
+-------+ $ $ +--------+
| $ $ | kp3=14 |
... $ $ +--------+
The entire graph is partitioned into "interval lists".
An interval list is a sequence of ordered disjoint intervals over the same
key part. SEL_ARG are linked via "next" and "prev" pointers. Additionally,
all intervals in the list form an RB-tree, linked via left/right/parent
pointers. The RB-tree root SEL_ARG object will be further called "root of the
interval list".
In the example pic, there are 4 interval lists:
"kp<1 OR kp1=2 OR kp1=3", "kp2=5", "kp3=10 OR kp3=12", "kp3=11 OR kp3=13".
The vertical lines represent SEL_ARG::next/prev pointers.
In an interval list, each member X may have SEL_ARG::next_key_part pointer
pointing to the root of another interval list Y. The pointed interval list
must cover a key part with greater number (i.e. Y->part > X->part).
In the example pic, the next_key_part pointers are represented by
horisontal lines.
2. SEL_ARG GRAPH SEMANTICS
It represents a condition in a special form (we don't have a name for it ATM)
The SEL_ARG::next/prev is "OR", and next_key_part is "AND".
For example, the picture represents the condition in form:
(kp1 < 1 AND kp2=5 AND (kp3=10 OR kp3=12)) OR
(kp1=2 AND (kp3=11 OR kp3=14)) OR
(kp1=3 AND (kp3=11 OR kp3=14))
3. SEL_ARG GRAPH USE
Use get_mm_tree() to construct SEL_ARG graph from WHERE condition.
Then walk the SEL_ARG graph and get a list of dijsoint ordered key
intervals (i.e. intervals in form
(constA1, .., const1_K) < (keypart1,.., keypartK) < (constB1, .., constB_K)
Those intervals can be used to access the index. The uses are in:
- check_quick_select() - Walk the SEL_ARG graph and find an estimate of
how many table records are contained within all
intervals.
- get_quick_select() - Walk the SEL_ARG, materialize the key intervals,
and create QUICK_RANGE_SELECT object that will
read records within these intervals.
4. SPACE COMPLEXITY NOTES
SEL_ARG graph is a representation of an ordered disjoint sequence of
intervals over the ordered set of index tuple values.
For multi-part keys, one can construct a WHERE expression such that its
list of intervals will be of combinatorial size. Here is an example:
(keypart1 IN (1,2, ..., n1)) AND
(keypart2 IN (1,2, ..., n2)) AND
(keypart3 IN (1,2, ..., n3))
For this WHERE clause the list of intervals will have n1*n2*n3 intervals
of form
(keypart1, keypart2, keypart3) = (k1, k2, k3), where 1 <= k{i} <= n{i}
SEL_ARG graph structure aims to reduce the amount of required space by
"sharing" the elementary intervals when possible (the pic at the
beginning of this comment has examples of such sharing). The sharing may
prevent combinatorial blowup:
There are WHERE clauses that have combinatorial-size interval lists but
will be represented by a compact SEL_ARG graph.
Example:
(keypartN IN (1,2, ..., n1)) AND
...
(keypart2 IN (1,2, ..., n2)) AND
(keypart1 IN (1,2, ..., n3))
but not in all cases:
- There are WHERE clauses that do have a compact SEL_ARG-graph
representation but get_mm_tree() and its callees will construct a
graph of combinatorial size.
Example:
(keypart1 IN (1,2, ..., n1)) AND
(keypart2 IN (1,2, ..., n2)) AND
...
(keypartN IN (1,2, ..., n3))
- There are WHERE clauses for which the minimal possible SEL_ARG graph
representation will have combinatorial size.
Example:
By induction: Let's take any interval on some keypart in the middle:
kp15=c0
Then let's AND it with this interval 'structure' from preceding and
following keyparts:
(kp14=c1 AND kp16=c3) OR keypart14=c2) (*)
We will obtain this SEL_ARG graph:
kp14 $ kp15 $ kp16
$ $
+---------+ $ +---------+ $ +---------+
| kp14=c1 |--$-->| kp15=c0 |--$-->| kp16=c3 |
+---------+ $ +---------+ $ +---------+
| $ $
+---------+ $ +---------+ $
| kp14=c2 |--$-->| kp15=c0 | $
+---------+ $ +---------+ $
$ $
Note that we had to duplicate "kp15=c0" and there was no way to avoid
that.
The induction step: AND the obtained expression with another "wrapping"
expression like (*).
When the process ends because of the limit on max. number of keyparts
we'll have:
WHERE clause length is O(3*#max_keyparts)
SEL_ARG graph size is O(2^(#max_keyparts/2))
(it is also possible to construct a case where instead of 2 in 2^n we
have a bigger constant, e.g. 4, and get a graph with 4^(31/2)= 2^31
nodes)
We avoid consuming too much memory by setting a limit on the number of
SEL_ARG object we can construct during one range analysis invocation.
*/
class SEL_ARG :public Sql_alloc
{
public:
uint8 min_flag,max_flag,maybe_flag;
uint8 part; // Which key part
uint8 maybe_null;
/*
The ordinal number the least significant component encountered in
the ranges of the SEL_ARG tree (the first component has number 1)
*/
uint16 max_part_no;
/*
Number of children of this element in the RB-tree, plus 1 for this
element itself.
*/
uint16 elements;
/*
Valid only for elements which are RB-tree roots: Number of times this
RB-tree is referred to (it is referred by SEL_ARG::next_key_part or by
SEL_TREE::keys[i] or by a temporary SEL_ARG* variable)
*/
ulong use_count;
Field *field;
uchar *min_value,*max_value; // Pointer to range
/*
eq_tree() requires that left == right == 0 if the type is MAYBE_KEY.
*/
SEL_ARG *left,*right; /* R-B tree children */
SEL_ARG *next,*prev; /* Links for bi-directional interval list */
SEL_ARG *parent; /* R-B tree parent */
SEL_ARG *next_key_part;
enum leaf_color { BLACK,RED } color;
enum Type { IMPOSSIBLE, MAYBE, MAYBE_KEY, KEY_RANGE } type;
enum { MAX_SEL_ARGS = 16000 };
SEL_ARG() {}
SEL_ARG(SEL_ARG &);
SEL_ARG(Field *,const uchar *, const uchar *);
SEL_ARG(Field *field, uint8 part, uchar *min_value, uchar *max_value,
uint8 min_flag, uint8 max_flag, uint8 maybe_flag);
SEL_ARG(enum Type type_arg)
:min_flag(0), max_part_no(0) /* first key part means 1. 0 mean 'no parts'*/,
elements(1),use_count(1),left(0),right(0),
next_key_part(0), color(BLACK), type(type_arg)
{}
/**
returns true if a range predicate is equal. Use all_same()
to check for equality of all the predicates on this keypart.
*/
inline bool is_same(const SEL_ARG *arg) const
{
if (type != arg->type || part != arg->part)
return false;
if (type != KEY_RANGE)
return true;
return cmp_min_to_min(arg) == 0 && cmp_max_to_max(arg) == 0;
}
/**
returns true if all the predicates in the keypart tree are equal
*/
bool all_same(const SEL_ARG *arg) const
{
if (type != arg->type || part != arg->part)
return false;
if (type != KEY_RANGE)
return true;
if (arg == this)
return true;
const SEL_ARG *cmp_arg= arg->first();
const SEL_ARG *cur_arg= first();
for (; cur_arg && cmp_arg && cur_arg->is_same(cmp_arg);
cur_arg= cur_arg->next, cmp_arg= cmp_arg->next) ;
if (cur_arg || cmp_arg)
return false;
return true;
}
inline void merge_flags(SEL_ARG *arg) { maybe_flag|=arg->maybe_flag; }
inline void maybe_smaller() { maybe_flag=1; }
/* Return true iff it's a single-point null interval */
inline bool is_null_interval() { return maybe_null && max_value[0] == 1; }
inline int cmp_min_to_min(const SEL_ARG* arg) const
{
return sel_cmp(field,min_value, arg->min_value, min_flag, arg->min_flag);
}
inline int cmp_min_to_max(const SEL_ARG* arg) const
{
return sel_cmp(field,min_value, arg->max_value, min_flag, arg->max_flag);
}
inline int cmp_max_to_max(const SEL_ARG* arg) const
{
return sel_cmp(field,max_value, arg->max_value, max_flag, arg->max_flag);
}
inline int cmp_max_to_min(const SEL_ARG* arg) const
{
return sel_cmp(field,max_value, arg->min_value, max_flag, arg->min_flag);
}
SEL_ARG *clone_and(SEL_ARG* arg)
{ // Get overlapping range
uchar *new_min,*new_max;
uint8 flag_min,flag_max;
if (cmp_min_to_min(arg) >= 0)
{
new_min=min_value; flag_min=min_flag;
}
else
{
new_min=arg->min_value; flag_min=arg->min_flag; /* purecov: deadcode */
}
if (cmp_max_to_max(arg) <= 0)
{
new_max=max_value; flag_max=max_flag;
}
else
{
new_max=arg->max_value; flag_max=arg->max_flag;
}
return new SEL_ARG(field, part, new_min, new_max, flag_min, flag_max,
MY_TEST(maybe_flag && arg->maybe_flag));
}
SEL_ARG *clone_first(SEL_ARG *arg)
{ // min <= X < arg->min
return new SEL_ARG(field,part, min_value, arg->min_value,
min_flag, arg->min_flag & NEAR_MIN ? 0 : NEAR_MAX,
maybe_flag | arg->maybe_flag);
}
SEL_ARG *clone_last(SEL_ARG *arg)
{ // min <= X <= key_max
return new SEL_ARG(field, part, min_value, arg->max_value,
min_flag, arg->max_flag, maybe_flag | arg->maybe_flag);
}
SEL_ARG *clone(RANGE_OPT_PARAM *param, SEL_ARG *new_parent, SEL_ARG **next);
bool copy_min(SEL_ARG* arg)
{ // Get overlapping range
if (cmp_min_to_min(arg) > 0)
{
min_value=arg->min_value; min_flag=arg->min_flag;
if ((max_flag & (NO_MAX_RANGE | NO_MIN_RANGE)) ==
(NO_MAX_RANGE | NO_MIN_RANGE))
return 1; // Full range
}
maybe_flag|=arg->maybe_flag;
return 0;
}
bool copy_max(SEL_ARG* arg)
{ // Get overlapping range
if (cmp_max_to_max(arg) <= 0)
{
max_value=arg->max_value; max_flag=arg->max_flag;
if ((max_flag & (NO_MAX_RANGE | NO_MIN_RANGE)) ==
(NO_MAX_RANGE | NO_MIN_RANGE))
return 1; // Full range
}
maybe_flag|=arg->maybe_flag;
return 0;
}
void copy_min_to_min(SEL_ARG *arg)
{
min_value=arg->min_value; min_flag=arg->min_flag;
}
void copy_min_to_max(SEL_ARG *arg)
{
max_value=arg->min_value;
max_flag=arg->min_flag & NEAR_MIN ? 0 : NEAR_MAX;
}
void copy_max_to_min(SEL_ARG *arg)
{
min_value=arg->max_value;
min_flag=arg->max_flag & NEAR_MAX ? 0 : NEAR_MIN;
}
/* returns a number of keypart values (0 or 1) appended to the key buffer */
int store_min(uint length, uchar **min_key,uint min_key_flag)
{
/* "(kp1 > c1) AND (kp2 OP c2) AND ..." -> (kp1 > c1) */
if ((min_flag & GEOM_FLAG) ||
(!(min_flag & NO_MIN_RANGE) &&
!(min_key_flag & (NO_MIN_RANGE | NEAR_MIN))))
{
if (maybe_null && *min_value)
{
**min_key=1;
bzero(*min_key+1,length-1);
}
else
memcpy(*min_key,min_value,length);
(*min_key)+= length;
return 1;
}
return 0;
}
/* returns a number of keypart values (0 or 1) appended to the key buffer */
int store_max(uint length, uchar **max_key, uint max_key_flag)
{
if (!(max_flag & NO_MAX_RANGE) &&
!(max_key_flag & (NO_MAX_RANGE | NEAR_MAX)))
{
if (maybe_null && *max_value)
{
**max_key=1;
bzero(*max_key+1,length-1);
}
else
memcpy(*max_key,max_value,length);
(*max_key)+= length;
return 1;
}
return 0;
}
/*
Returns a number of keypart values appended to the key buffer
for min key and max key. This function is used by both Range
Analysis and Partition pruning. For partition pruning we have
to ensure that we don't store also subpartition fields. Thus
we have to stop at the last partition part and not step into
the subpartition fields. For Range Analysis we set last_part
to MAX_KEY which we should never reach.
*/
int store_min_key(KEY_PART *key,
uchar **range_key,
uint *range_key_flag,
uint last_part)
{
SEL_ARG *key_tree= first();
uint res= key_tree->store_min(key[key_tree->part].store_length,
range_key, *range_key_flag);
*range_key_flag|= key_tree->min_flag;
if (key_tree->next_key_part &&
key_tree->next_key_part->type == SEL_ARG::KEY_RANGE &&
key_tree->part != last_part &&
key_tree->next_key_part->part == key_tree->part+1 &&
!(*range_key_flag & (NO_MIN_RANGE | NEAR_MIN)))
res+= key_tree->next_key_part->store_min_key(key,
range_key,
range_key_flag,
last_part);
return res;
}
/* returns a number of keypart values appended to the key buffer */
int store_max_key(KEY_PART *key,
uchar **range_key,
uint *range_key_flag,
uint last_part)
{
SEL_ARG *key_tree= last();
uint res=key_tree->store_max(key[key_tree->part].store_length,
range_key, *range_key_flag);
(*range_key_flag)|= key_tree->max_flag;
if (key_tree->next_key_part &&
key_tree->next_key_part->type == SEL_ARG::KEY_RANGE &&
key_tree->part != last_part &&
key_tree->next_key_part->part == key_tree->part+1 &&
!(*range_key_flag & (NO_MAX_RANGE | NEAR_MAX)))
res+= key_tree->next_key_part->store_max_key(key,
range_key,
range_key_flag,
last_part);
return res;
}
SEL_ARG *insert(SEL_ARG *key);
SEL_ARG *tree_delete(SEL_ARG *key);
SEL_ARG *find_range(SEL_ARG *key);
SEL_ARG *rb_insert(SEL_ARG *leaf);
friend SEL_ARG *rb_delete_fixup(SEL_ARG *root,SEL_ARG *key, SEL_ARG *par);
#ifdef EXTRA_DEBUG
friend int test_rb_tree(SEL_ARG *element,SEL_ARG *parent);
void test_use_count(SEL_ARG *root);
#endif
SEL_ARG *first();
const SEL_ARG *first() const;
SEL_ARG *last();
void make_root();
inline bool simple_key()
{
return !next_key_part && elements == 1;
}
void increment_use_count(long count)
{
if (next_key_part)
{
next_key_part->use_count+=count;
count*= (next_key_part->use_count-count);
for (SEL_ARG *pos=next_key_part->first(); pos ; pos=pos->next)
if (pos->next_key_part)
pos->increment_use_count(count);
}
}
void incr_refs()
{
increment_use_count(1);
use_count++;
}
void incr_refs_all()
{
for (SEL_ARG *pos=first(); pos ; pos=pos->next)
{
pos->increment_use_count(1);
}
use_count++;
}
void free_tree()
{
for (SEL_ARG *pos=first(); pos ; pos=pos->next)
if (pos->next_key_part)
{
pos->next_key_part->use_count--;
pos->next_key_part->free_tree();
}
}
inline SEL_ARG **parent_ptr()
{
return parent->left == this ? &parent->left : &parent->right;
}
/*
Check if this SEL_ARG object represents a single-point interval
SYNOPSIS
is_singlepoint()
DESCRIPTION
Check if this SEL_ARG object (not tree) represents a single-point
interval, i.e. if it represents a "keypart = const" or
"keypart IS NULL".
RETURN
TRUE This SEL_ARG object represents a singlepoint interval
FALSE Otherwise
*/
bool is_singlepoint()
{
/*
Check for NEAR_MIN ("strictly less") and NO_MIN_RANGE (-inf < field)
flags, and the same for right edge.
*/
if (min_flag || max_flag)
return FALSE;
uchar *min_val= min_value;
uchar *max_val= max_value;
if (maybe_null)
{
/* First byte is a NULL value indicator */
if (*min_val != *max_val)
return FALSE;
if (*min_val)
return TRUE; /* This "x IS NULL" */
min_val++;
max_val++;
}
return !field->key_cmp(min_val, max_val);
}
SEL_ARG *clone_tree(RANGE_OPT_PARAM *param);
};
/**
Helper function to compare two SEL_ARG's.
*/
static bool all_same(const SEL_ARG *sa1, const SEL_ARG *sa2)
{
if (sa1 == NULL && sa2 == NULL)
return true;
if ((sa1 != NULL && sa2 == NULL) || (sa1 == NULL && sa2 != NULL))
return false;
return sa1->all_same(sa2);
}
class SEL_IMERGE;
#define CLONE_KEY1_MAYBE 1
#define CLONE_KEY2_MAYBE 2
#define swap_clone_flag(A) ((A & 1) << 1) | ((A & 2) >> 1)
/*
While objects of the class SEL_ARG represent ranges for indexes or
index infixes (including ranges for index prefixes and index suffixes),
objects of the class SEL_TREE represent AND/OR formulas of such ranges.
Currently an AND/OR formula represented by a SEL_TREE object can have
at most three levels:
<SEL_TREE formula> ::=
[ <SEL_RANGE_TREE formula> AND ]
[ <SEL_IMERGE formula> [ AND <SEL_IMERGE formula> ...] ]
<SEL_RANGE_TREE formula> ::=
<SEL_ARG formula> [ AND <SEL_ARG_formula> ... ]
<SEL_IMERGE formula> ::=
<SEL_RANGE_TREE formula> [ OR <SEL_RANGE_TREE formula> ]
As we can see from the above definitions:
- SEL_RANGE_TREE formula is a conjunction of SEL_ARG formulas
- SEL_IMERGE formula is a disjunction of SEL_RANGE_TREE formulas
- SEL_TREE formula is a conjunction of a SEL_RANGE_TREE formula
and SEL_IMERGE formulas.
It's required above that a SEL_TREE formula has at least one conjunct.
Usually we will consider normalized SEL_RANGE_TREE formulas where we use
TRUE as conjunct members for those indexes whose SEL_ARG trees are empty.
We will call an SEL_TREE object simply 'tree'.
The part of a tree that represents SEL_RANGE_TREE formula is called
'range part' of the tree while the remaining part is called 'imerge part'.
If a tree contains only a range part then we call such a tree 'range tree'.
Components of a range tree that represent SEL_ARG formulas are called ranges.
If a tree does not contain any range part we call such a tree 'imerge tree'.
Components of the imerge part of a tree that represent SEL_IMERGE formula
are called imerges.
Usually we'll designate:
SEL_TREE formulas by T_1,...,T_k
SEL_ARG formulas by R_1,...,R_k
SEL_RANGE_TREE formulas by RT_1,...,RT_k
SEL_IMERGE formulas by M_1,...,M_k
Accordingly we'll use:
t_1,...,t_k - to designate trees representing T_1,...,T_k
r_1,...,r_k - to designate ranges representing R_1,...,R_k
rt_1,...,r_tk - to designate range trees representing RT_1,...,RT_k
m_1,...,m_k - to designate imerges representing M_1,...,M_k
SEL_TREE objects are usually built from WHERE conditions or
ON expressions.
A SEL_TREE object always represents an inference of the condition it is
built from. Therefore, if a row satisfies a SEL_TREE formula it also
satisfies the condition it is built from.
The following transformations of tree t representing SEL_TREE formula T
yield a new tree t1 thar represents an inference of T: T=>T1.
(1) remove any of SEL_ARG tree from the range part of t
(2) remove any imerge from the tree t
(3) remove any of SEL_ARG tree from any range tree contained
in any imerge of tree
Since the basic blocks of any SEL_TREE objects are ranges, SEL_TREE
objects in many cases can be effectively used to filter out a big part
of table rows that do not satisfy WHERE/IN conditions utilizing
only single or multiple range index scans.
A single range index scan is constructed for a range tree that contains
only one SEL_ARG object for an index or an index prefix.
An index intersection scan can be constructed for a range tree
that contains several SEL_ARG objects. Currently index intersection
scans are constructed only for single-point ranges.
An index merge scan is constructed for a imerge tree that contains only
one imerge. If range trees of this imerge contain only single-point merges
than a union of index intersections can be built.
Usually the tree built by the range optimizer for a query table contains
more than one range in the range part, and additionally may contain some
imerges in the imerge part. The range optimizer evaluates all of them one
by one and chooses the range or the imerge that provides the cheapest
single or multiple range index scan of the table. According to rules
(1)-(3) this scan always filter out only those rows that do not satisfy
the query conditions.
For any condition the SEL_TREE object for it is built in a bottom up
manner starting from the range trees for the predicates. The tree_and
function builds a tree for any conjunction of formulas from the trees
for its conjuncts. The tree_or function builds a tree for any disjunction
of formulas from the trees for its disjuncts.
*/
class SEL_TREE :public Sql_alloc
{
public:
/*
Starting an effort to document this field:
(for some i, keys[i]->type == SEL_ARG::IMPOSSIBLE) =>
(type == SEL_TREE::IMPOSSIBLE)
*/
enum Type { IMPOSSIBLE, ALWAYS, MAYBE, KEY, KEY_SMALLER } type;
SEL_TREE(enum Type type_arg) :type(type_arg) {}
SEL_TREE() :type(KEY)
{
keys_map.clear_all();
bzero((char*) keys,sizeof(keys));
}
SEL_TREE(SEL_TREE *arg, bool without_merges, RANGE_OPT_PARAM *param);
/*
Note: there may exist SEL_TREE objects with sel_tree->type=KEY and
keys[i]=0 for all i. (SergeyP: it is not clear whether there is any
merit in range analyzer functions (e.g. get_mm_parts) returning a
pointer to such SEL_TREE instead of NULL)
*/
SEL_ARG *keys[MAX_KEY];
key_map keys_map; /* bitmask of non-NULL elements in keys */
/*
Possible ways to read rows using index_merge. The list is non-empty only
if type==KEY. Currently can be non empty only if keys_map.is_clear_all().
*/
List<SEL_IMERGE> merges;
/* The members below are filled/used only after get_mm_tree is done */
key_map ror_scans_map; /* bitmask of ROR scan-able elements in keys */
uint n_ror_scans; /* number of set bits in ror_scans_map */
struct st_index_scan_info **index_scans; /* list of index scans */
struct st_index_scan_info **index_scans_end; /* last index scan */
struct st_ror_scan_info **ror_scans; /* list of ROR key scans */
struct st_ror_scan_info **ror_scans_end; /* last ROR scan */
/* Note that #records for each key scan is stored in table->quick_rows */
bool without_ranges() { return keys_map.is_clear_all(); }
bool without_imerges() { return merges.is_empty(); }
};
class RANGE_OPT_PARAM
{
public:
THD *thd; /* Current thread handle */
TABLE *table; /* Table being analyzed */
COND *cond; /* Used inside get_mm_tree(). */
table_map prev_tables;
table_map read_tables;
table_map current_table; /* Bit of the table being analyzed */
/* Array of parts of all keys for which range analysis is performed */
KEY_PART *key_parts;
KEY_PART *key_parts_end;
MEM_ROOT *mem_root; /* Memory that will be freed when range analysis completes */
MEM_ROOT *old_root; /* Memory that will last until the query end */
/*
Number of indexes used in range analysis (In SEL_TREE::keys only first
#keys elements are not empty)
*/
uint keys;
/*
If true, the index descriptions describe real indexes (and it is ok to
call field->optimize_range(real_keynr[...], ...).
Otherwise index description describes fake indexes.
*/
bool using_real_indexes;
/*
Aggressively remove "scans" that do not have conditions on first
keyparts. Such scans are usable when doing partition pruning but not
regular range optimization.
*/
bool remove_jump_scans;
/*
used_key_no -> table_key_no translation table. Only makes sense if
using_real_indexes==TRUE
*/
uint real_keynr[MAX_KEY];
/*
Used to store 'current key tuples', in both range analysis and
partitioning (list) analysis
*/
uchar min_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH],
max_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH];
/* Number of SEL_ARG objects allocated by SEL_ARG::clone_tree operations */
uint alloced_sel_args;
bool force_default_mrr;
KEY_PART *key[MAX_KEY]; /* First key parts of keys used in the query */
bool statement_should_be_aborted() const
{
return
thd->is_fatal_error ||
thd->is_error() ||
alloced_sel_args > SEL_ARG::MAX_SEL_ARGS;
}
};
class PARAM : public RANGE_OPT_PARAM
{
public:
ha_rows quick_rows[MAX_KEY];
/*
This will collect 'possible keys' based on the range optimization.
Queries with a JOIN object actually use ref optimizer (see add_key_field)
to collect possible_keys. This is used by single table UPDATE/DELETE.
*/
key_map possible_keys;
longlong baseflag;
uint max_key_part, range_count;
bool quick; // Don't calulate possible keys
uint fields_bitmap_size;
MY_BITMAP needed_fields; /* bitmask of fields needed by the query */
MY_BITMAP tmp_covered_fields;
key_map *needed_reg; /* ptr to SQL_SELECT::needed_reg */
uint *imerge_cost_buff; /* buffer for index_merge cost estimates */
uint imerge_cost_buff_size; /* size of the buffer */
/* TRUE if last checked tree->key can be used for ROR-scan */
bool is_ror_scan;
/* Number of ranges in the last checked tree->key */
uint n_ranges;
uint8 first_null_comp; /* first null component if any, 0 - otherwise */
};
class TABLE_READ_PLAN;
class TRP_RANGE;
class TRP_ROR_INTERSECT;
class TRP_ROR_UNION;
class TRP_INDEX_INTERSECT;
class TRP_INDEX_MERGE;
class TRP_GROUP_MIN_MAX;
struct st_index_scan_info;
struct st_ror_scan_info;
static SEL_TREE * get_mm_parts(RANGE_OPT_PARAM *param,COND *cond_func,Field *field,
Item_func::Functype type,Item *value,
Item_result cmp_type);
static SEL_ARG *get_mm_leaf(RANGE_OPT_PARAM *param,COND *cond_func,Field *field,
KEY_PART *key_part,
Item_func::Functype type,Item *value);
static SEL_TREE *get_mm_tree(RANGE_OPT_PARAM *param,COND *cond);
static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts);
static ha_rows check_quick_select(PARAM *param, uint idx, bool index_only,
SEL_ARG *tree, bool update_tbl_stats,
uint *mrr_flags, uint *bufsize,
Cost_estimate *cost);
QUICK_RANGE_SELECT *get_quick_select(PARAM *param,uint index,
SEL_ARG *key_tree, uint mrr_flags,
uint mrr_buf_size, MEM_ROOT *alloc);
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
bool index_read_must_be_used,
bool update_tbl_stats,
double read_time);
static
TRP_INDEX_INTERSECT *get_best_index_intersect(PARAM *param, SEL_TREE *tree,
double read_time);
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
double read_time,
bool *are_all_covering);
static
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
SEL_TREE *tree,
double read_time);
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
double read_time);
static
TABLE_READ_PLAN *merge_same_index_scans(PARAM *param, SEL_IMERGE *imerge,
TRP_INDEX_MERGE *imerge_trp,
double read_time);
static
TRP_GROUP_MIN_MAX *get_best_group_min_max(PARAM *param, SEL_TREE *tree,
double read_time);
#ifndef DBUG_OFF
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
const char *msg);
static void print_ror_scans_arr(TABLE *table, const char *msg,
struct st_ror_scan_info **start,
struct st_ror_scan_info **end);
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg);
#endif
static SEL_TREE *tree_and(RANGE_OPT_PARAM *param,
SEL_TREE *tree1, SEL_TREE *tree2);
static SEL_TREE *tree_or(RANGE_OPT_PARAM *param,
SEL_TREE *tree1,SEL_TREE *tree2);
static SEL_ARG *sel_add(SEL_ARG *key1,SEL_ARG *key2);
static SEL_ARG *key_or(RANGE_OPT_PARAM *param,
SEL_ARG *key1, SEL_ARG *key2);
static SEL_ARG *key_and(RANGE_OPT_PARAM *param,
SEL_ARG *key1, SEL_ARG *key2,
uint clone_flag);
static bool get_range(SEL_ARG **e1,SEL_ARG **e2,SEL_ARG *root1);
bool get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
SEL_ARG *key_tree, uchar *min_key,uint min_key_flag,
uchar *max_key,uint max_key_flag);
static bool eq_tree(SEL_ARG* a,SEL_ARG *b);
static SEL_ARG null_element(SEL_ARG::IMPOSSIBLE);
static bool null_part_in_key(KEY_PART *key_part, const uchar *key,
uint length);
static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts);
#include "opt_range_mrr.cc"
static bool sel_trees_have_common_keys(SEL_TREE *tree1, SEL_TREE *tree2,
key_map *common_keys);
static void eliminate_single_tree_imerges(RANGE_OPT_PARAM *param,
SEL_TREE *tree);
static bool sel_trees_can_be_ored(RANGE_OPT_PARAM* param,
SEL_TREE *tree1, SEL_TREE *tree2,
key_map *common_keys);
static bool sel_trees_must_be_ored(RANGE_OPT_PARAM* param,
SEL_TREE *tree1, SEL_TREE *tree2,
key_map common_keys);
static int and_range_trees(RANGE_OPT_PARAM *param,
SEL_TREE *tree1, SEL_TREE *tree2,
SEL_TREE *result);
static bool remove_nonrange_trees(RANGE_OPT_PARAM *param, SEL_TREE *tree);
/*
SEL_IMERGE is a list of possible ways to do index merge, i.e. it is
a condition in the following form:
(t_1||t_2||...||t_N) && (next)
where all t_i are SEL_TREEs, next is another SEL_IMERGE and no pair
(t_i,t_j) contains SEL_ARGS for the same index.
SEL_TREE contained in SEL_IMERGE always has merges=NULL.
This class relies on memory manager to do the cleanup.
*/
class SEL_IMERGE : public Sql_alloc
{
enum { PREALLOCED_TREES= 10};
public:
SEL_TREE *trees_prealloced[PREALLOCED_TREES];
SEL_TREE **trees; /* trees used to do index_merge */
SEL_TREE **trees_next; /* last of these trees */
SEL_TREE **trees_end; /* end of allocated space */
SEL_ARG ***best_keys; /* best keys to read in SEL_TREEs */
SEL_IMERGE() :
trees(&trees_prealloced[0]),
trees_next(trees),
trees_end(trees + PREALLOCED_TREES)
{}
SEL_IMERGE (SEL_IMERGE *arg, uint cnt, RANGE_OPT_PARAM *param);
int or_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree);
bool have_common_keys(RANGE_OPT_PARAM *param, SEL_TREE *tree);
int and_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree,
SEL_IMERGE *new_imerge);
int or_sel_tree_with_checks(RANGE_OPT_PARAM *param,
uint n_init_trees,
SEL_TREE *new_tree,
bool is_first_check_pass,
bool *is_last_check_pass);
int or_sel_imerge_with_checks(RANGE_OPT_PARAM *param,
uint n_init_trees,
SEL_IMERGE* imerge,
bool is_first_check_pass,
bool *is_last_check_pass);
};
/*
Add a range tree to the range trees of this imerge
SYNOPSIS
or_sel_tree()
param Context info for the operation
tree SEL_TREE to add to this imerge
DESCRIPTION
The function just adds the range tree 'tree' to the range trees
of this imerge.
RETURN
0 if the operation is success
-1 if the function runs out memory
*/
int SEL_IMERGE::or_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree)
{
if (trees_next == trees_end)
{
const int realloc_ratio= 2; /* Double size for next round */
uint old_elements= (trees_end - trees);
uint old_size= sizeof(SEL_TREE**) * old_elements;
uint new_size= old_size * realloc_ratio;
SEL_TREE **new_trees;
if (!(new_trees= (SEL_TREE**)alloc_root(param->mem_root, new_size)))
return -1;
memcpy(new_trees, trees, old_size);
trees= new_trees;
trees_next= trees + old_elements;
trees_end= trees + old_elements * realloc_ratio;
}
*(trees_next++)= tree;
return 0;
}
/*
Check if any of the range trees of this imerge intersects with a given tree
SYNOPSIS
have_common_keys()
param Context info for the function
tree SEL_TREE intersection with the imerge range trees is checked for
DESCRIPTION
The function checks whether there is any range tree rt_i in this imerge
such that there are some indexes for which ranges are defined in both
rt_i and the range part of the SEL_TREE tree.
To check this the function calls the function sel_trees_have_common_keys.
RETURN
TRUE if there are such range trees in this imerge
FALSE otherwise
*/
bool SEL_IMERGE::have_common_keys(RANGE_OPT_PARAM *param, SEL_TREE *tree)
{
for (SEL_TREE** or_tree= trees, **bound= trees_next;
or_tree != bound; or_tree++)
{
key_map common_keys;
if (sel_trees_have_common_keys(*or_tree, tree, &common_keys))
return TRUE;
}
return FALSE;
}
/*
Perform AND operation for this imerge and the range part of a tree
SYNOPSIS
and_sel_tree()
param Context info for the operation
tree SEL_TREE for the second operand of the operation
new_imerge OUT imerge for the result of the operation
DESCRIPTION
This function performs AND operation for this imerge m and the
range part of the SEL_TREE tree rt. In other words the function
pushes rt into this imerge. The resulting imerge is returned in
the parameter new_imerge.
If this imerge m represent the formula
RT_1 OR ... OR RT_k
then the resulting imerge of the function represents the formula
(RT_1 AND RT) OR ... OR (RT_k AND RT)
The function calls the function and_range_trees to construct the
range tree representing (RT_i AND RT).
NOTE
The function may return an empty imerge without any range trees.
This happens when each call of and_range_trees returns an
impossible range tree (SEL_TREE::IMPOSSIBLE).
Example: (key1 < 2 AND key2 > 10) AND (key1 > 4 OR key2 < 6).
RETURN
0 if the operation is a success
-1 otherwise: there is not enough memory to perform the operation
*/
int SEL_IMERGE::and_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree,
SEL_IMERGE *new_imerge)
{
for (SEL_TREE** or_tree= trees; or_tree != trees_next; or_tree++)
{
SEL_TREE *res_or_tree= 0;
SEL_TREE *and_tree= 0;
if (!(res_or_tree= new SEL_TREE()) ||
!(and_tree= new SEL_TREE(tree, TRUE, param)))
return (-1);
if (!and_range_trees(param, *or_tree, and_tree, res_or_tree))
{
if (new_imerge->or_sel_tree(param, res_or_tree))
return (-1);
}
}
return 0;
}
/*
Perform OR operation on this imerge and the range part of a tree
SYNOPSIS
or_sel_tree_with_checks()
param Context info for the operation
n_trees Number of trees in this imerge to check for oring
tree SEL_TREE whose range part is to be ored
is_first_check_pass <=> the first call of the function for this imerge
is_last_check_pass OUT <=> no more calls of the function for this imerge
DESCRIPTION
The function performs OR operation on this imerge m and the range part
of the SEL_TREE tree rt. It always replaces this imerge with the result
of the operation.
The operation can be performed in two different modes: with
is_first_check_pass==TRUE and is_first_check_pass==FALSE, transforming
this imerge differently.
Given this imerge represents the formula
RT_1 OR ... OR RT_k:
1. In the first mode, when is_first_check_pass==TRUE :
1.1. If rt must be ored(see the function sel_trees_must_be_ored) with
some rt_j (there may be only one such range tree in the imerge)
then the function produces an imerge representing the formula
RT_1 OR ... OR (RT_j OR RT) OR ... OR RT_k,
where the tree for (RT_j OR RT) is built by oring the pairs
of SEL_ARG trees for the corresponding indexes
1.2. Otherwise the function produces the imerge representing the formula:
RT_1 OR ... OR RT_k OR RT.
2. In the second mode, when is_first_check_pass==FALSE :
2.1. For each rt_j in the imerge that can be ored (see the function
sel_trees_can_be_ored) with rt the function replaces rt_j for a
range tree such that for each index for which ranges are defined
in both in rt_j and rt the tree contains the result of oring of
these ranges.
2.2. In other cases the function does not produce any imerge.
When is_first_check==TRUE the function returns FALSE in the parameter
is_last_check_pass if there is no rt_j such that rt_j can be ored with rt,
but, at the same time, it's not true that rt_j must be ored with rt.
When is_first_check==FALSE the function always returns FALSE in the
parameter is_last_check_pass.
RETURN
1 The result of oring of rt_j and rt that must be ored returns the
the range tree with type==SEL_TREE::ALWAYS
(in this case the imerge m should be discarded)
-1 The function runs out of memory
0 in all other cases
*/
int SEL_IMERGE::or_sel_tree_with_checks(RANGE_OPT_PARAM *param,
uint n_trees,
SEL_TREE *tree,
bool is_first_check_pass,
bool *is_last_check_pass)
{
bool was_ored= FALSE;
*is_last_check_pass= is_first_check_pass;
SEL_TREE** or_tree = trees;
for (uint i= 0; i < n_trees; i++, or_tree++)
{
SEL_TREE *result= 0;
key_map result_keys;
key_map ored_keys;
if (sel_trees_can_be_ored(param, *or_tree, tree, &ored_keys))
{
bool must_be_ored= sel_trees_must_be_ored(param, *or_tree, tree,
ored_keys);
if (must_be_ored || !is_first_check_pass)
{
result_keys.clear_all();
result= *or_tree;
for (uint key_no= 0; key_no < param->keys; key_no++)
{
if (!ored_keys.is_set(key_no))
{
result->keys[key_no]= 0;
continue;
}
SEL_ARG *key1= (*or_tree)->keys[key_no];
SEL_ARG *key2= tree->keys[key_no];
key2->incr_refs();
if ((result->keys[key_no]= key_or(param, key1, key2)))
{
result_keys.set_bit(key_no);
#ifdef EXTRA_DEBUG
if (param->alloced_sel_args < SEL_ARG::MAX_SEL_ARGS)
{
key1= result->keys[key_no];
(key1)->test_use_count(key1);
}
#endif
}
}
}
else if(is_first_check_pass)
*is_last_check_pass= FALSE;
}
if (result)
{
result->keys_map= result_keys;
if (result_keys.is_clear_all())
result->type= SEL_TREE::ALWAYS;
if ((result->type == SEL_TREE::MAYBE) ||
(result->type == SEL_TREE::ALWAYS))
return 1;
/* SEL_TREE::IMPOSSIBLE is impossible here */
*or_tree= result;
was_ored= TRUE;
}
}
if (was_ored)
return 0;
if (is_first_check_pass && !*is_last_check_pass &&
!(tree= new SEL_TREE(tree, FALSE, param)))
return (-1);
return or_sel_tree(param, tree);
}
/*
Perform OR operation on this imerge and and another imerge
SYNOPSIS
or_sel_imerge_with_checks()
param Context info for the operation
n_trees Number of trees in this imerge to check for oring
imerge The second operand of the operation
is_first_check_pass <=> the first call of the function for this imerge
is_last_check_pass OUT <=> no more calls of the function for this imerge
DESCRIPTION
For each range tree rt from 'imerge' the function calls the method
SEL_IMERGE::or_sel_tree_with_checks that performs OR operation on this
SEL_IMERGE object m and the tree rt. The mode of the operation is
specified by the parameter is_first_check_pass. Each call of
SEL_IMERGE::or_sel_tree_with_checks transforms this SEL_IMERGE object m.
The function returns FALSE in the prameter is_last_check_pass if
at least one of the calls of SEL_IMERGE::or_sel_tree_with_checks
returns FALSE as the value of its last parameter.
RETURN
1 One of the calls of SEL_IMERGE::or_sel_tree_with_checks returns 1.
(in this case the imerge m should be discarded)
-1 The function runs out of memory
0 in all other cases
*/
int SEL_IMERGE::or_sel_imerge_with_checks(RANGE_OPT_PARAM *param,
uint n_trees,
SEL_IMERGE* imerge,
bool is_first_check_pass,
bool *is_last_check_pass)
{
*is_last_check_pass= TRUE;
SEL_TREE** tree= imerge->trees;
SEL_TREE** tree_end= imerge->trees_next;
for ( ; tree < tree_end; tree++)
{
uint rc;
bool is_last= TRUE;
rc= or_sel_tree_with_checks(param, n_trees, *tree,
is_first_check_pass, &is_last);
if (!is_last)
*is_last_check_pass= FALSE;
if (rc)
return rc;
}
return 0;
}
/*
Copy constructor for SEL_TREE objects
SYNOPSIS
SEL_TREE
arg The source tree for the constructor
without_merges <=> only the range part of the tree arg is copied
param Context info for the operation
DESCRIPTION
The constructor creates a full copy of the SEL_TREE arg if
the prameter without_merges==FALSE. Otherwise a tree is created
that contains the copy only of the range part of the tree arg.
*/
SEL_TREE::SEL_TREE(SEL_TREE *arg, bool without_merges,
RANGE_OPT_PARAM *param): Sql_alloc()
{
keys_map= arg->keys_map;
type= arg->type;
for (uint idx= 0; idx < param->keys; idx++)
{
if ((keys[idx]= arg->keys[idx]))
keys[idx]->incr_refs_all();
}
if (without_merges)
return;
List_iterator<SEL_IMERGE> it(arg->merges);
for (SEL_IMERGE *el= it++; el; el= it++)
{
SEL_IMERGE *merge= new SEL_IMERGE(el, 0, param);
if (!merge || merge->trees == merge->trees_next)
{
merges.empty();
return;
}
merges.push_back (merge);
}
}
/*
Copy constructor for SEL_IMERGE objects
SYNOPSIS
SEL_IMERGE
arg The source imerge for the constructor
cnt How many trees from arg are to be copied
param Context info for the operation
DESCRIPTION
The cnt==0 then the constructor creates a full copy of the
imerge arg. Otherwise only the first cnt trees of the imerge
are copied.
*/
SEL_IMERGE::SEL_IMERGE(SEL_IMERGE *arg, uint cnt,
RANGE_OPT_PARAM *param) : Sql_alloc()
{
uint elements= (arg->trees_end - arg->trees);
if (elements > PREALLOCED_TREES)
{
uint size= elements * sizeof (SEL_TREE **);
if (!(trees= (SEL_TREE **)alloc_root(param->mem_root, size)))
goto mem_err;
}
else
trees= &trees_prealloced[0];
trees_next= trees + (cnt ? cnt : arg->trees_next-arg->trees);
trees_end= trees + elements;
for (SEL_TREE **tree = trees, **arg_tree= arg->trees; tree < trees_next;
tree++, arg_tree++)
{
if (!(*tree= new SEL_TREE(*arg_tree, TRUE, param)))
goto mem_err;
}
return;
mem_err:
trees= &trees_prealloced[0];
trees_next= trees;
trees_end= trees;
}
/*
Perform AND operation on two imerge lists
SYNOPSIS
imerge_list_and_list()
param Context info for the operation
im1 The first imerge list for the operation
im2 The second imerge list for the operation
DESCRIPTION
The function just appends the imerge list im2 to the imerge list im1
RETURN VALUE
none
*/
inline void imerge_list_and_list(List<SEL_IMERGE> *im1, List<SEL_IMERGE> *im2)
{
im1->concat(im2);
}
/*
Perform OR operation on two imerge lists
SYNOPSIS
imerge_list_or_list()
param Context info for the operation
im1 The first imerge list for the operation
im2 The second imerge list for the operation
DESCRIPTION
Assuming that the first imerge list represents the formula
F1= M1_1 AND ... AND M1_k1
while the second imerge list represents the formula
F2= M2_1 AND ... AND M2_k2,
where M1_i= RT1_i_1 OR ... OR RT1_i_l1i (i in [1..k1])
and M2_i = RT2_i_1 OR ... OR RT2_i_l2i (i in [1..k2]),
the function builds a list of imerges for some formula that can be
inferred from the formula (F1 OR F2).
More exactly the function builds imerges for the formula (M1_1 OR M2_1).
Note that
(F1 OR F2) = (M1_1 AND ... AND M1_k1) OR (M2_1 AND ... AND M2_k2) =
AND (M1_i OR M2_j) (i in [1..k1], j in [1..k2]) =>
M1_1 OR M2_1.
So (M1_1 OR M2_1) is indeed an inference formula for (F1 OR F2).
To build imerges for the formula (M1_1 OR M2_1) the function invokes,
possibly twice, the method SEL_IMERGE::or_sel_imerge_with_checks
for the imerge m1_1.
At its first invocation the method SEL_IMERGE::or_sel_imerge_with_checks
performs OR operation on the imerge m1_1 and the range tree rt2_1_1 by
calling SEL_IMERGE::or_sel_tree_with_checks with is_first_pass_check==TRUE.
The resulting imerge of the operation is ored with the next range tree of
the imerge m2_1. This oring continues until the last range tree from
m2_1 has been ored.
At its second invocation the method SEL_IMERGE::or_sel_imerge_with_checks
performs the same sequence of OR operations, but now calling
SEL_IMERGE::or_sel_tree_with_checks with is_first_pass_check==FALSE.
The imerges that the operation produces replace those in the list im1
RETURN
0 if the operation is a success
-1 if the function has run out of memory
*/
int imerge_list_or_list(RANGE_OPT_PARAM *param,
List<SEL_IMERGE> *im1,
List<SEL_IMERGE> *im2)
{
uint rc;
bool is_last_check_pass= FALSE;
SEL_IMERGE *imerge= im1->head();
uint elems= imerge->trees_next-imerge->trees;
im1->empty();
im1->push_back(imerge);
rc= imerge->or_sel_imerge_with_checks(param, elems, im2->head(),
TRUE, &is_last_check_pass);
if (rc)
{
if (rc == 1)
{
im1->empty();
rc= 0;
}
return rc;
}
if (!is_last_check_pass)
{
SEL_IMERGE* new_imerge= new SEL_IMERGE(imerge, elems, param);
if (new_imerge)
{
is_last_check_pass= TRUE;
rc= new_imerge->or_sel_imerge_with_checks(param, elems, im2->head(),
FALSE, &is_last_check_pass);
if (!rc)
im1->push_back(new_imerge);
}
}
return rc;
}
/*
Perform OR operation for each imerge from a list and the range part of a tree
SYNOPSIS
imerge_list_or_tree()
param Context info for the operation
merges The list of imerges to be ored with the range part of tree
tree SEL_TREE whose range part is to be ored with the imerges
DESCRIPTION
For each imerge mi from the list 'merges' the function performes OR
operation with mi and the range part of 'tree' rt, producing one or
two imerges.
Given the merge mi represent the formula RTi_1 OR ... OR RTi_k,
the function forms the merges by the following rules:
1. If rt cannot be ored with any of the trees rti the function just
produces an imerge that represents the formula
RTi_1 OR ... RTi_k OR RT.
2. If there exist a tree rtj that must be ored with rt the function
produces an imerge the represents the formula
RTi_1 OR ... OR (RTi_j OR RT) OR ... OR RTi_k,
where the range tree for (RTi_j OR RT) is constructed by oring the
SEL_ARG trees that must be ored.
3. For each rti_j that can be ored with rt the function produces
the new tree rti_j' and substitutes rti_j for this new range tree.
In any case the function removes mi from the list and then adds all
produced imerges.
To build imerges by rules 1-3 the function calls the method
SEL_IMERGE::or_sel_tree_with_checks, possibly twice. With the first
call it passes TRUE for the third parameter of the function.
At this first call imerges by rules 1-2 are built. If the call
returns FALSE as the return value of its fourth parameter then the
function are called for the second time. At this call the imerge
of rule 3 is produced.
If a call of SEL_IMERGE::or_sel_tree_with_checks returns 1 then
then it means that the produced tree contains an always true
range tree and the whole imerge can be discarded.
RETURN
1 if no imerges are produced
0 otherwise
*/
static
int imerge_list_or_tree(RANGE_OPT_PARAM *param,
List<SEL_IMERGE> *merges,
SEL_TREE *tree)
{
SEL_IMERGE *imerge;
List<SEL_IMERGE> additional_merges;
List_iterator<SEL_IMERGE> it(*merges);
while ((imerge= it++))
{
bool is_last_check_pass;
int rc= 0;
int rc1= 0;
SEL_TREE *or_tree= new SEL_TREE (tree, FALSE, param);
if (or_tree)
{
uint elems= imerge->trees_next-imerge->trees;
rc= imerge->or_sel_tree_with_checks(param, elems, or_tree,
TRUE, &is_last_check_pass);
if (!is_last_check_pass)
{
SEL_IMERGE *new_imerge= new SEL_IMERGE(imerge, elems, param);
if (new_imerge)
{
rc1= new_imerge->or_sel_tree_with_checks(param, elems, or_tree,
FALSE, &is_last_check_pass);
if (!rc1)
additional_merges.push_back(new_imerge);
}
}
}
if (rc || rc1 || !or_tree)
it.remove();
}
merges->concat(&additional_merges);
return merges->is_empty();
}
/*
Perform pushdown operation of the range part of a tree into given imerges
SYNOPSIS
imerge_list_and_tree()
param Context info for the operation
merges IN/OUT List of imerges to push the range part of 'tree' into
tree SEL_TREE whose range part is to be pushed into imerges
replace if the pushdow operation for a imerge is a success
then the original imerge is replaced for the result
of the pushdown
DESCRIPTION
For each imerge from the list merges the function pushes the range part
rt of 'tree' into the imerge.
More exactly if the imerge mi from the list represents the formula
RTi_1 OR ... OR RTi_k
the function bulds a new imerge that represents the formula
(RTi_1 AND RT) OR ... OR (RTi_k AND RT)
and adds this imerge to the list merges.
To perform this pushdown operation the function calls the method
SEL_IMERGE::and_sel_tree.
For any imerge mi the new imerge is not created if for each pair of
trees rti_j and rt the intersection of the indexes with defined ranges
is empty.
If the result of the pushdown operation for the imerge mi returns an
imerge with no trees then then not only nothing is added to the list
merges but mi itself is removed from the list.
TODO
Optimize the code in order to not create new SEL_IMERGE and new SER_TREE
objects when 'replace' is TRUE. (Currently this function is called always
with this parameter equal to TRUE.)
RETURN
1 if no imerges are left in the list merges
0 otherwise
*/
static
int imerge_list_and_tree(RANGE_OPT_PARAM *param,
List<SEL_IMERGE> *merges,
SEL_TREE *tree,
bool replace)
{
SEL_IMERGE *imerge;
SEL_IMERGE *new_imerge= NULL;
List<SEL_IMERGE> new_merges;
List_iterator<SEL_IMERGE> it(*merges);
while ((imerge= it++))
{
if (!new_imerge)
new_imerge= new SEL_IMERGE();
if (imerge->have_common_keys(param, tree) &&
new_imerge && !imerge->and_sel_tree(param, tree, new_imerge))
{
if (new_imerge->trees == new_imerge->trees_next)
it.remove();
else
{
if (replace)
it.replace(new_imerge);
else
new_merges.push_back(new_imerge);
new_imerge= NULL;
}
}
}
imerge_list_and_list(&new_merges, merges);
*merges= new_merges;
return merges->is_empty();
}
/***************************************************************************
** Basic functions for SQL_SELECT and QUICK_RANGE_SELECT
***************************************************************************/
/* make a select from mysql info
Error is set as following:
0 = ok
1 = Got some error (out of memory?)
*/
SQL_SELECT *make_select(TABLE *head, table_map const_tables,
table_map read_tables, COND *conds,
bool allow_null_cond,
int *error)
{
SQL_SELECT *select;
DBUG_ENTER("make_select");
*error=0;
if (!conds && !allow_null_cond)
DBUG_RETURN(0);
if (!(select= new SQL_SELECT))
{
*error= 1; // out of memory
DBUG_RETURN(0); /* purecov: inspected */
}
select->read_tables=read_tables;
select->const_tables=const_tables;
select->head=head;
select->cond= conds;
if (head->sort.io_cache)
{
select->file= *head->sort.io_cache;
select->records=(ha_rows) (select->file.end_of_file/
head->file->ref_length);
my_free(head->sort.io_cache);
head->sort.io_cache=0;
}
DBUG_RETURN(select);
}
SQL_SELECT::SQL_SELECT() :quick(0),cond(0),pre_idx_push_select_cond(NULL),free_cond(0)
{
quick_keys.clear_all(); needed_reg.clear_all();
my_b_clear(&file);
}
void SQL_SELECT::cleanup()
{
delete quick;
quick= 0;
if (free_cond)
{
free_cond=0;
delete cond;
cond= 0;
}
close_cached_file(&file);
}
SQL_SELECT::~SQL_SELECT()
{
cleanup();
}
#undef index // Fix for Unixware 7
QUICK_SELECT_I::QUICK_SELECT_I()
:max_used_key_length(0),
used_key_parts(0)
{}
QUICK_RANGE_SELECT::QUICK_RANGE_SELECT(THD *thd, TABLE *table, uint key_nr,
bool no_alloc, MEM_ROOT *parent_alloc,
bool *create_error)
:doing_key_read(0),free_file(0),cur_range(NULL),last_range(0),dont_free(0)
{
my_bitmap_map *bitmap;
DBUG_ENTER("QUICK_RANGE_SELECT::QUICK_RANGE_SELECT");
in_ror_merged_scan= 0;
index= key_nr;
head= table;
key_part_info= head->key_info[index].key_part;
my_init_dynamic_array(&ranges, sizeof(QUICK_RANGE*), 16, 16,
MYF(MY_THREAD_SPECIFIC));
/* 'thd' is not accessible in QUICK_RANGE_SELECT::reset(). */
mrr_buf_size= thd->variables.mrr_buff_size;
mrr_buf_desc= NULL;
if (!no_alloc && !parent_alloc)
{
// Allocates everything through the internal memroot
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
thd->mem_root= &alloc;
}
else
bzero((char*) &alloc,sizeof(alloc));
file= head->file;
record= head->record[0];
/* Allocate a bitmap for used columns (Q: why not on MEM_ROOT?) */
if (!(bitmap= (my_bitmap_map*) my_malloc(head->s->column_bitmap_size,
MYF(MY_WME | MY_THREAD_SPECIFIC))))
{
column_bitmap.bitmap= 0;
*create_error= 1;
}
else
my_bitmap_init(&column_bitmap, bitmap, head->s->fields, FALSE);
DBUG_VOID_RETURN;
}
void QUICK_RANGE_SELECT::need_sorted_output()
{
if (!(mrr_flags & HA_MRR_SORTED))
{
/*
Native implementation can't produce sorted output. We'll have to
switch to default
*/
mrr_flags |= HA_MRR_USE_DEFAULT_IMPL;
}
mrr_flags |= HA_MRR_SORTED;
}
int QUICK_RANGE_SELECT::init()
{
DBUG_ENTER("QUICK_RANGE_SELECT::init");
if (file->inited != handler::NONE)
file->ha_index_or_rnd_end();
DBUG_RETURN(FALSE);
}
void QUICK_RANGE_SELECT::range_end()
{
if (file->inited != handler::NONE)
file->ha_index_or_rnd_end();
}
QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT()
{
DBUG_ENTER("QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT");
if (!dont_free)
{
/* file is NULL for CPK scan on covering ROR-intersection */
if (file)
{
range_end();
if (doing_key_read)
file->extra(HA_EXTRA_NO_KEYREAD);
if (free_file)
{
DBUG_PRINT("info", ("Freeing separate handler 0x%lx (free: %d)", (long) file,
free_file));
file->ha_external_lock(current_thd, F_UNLCK);
file->ha_close();
delete file;
}
}
delete_dynamic(&ranges); /* ranges are allocated in alloc */
free_root(&alloc,MYF(0));
my_free(column_bitmap.bitmap);
}
my_free(mrr_buf_desc);
DBUG_VOID_RETURN;
}
/*
QUICK_INDEX_SORT_SELECT works as follows:
- Do index scans, accumulate rowids in the Unique object
(Unique will also sort and de-duplicate rowids)
- Use rowids from unique to run a disk-ordered sweep
*/
QUICK_INDEX_SORT_SELECT::QUICK_INDEX_SORT_SELECT(THD *thd_param,
TABLE *table)
:unique(NULL), pk_quick_select(NULL), thd(thd_param)
{
DBUG_ENTER("QUICK_INDEX_SORT_SELECT::QUICK_INDEX_SORT_SELECT");
index= MAX_KEY;
head= table;
bzero(&read_record, sizeof(read_record));
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
DBUG_VOID_RETURN;
}
int QUICK_INDEX_SORT_SELECT::init()
{
DBUG_ENTER("QUICK_INDEX_SORT_SELECT::init");
DBUG_RETURN(0);
}
int QUICK_INDEX_SORT_SELECT::reset()
{
DBUG_ENTER("QUICK_INDEX_SORT_SELECT::reset");
const int retval= read_keys_and_merge();
DBUG_RETURN(retval);
}
bool
QUICK_INDEX_SORT_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick_sel_range)
{
DBUG_ENTER("QUICK_INDEX_SORT_SELECT::push_quick_back");
if (head->file->primary_key_is_clustered() &&
quick_sel_range->index == head->s->primary_key)
{
/*
A quick_select over a clustered primary key is handled specifically
Here we assume:
- PK columns are included in any other merged index
- Scan on the PK is disk-ordered.
(not meeting #2 will only cause performance degradation)
We could treat clustered PK as any other index, but that would
be inefficient. There is no point in doing scan on
CPK, remembering the rowid, then making rnd_pos() call with
that rowid.
*/
pk_quick_select= quick_sel_range;
DBUG_RETURN(0);
}
DBUG_RETURN(quick_selects.push_back(quick_sel_range));
}
QUICK_INDEX_SORT_SELECT::~QUICK_INDEX_SORT_SELECT()
{
List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
QUICK_RANGE_SELECT* quick;
DBUG_ENTER("QUICK_INDEX_SORT_SELECT::~QUICK_INDEX_SORT_SELECT");
delete unique;
quick_it.rewind();
while ((quick= quick_it++))
quick->file= NULL;
quick_selects.delete_elements();
delete pk_quick_select;
/* It's ok to call the next two even if they are already deinitialized */
end_read_record(&read_record);
free_io_cache(head);
free_root(&alloc,MYF(0));
DBUG_VOID_RETURN;
}
QUICK_ROR_INTERSECT_SELECT::QUICK_ROR_INTERSECT_SELECT(THD *thd_param,
TABLE *table,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
: cpk_quick(NULL), thd(thd_param), need_to_fetch_row(retrieve_full_rows),
scans_inited(FALSE)
{
index= MAX_KEY;
head= table;
record= head->record[0];
if (!parent_alloc)
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
else
bzero(&alloc, sizeof(MEM_ROOT));
last_rowid= (uchar*) alloc_root(parent_alloc? parent_alloc : &alloc,
head->file->ref_length);
}
/*
Do post-constructor initialization.
SYNOPSIS
QUICK_ROR_INTERSECT_SELECT::init()
RETURN
0 OK
other Error code
*/
int QUICK_ROR_INTERSECT_SELECT::init()
{
DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init");
/* Check if last_rowid was successfully allocated in ctor */
DBUG_RETURN(!last_rowid);
}
/*
Initialize this quick select to be a ROR-merged scan.
SYNOPSIS
QUICK_RANGE_SELECT::init_ror_merged_scan()
reuse_handler If TRUE, use head->file, otherwise create a separate
handler object
NOTES
This function creates and prepares for subsequent use a separate handler
object if it can't reuse head->file. The reason for this is that during
ROR-merge several key scans are performed simultaneously, and a single
handler is only capable of preserving context of a single key scan.
In ROR-merge the quick select doing merge does full records retrieval,
merged quick selects read only keys.
RETURN
0 ROR child scan initialized, ok to use.
1 error
*/
int QUICK_RANGE_SELECT::init_ror_merged_scan(bool reuse_handler, MEM_ROOT *alloc)
{
handler *save_file= file, *org_file;
my_bool org_key_read;
THD *thd= head->in_use;
MY_BITMAP * const save_read_set= head->read_set;
MY_BITMAP * const save_write_set= head->write_set;
DBUG_ENTER("QUICK_RANGE_SELECT::init_ror_merged_scan");
in_ror_merged_scan= 1;
if (reuse_handler)
{
DBUG_PRINT("info", ("Reusing handler 0x%lx", (long) file));
if (init())
{
DBUG_RETURN(1);
}
head->column_bitmaps_set(&column_bitmap, &column_bitmap);
goto end;
}
/* Create a separate handler object for this quick select */
if (free_file)
{
/* already have own 'handler' object. */
DBUG_RETURN(0);
}
if (!(file= head->file->clone(head->s->normalized_path.str, alloc)))
{
/*
Manually set the error flag. Note: there seems to be quite a few
places where a failure could cause the server to "hang" the client by
sending no response to a query. ATM those are not real errors because
the storage engine calls in question happen to never fail with the
existing storage engines.
*/
my_error(ER_OUT_OF_RESOURCES, MYF(0)); /* purecov: inspected */
/* Caller will free the memory */
goto failure; /* purecov: inspected */
}
head->column_bitmaps_set(&column_bitmap, &column_bitmap);
if (file->ha_external_lock(thd, F_RDLCK))
goto failure;
if (init())
{
file->ha_external_lock(thd, F_UNLCK);
file->ha_close();
goto failure;
}
free_file= TRUE;
last_rowid= file->ref;
end:
DBUG_ASSERT(head->read_set == &column_bitmap);
/*
We are only going to read key fields and call position() on 'file'
The following sets head->tmp_set to only use this key and then updates
head->read_set and head->write_set to use this bitmap.
The now bitmap is stored in 'column_bitmap' which is used in ::get_next()
*/
org_file= head->file;
org_key_read= head->key_read;
head->file= file;
head->key_read= 0;
head->mark_columns_used_by_index_no_reset(index, head->read_set);
if (!head->no_keyread)
{
doing_key_read= 1;
head->enable_keyread();
}
head->prepare_for_position();
head->file= org_file;
head->key_read= org_key_read;
/* Restore head->read_set (and write_set) to what they had before the call */
head->column_bitmaps_set(save_read_set, save_write_set);
if (reset())
{
if (!reuse_handler)
{
file->ha_external_lock(thd, F_UNLCK);
file->ha_close();
goto failure;
}
else
DBUG_RETURN(1);
}
DBUG_RETURN(0);
failure:
head->column_bitmaps_set(save_read_set, save_write_set);
delete file;
file= save_file;
DBUG_RETURN(1);
}
/*
Initialize this quick select to be a part of a ROR-merged scan.
SYNOPSIS
QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan()
reuse_handler If TRUE, use head->file, otherwise create separate
handler object.
RETURN
0 OK
other error code
*/
int QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan(bool reuse_handler,
MEM_ROOT *alloc)
{
List_iterator_fast<QUICK_SELECT_WITH_RECORD> quick_it(quick_selects);
QUICK_SELECT_WITH_RECORD *cur;
QUICK_RANGE_SELECT *quick;
DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan");
/* Initialize all merged "children" quick selects */
DBUG_ASSERT(!need_to_fetch_row || reuse_handler);
if (!need_to_fetch_row && reuse_handler)
{
cur= quick_it++;
quick= cur->quick;
/*
There is no use of this->file. Use it for the first of merged range
selects.
*/
int error= quick->init_ror_merged_scan(TRUE, alloc);
if (error)
DBUG_RETURN(error);
quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
}
while ((cur= quick_it++))
{
quick= cur->quick;
#ifndef DBUG_OFF
const MY_BITMAP * const save_read_set= quick->head->read_set;
const MY_BITMAP * const save_write_set= quick->head->write_set;
#endif
if (quick->init_ror_merged_scan(FALSE, alloc))
DBUG_RETURN(1);
quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
// Sets are shared by all members of "quick_selects" so must not change
#ifndef DBUG_OFF
DBUG_ASSERT(quick->head->read_set == save_read_set);
DBUG_ASSERT(quick->head->write_set == save_write_set);
#endif
/* All merged scans share the same record buffer in intersection. */
quick->record= head->record[0];
}
if (need_to_fetch_row && head->file->ha_rnd_init_with_error(false))
{
DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
/*
Initialize quick select for row retrieval.
SYNOPSIS
reset()
RETURN
0 OK
other Error code
*/
int QUICK_ROR_INTERSECT_SELECT::reset()
{
DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::reset");
if (!scans_inited && init_ror_merged_scan(TRUE, &alloc))
DBUG_RETURN(1);
scans_inited= TRUE;
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
QUICK_SELECT_WITH_RECORD *qr;
while ((qr= it++))
qr->quick->reset();
DBUG_RETURN(0);
}
/*
Add a merged quick select to this ROR-intersection quick select.
SYNOPSIS
QUICK_ROR_INTERSECT_SELECT::push_quick_back()
alloc Mem root to create auxiliary structures on
quick Quick select to be added. The quick select must return
rows in rowid order.
NOTES
This call can only be made before init() is called.
RETURN
FALSE OK
TRUE Out of memory.
*/
bool
QUICK_ROR_INTERSECT_SELECT::push_quick_back(MEM_ROOT *alloc, QUICK_RANGE_SELECT *quick)
{
QUICK_SELECT_WITH_RECORD *qr;
if (!(qr= new QUICK_SELECT_WITH_RECORD) ||
!(qr->key_tuple= (uchar*)alloc_root(alloc, quick->max_used_key_length)))
return TRUE;
qr->quick= quick;
return quick_selects.push_back(qr);
}
QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT()
{
DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT");
quick_selects.delete_elements();
delete cpk_quick;
free_root(&alloc,MYF(0));
if (need_to_fetch_row && head->file->inited != handler::NONE)
head->file->ha_rnd_end();
DBUG_VOID_RETURN;
}
QUICK_ROR_UNION_SELECT::QUICK_ROR_UNION_SELECT(THD *thd_param,
TABLE *table)
: thd(thd_param), scans_inited(FALSE)
{
index= MAX_KEY;
head= table;
rowid_length= table->file->ref_length;
record= head->record[0];
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
thd_param->mem_root= &alloc;
}
/*
Comparison function to be used QUICK_ROR_UNION_SELECT::queue priority
queue.
SYNPOSIS
QUICK_ROR_UNION_SELECT_queue_cmp()
arg Pointer to QUICK_ROR_UNION_SELECT
val1 First merged select
val2 Second merged select
*/
C_MODE_START
static int QUICK_ROR_UNION_SELECT_queue_cmp(void *arg, uchar *val1, uchar *val2)
{
QUICK_ROR_UNION_SELECT *self= (QUICK_ROR_UNION_SELECT*)arg;
return self->head->file->cmp_ref(((QUICK_SELECT_I*)val1)->last_rowid,
((QUICK_SELECT_I*)val2)->last_rowid);
}
C_MODE_END
/*
Do post-constructor initialization.
SYNOPSIS
QUICK_ROR_UNION_SELECT::init()
RETURN
0 OK
other Error code
*/
int QUICK_ROR_UNION_SELECT::init()
{
DBUG_ENTER("QUICK_ROR_UNION_SELECT::init");
if (init_queue(&queue, quick_selects.elements, 0,
FALSE , QUICK_ROR_UNION_SELECT_queue_cmp,
(void*) this, 0, 0))
{
bzero(&queue, sizeof(QUEUE));
DBUG_RETURN(1);
}
if (!(cur_rowid= (uchar*) alloc_root(&alloc, 2*head->file->ref_length)))
DBUG_RETURN(1);
prev_rowid= cur_rowid + head->file->ref_length;
DBUG_RETURN(0);
}
/*
Initialize quick select for row retrieval.
SYNOPSIS
reset()
RETURN
0 OK
other Error code
*/
int QUICK_ROR_UNION_SELECT::reset()
{
QUICK_SELECT_I *quick;
int error;
DBUG_ENTER("QUICK_ROR_UNION_SELECT::reset");
have_prev_rowid= FALSE;
if (!scans_inited)
{
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
if (quick->init_ror_merged_scan(FALSE, &alloc))
DBUG_RETURN(1);
}
scans_inited= TRUE;
}
queue_remove_all(&queue);
/*
Initialize scans for merged quick selects and put all merged quick
selects into the queue.
*/
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
if ((error= quick->reset()))
DBUG_RETURN(error);
if ((error= quick->get_next()))
{
if (error == HA_ERR_END_OF_FILE)
continue;
DBUG_RETURN(error);
}
quick->save_last_pos();
queue_insert(&queue, (uchar*)quick);
}
/* Prepare for ha_rnd_pos calls. */
if (head->file->inited && (error= head->file->ha_rnd_end()))
{
DBUG_PRINT("error", ("ROR index_merge rnd_end call failed"));
DBUG_RETURN(error);
}
if ((error= head->file->ha_rnd_init(false)))
{
DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
DBUG_RETURN(error);
}
DBUG_RETURN(0);
}
bool
QUICK_ROR_UNION_SELECT::push_quick_back(QUICK_SELECT_I *quick_sel_range)
{
return quick_selects.push_back(quick_sel_range);
}
QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT()
{
DBUG_ENTER("QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT");
delete_queue(&queue);
quick_selects.delete_elements();
if (head->file->inited != handler::NONE)
head->file->ha_rnd_end();
free_root(&alloc,MYF(0));
DBUG_VOID_RETURN;
}
QUICK_RANGE::QUICK_RANGE()
:min_key(0),max_key(0),min_length(0),max_length(0),
flag(NO_MIN_RANGE | NO_MAX_RANGE),
min_keypart_map(0), max_keypart_map(0)
{}
SEL_ARG::SEL_ARG(SEL_ARG &arg) :Sql_alloc()
{
type=arg.type;
min_flag=arg.min_flag;
max_flag=arg.max_flag;
maybe_flag=arg.maybe_flag;
maybe_null=arg.maybe_null;
part=arg.part;
field=arg.field;
min_value=arg.min_value;
max_value=arg.max_value;
next_key_part=arg.next_key_part;
max_part_no= arg.max_part_no;
use_count=1; elements=1;
}
inline void SEL_ARG::make_root()
{
left=right= &null_element;
color=BLACK;
next=prev=0;
use_count=0; elements=1;
}
SEL_ARG::SEL_ARG(Field *f,const uchar *min_value_arg,
const uchar *max_value_arg)
:min_flag(0), max_flag(0), maybe_flag(0), maybe_null(f->real_maybe_null()),
elements(1), use_count(1), field(f), min_value((uchar*) min_value_arg),
max_value((uchar*) max_value_arg), next(0),prev(0),
next_key_part(0), color(BLACK), type(KEY_RANGE)
{
left=right= &null_element;
max_part_no= 1;
}
SEL_ARG::SEL_ARG(Field *field_,uint8 part_,
uchar *min_value_, uchar *max_value_,
uint8 min_flag_,uint8 max_flag_,uint8 maybe_flag_)
:min_flag(min_flag_),max_flag(max_flag_),maybe_flag(maybe_flag_),
part(part_),maybe_null(field_->real_maybe_null()), elements(1),use_count(1),
field(field_), min_value(min_value_), max_value(max_value_),
next(0),prev(0),next_key_part(0),color(BLACK),type(KEY_RANGE)
{
max_part_no= part+1;
left=right= &null_element;
}
SEL_ARG *SEL_ARG::clone(RANGE_OPT_PARAM *param, SEL_ARG *new_parent,
SEL_ARG **next_arg)
{
SEL_ARG *tmp;
/* Bail out if we have already generated too many SEL_ARGs */
if (++param->alloced_sel_args > MAX_SEL_ARGS)
return 0;
if (type != KEY_RANGE)
{
if (!(tmp= new (param->mem_root) SEL_ARG(type)))
return 0; // out of memory
tmp->prev= *next_arg; // Link into next/prev chain
(*next_arg)->next=tmp;
(*next_arg)= tmp;
tmp->part= this->part;
}
else
{
if (!(tmp= new (param->mem_root) SEL_ARG(field,part, min_value,max_value,
min_flag, max_flag, maybe_flag)))
return 0; // OOM
tmp->parent=new_parent;
tmp->next_key_part=next_key_part;
if (left != &null_element)
if (!(tmp->left=left->clone(param, tmp, next_arg)))
return 0; // OOM
tmp->prev= *next_arg; // Link into next/prev chain
(*next_arg)->next=tmp;
(*next_arg)= tmp;
if (right != &null_element)
if (!(tmp->right= right->clone(param, tmp, next_arg)))
return 0; // OOM
}
increment_use_count(1);
tmp->color= color;
tmp->elements= this->elements;
tmp->max_part_no= max_part_no;
return tmp;
}
/**
This gives the first SEL_ARG in the interval list, and the minimal element
in the red-black tree
@return
SEL_ARG first SEL_ARG in the interval list
*/
SEL_ARG *SEL_ARG::first()
{
SEL_ARG *next_arg=this;
if (!next_arg->left)
return 0; // MAYBE_KEY
while (next_arg->left != &null_element)
next_arg=next_arg->left;
return next_arg;
}
const SEL_ARG *SEL_ARG::first() const
{
return const_cast<SEL_ARG*>(this)->first();
}
SEL_ARG *SEL_ARG::last()
{
SEL_ARG *next_arg=this;
if (!next_arg->right)
return 0; // MAYBE_KEY
while (next_arg->right != &null_element)
next_arg=next_arg->right;
return next_arg;
}
/*
Check if a compare is ok, when one takes ranges in account
Returns -2 or 2 if the ranges where 'joined' like < 2 and >= 2
*/
static int sel_cmp(Field *field, uchar *a, uchar *b, uint8 a_flag,
uint8 b_flag)
{
int cmp;
/* First check if there was a compare to a min or max element */
if (a_flag & (NO_MIN_RANGE | NO_MAX_RANGE))
{
if ((a_flag & (NO_MIN_RANGE | NO_MAX_RANGE)) ==
(b_flag & (NO_MIN_RANGE | NO_MAX_RANGE)))
return 0;
return (a_flag & NO_MIN_RANGE) ? -1 : 1;
}
if (b_flag & (NO_MIN_RANGE | NO_MAX_RANGE))
return (b_flag & NO_MIN_RANGE) ? 1 : -1;
if (field->real_maybe_null()) // If null is part of key
{
if (*a != *b)
{
return *a ? -1 : 1;
}
if (*a)
goto end; // NULL where equal
a++; b++; // Skip NULL marker
}
cmp=field->key_cmp(a , b);
if (cmp) return cmp < 0 ? -1 : 1; // The values differed
// Check if the compared equal arguments was defined with open/closed range
end:
if (a_flag & (NEAR_MIN | NEAR_MAX))
{
if ((a_flag & (NEAR_MIN | NEAR_MAX)) == (b_flag & (NEAR_MIN | NEAR_MAX)))
return 0;
if (!(b_flag & (NEAR_MIN | NEAR_MAX)))
return (a_flag & NEAR_MIN) ? 2 : -2;
return (a_flag & NEAR_MIN) ? 1 : -1;
}
if (b_flag & (NEAR_MIN | NEAR_MAX))
return (b_flag & NEAR_MIN) ? -2 : 2;
return 0; // The elements where equal
}
SEL_ARG *SEL_ARG::clone_tree(RANGE_OPT_PARAM *param)
{
SEL_ARG tmp_link,*next_arg,*root;
next_arg= &tmp_link;
if (!(root= clone(param, (SEL_ARG *) 0, &next_arg)))
return 0;
next_arg->next=0; // Fix last link
tmp_link.next->prev=0; // Fix first link
if (root) // If not OOM
root->use_count= 0;
return root;
}
/*
Table rows retrieval plan. Range optimizer creates QUICK_SELECT_I-derived
objects from table read plans.
*/
class TABLE_READ_PLAN
{
public:
/*
Plan read cost, with or without cost of full row retrieval, depending
on plan creation parameters.
*/
double read_cost;
ha_rows records; /* estimate of #rows to be examined */
/*
If TRUE, the scan returns rows in rowid order. This is used only for
scans that can be both ROR and non-ROR.
*/
bool is_ror;
/*
Create quick select for this plan.
SYNOPSIS
make_quick()
param Parameter from test_quick_select
retrieve_full_rows If TRUE, created quick select will do full record
retrieval.
parent_alloc Memory pool to use, if any.
NOTES
retrieve_full_rows is ignored by some implementations.
RETURN
created quick select
NULL on any error.
*/
virtual QUICK_SELECT_I *make_quick(PARAM *param,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc=NULL) = 0;
/* Table read plans are allocated on MEM_ROOT and are never deleted */
static void *operator new(size_t size, MEM_ROOT *mem_root)
{ return (void*) alloc_root(mem_root, (uint) size); }
static void operator delete(void *ptr,size_t size) { TRASH(ptr, size); }
static void operator delete(void *ptr, MEM_ROOT *mem_root) { /* Never called */ }
virtual ~TABLE_READ_PLAN() {} /* Remove gcc warning */
};
class TRP_ROR_INTERSECT;
class TRP_ROR_UNION;
class TRP_INDEX_MERGE;
/*
Plan for a QUICK_RANGE_SELECT scan.
TRP_RANGE::make_quick ignores retrieve_full_rows parameter because
QUICK_RANGE_SELECT doesn't distinguish between 'index only' scans and full
record retrieval scans.
*/
class TRP_RANGE : public TABLE_READ_PLAN
{
public:
SEL_ARG *key; /* set of intervals to be used in "range" method retrieval */
uint key_idx; /* key number in PARAM::key */
uint mrr_flags;
uint mrr_buf_size;
TRP_RANGE(SEL_ARG *key_arg, uint idx_arg, uint mrr_flags_arg)
: key(key_arg), key_idx(idx_arg), mrr_flags(mrr_flags_arg)
{}
virtual ~TRP_RANGE() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
DBUG_ENTER("TRP_RANGE::make_quick");
QUICK_RANGE_SELECT *quick;
if ((quick= get_quick_select(param, key_idx, key, mrr_flags,
mrr_buf_size, parent_alloc)))
{
quick->records= records;
quick->read_time= read_cost;
}
DBUG_RETURN(quick);
}
};
/* Plan for QUICK_ROR_INTERSECT_SELECT scan. */
class TRP_ROR_INTERSECT : public TABLE_READ_PLAN
{
public:
TRP_ROR_INTERSECT() {} /* Remove gcc warning */
virtual ~TRP_ROR_INTERSECT() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc);
/* Array of pointers to ROR range scans used in this intersection */
struct st_ror_scan_info **first_scan;
struct st_ror_scan_info **last_scan; /* End of the above array */
struct st_ror_scan_info *cpk_scan; /* Clustered PK scan, if there is one */
bool is_covering; /* TRUE if no row retrieval phase is necessary */
double index_scan_costs; /* SUM(cost(index_scan)) */
};
/*
Plan for QUICK_ROR_UNION_SELECT scan.
QUICK_ROR_UNION_SELECT always retrieves full rows, so retrieve_full_rows
is ignored by make_quick.
*/
class TRP_ROR_UNION : public TABLE_READ_PLAN
{
public:
TRP_ROR_UNION() {} /* Remove gcc warning */
virtual ~TRP_ROR_UNION() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc);
TABLE_READ_PLAN **first_ror; /* array of ptrs to plans for merged scans */
TABLE_READ_PLAN **last_ror; /* end of the above array */
};
/*
Plan for QUICK_INDEX_INTERSECT_SELECT scan.
QUICK_INDEX_INTERSECT_SELECT always retrieves full rows, so retrieve_full_rows
is ignored by make_quick.
*/
class TRP_INDEX_INTERSECT : public TABLE_READ_PLAN
{
public:
TRP_INDEX_INTERSECT() {} /* Remove gcc warning */
virtual ~TRP_INDEX_INTERSECT() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc);
TRP_RANGE **range_scans; /* array of ptrs to plans of intersected scans */
TRP_RANGE **range_scans_end; /* end of the array */
/* keys whose scans are to be filtered by cpk conditions */
key_map filtered_scans;
};
/*
Plan for QUICK_INDEX_MERGE_SELECT scan.
QUICK_ROR_INTERSECT_SELECT always retrieves full rows, so retrieve_full_rows
is ignored by make_quick.
*/
class TRP_INDEX_MERGE : public TABLE_READ_PLAN
{
public:
TRP_INDEX_MERGE() {} /* Remove gcc warning */
virtual ~TRP_INDEX_MERGE() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc);
TRP_RANGE **range_scans; /* array of ptrs to plans of merged scans */
TRP_RANGE **range_scans_end; /* end of the array */
};
/*
Plan for a QUICK_GROUP_MIN_MAX_SELECT scan.
*/
class TRP_GROUP_MIN_MAX : public TABLE_READ_PLAN
{
private:
bool have_min, have_max, have_agg_distinct;
KEY_PART_INFO *min_max_arg_part;
uint group_prefix_len;
uint used_key_parts;
uint group_key_parts;
KEY *index_info;
uint index;
uint key_infix_len;
uchar key_infix[MAX_KEY_LENGTH];
SEL_TREE *range_tree; /* Represents all range predicates in the query. */
SEL_ARG *index_tree; /* The SEL_ARG sub-tree corresponding to index_info. */
uint param_idx; /* Index of used key in param->key. */
bool is_index_scan; /* Use index_next() instead of random read */
public:
/* Number of records selected by the ranges in index_tree. */
ha_rows quick_prefix_records;
public:
TRP_GROUP_MIN_MAX(bool have_min_arg, bool have_max_arg,
bool have_agg_distinct_arg,
KEY_PART_INFO *min_max_arg_part_arg,
uint group_prefix_len_arg, uint used_key_parts_arg,
uint group_key_parts_arg, KEY *index_info_arg,
uint index_arg, uint key_infix_len_arg,
uchar *key_infix_arg,
SEL_TREE *tree_arg, SEL_ARG *index_tree_arg,
uint param_idx_arg, ha_rows quick_prefix_records_arg)
: have_min(have_min_arg), have_max(have_max_arg),
have_agg_distinct(have_agg_distinct_arg),
min_max_arg_part(min_max_arg_part_arg),
group_prefix_len(group_prefix_len_arg), used_key_parts(used_key_parts_arg),
group_key_parts(group_key_parts_arg), index_info(index_info_arg),
index(index_arg), key_infix_len(key_infix_len_arg), range_tree(tree_arg),
index_tree(index_tree_arg), param_idx(param_idx_arg), is_index_scan(FALSE),
quick_prefix_records(quick_prefix_records_arg)
{
if (key_infix_len)
memcpy(this->key_infix, key_infix_arg, key_infix_len);
}
virtual ~TRP_GROUP_MIN_MAX() {} /* Remove gcc warning */
QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc);
void use_index_scan() { is_index_scan= TRUE; }
};
typedef struct st_index_scan_info
{
uint idx; /* # of used key in param->keys */
uint keynr; /* # of used key in table */
uint range_count;
ha_rows records; /* estimate of # records this scan will return */
/* Set of intervals over key fields that will be used for row retrieval. */
SEL_ARG *sel_arg;
KEY *key_info;
uint used_key_parts;
/* Estimate of # records filtered out by intersection with cpk */
ha_rows filtered_out;
/* Bitmap of fields used in index intersection */
MY_BITMAP used_fields;
/* Fields used in the query and covered by ROR scan. */
MY_BITMAP covered_fields;
uint used_fields_covered; /* # of set bits in covered_fields */
int key_rec_length; /* length of key record (including rowid) */
/*
Cost of reading all index records with values in sel_arg intervals set
(assuming there is no need to access full table records)
*/
double index_read_cost;
uint first_uncovered_field; /* first unused bit in covered_fields */
uint key_components; /* # of parts in the key */
} INDEX_SCAN_INFO;
/*
Fill param->needed_fields with bitmap of fields used in the query.
SYNOPSIS
fill_used_fields_bitmap()
param Parameter from test_quick_select function.
NOTES
Clustered PK members are not put into the bitmap as they are implicitly
present in all keys (and it is impossible to avoid reading them).
RETURN
0 Ok
1 Out of memory.
*/
static int fill_used_fields_bitmap(PARAM *param)
{
TABLE *table= param->table;
my_bitmap_map *tmp;
uint pk;
param->tmp_covered_fields.bitmap= 0;
param->fields_bitmap_size= table->s->column_bitmap_size;
if (!(tmp= (my_bitmap_map*) alloc_root(param->mem_root,
param->fields_bitmap_size)) ||
my_bitmap_init(¶m->needed_fields, tmp, table->s->fields, FALSE))
return 1;
bitmap_copy(¶m->needed_fields, table->read_set);
bitmap_union(¶m->needed_fields, table->write_set);
pk= param->table->s->primary_key;
if (pk != MAX_KEY && param->table->file->primary_key_is_clustered())
{
/* The table uses clustered PK and it is not internally generated */
KEY_PART_INFO *key_part= param->table->key_info[pk].key_part;
KEY_PART_INFO *key_part_end= key_part +
param->table->key_info[pk].user_defined_key_parts;
for (;key_part != key_part_end; ++key_part)
bitmap_clear_bit(¶m->needed_fields, key_part->fieldnr-1);
}
return 0;
}
/*
Test if a key can be used in different ranges
SYNOPSIS
SQL_SELECT::test_quick_select()
thd Current thread
keys_to_use Keys to use for range retrieval
prev_tables Tables assumed to be already read when the scan is
performed (but not read at the moment of this call)
limit Query limit
force_quick_range Prefer to use range (instead of full table scan) even
if it is more expensive.
NOTES
Updates the following in the select parameter:
needed_reg - Bits for keys with may be used if all prev regs are read
quick - Parameter to use when reading records.
In the table struct the following information is updated:
quick_keys - Which keys can be used
quick_rows - How many rows the key matches
quick_condition_rows - E(# rows that will satisfy the table condition)
IMPLEMENTATION
quick_condition_rows value is obtained as follows:
It is a minimum of E(#output rows) for all considered table access
methods (range and index_merge accesses over various indexes).
The obtained value is not a true E(#rows that satisfy table condition)
but rather a pessimistic estimate. To obtain a true E(#...) one would
need to combine estimates of various access methods, taking into account
correlations between sets of rows they will return.
For example, if values of tbl.key1 and tbl.key2 are independent (a right
assumption if we have no information about their correlation) then the
correct estimate will be:
E(#rows("tbl.key1 < c1 AND tbl.key2 < c2")) =
= E(#rows(tbl.key1 < c1)) / total_rows(tbl) * E(#rows(tbl.key2 < c2)
which is smaller than
MIN(E(#rows(tbl.key1 < c1), E(#rows(tbl.key2 < c2)))
which is currently produced.
TODO
* Change the value returned in quick_condition_rows from a pessimistic
estimate to true E(#rows that satisfy table condition).
(we can re-use some of E(#rows) calcuation code from index_merge/intersection
for this)
* Check if this function really needs to modify keys_to_use, and change the
code to pass it by reference if it doesn't.
* In addition to force_quick_range other means can be (an usually are) used
to make this function prefer range over full table scan. Figure out if
force_quick_range is really needed.
RETURN
-1 if impossible select (i.e. certainly no rows will be selected)
0 if can't use quick_select
1 if found usable ranges and quick select has been successfully created.
*/
int SQL_SELECT::test_quick_select(THD *thd, key_map keys_to_use,
table_map prev_tables,
ha_rows limit, bool force_quick_range,
bool ordered_output)
{
uint idx;
double scan_time;
DBUG_ENTER("SQL_SELECT::test_quick_select");
DBUG_PRINT("enter",("keys_to_use: %lu prev_tables: %lu const_tables: %lu",
(ulong) keys_to_use.to_ulonglong(), (ulong) prev_tables,
(ulong) const_tables));
DBUG_PRINT("info", ("records: %lu", (ulong) head->stat_records()));
delete quick;
quick=0;
needed_reg.clear_all();
quick_keys.clear_all();
DBUG_ASSERT(!head->is_filled_at_execution());
if (keys_to_use.is_clear_all() || head->is_filled_at_execution())
DBUG_RETURN(0);
records= head->stat_records();
if (!records)
records++; /* purecov: inspected */
scan_time= (double) records / TIME_FOR_COMPARE + 1;
read_time= (double) head->file->scan_time() + scan_time + 1.1;
if (head->force_index)
scan_time= read_time= DBL_MAX;
if (limit < records)
read_time= (double) records + scan_time + 1; // Force to use index
else if (read_time <= 2.0 && !force_quick_range)
DBUG_RETURN(0); /* No need for quick select */
possible_keys.clear_all();
DBUG_PRINT("info",("Time to scan table: %g", read_time));
keys_to_use.intersect(head->keys_in_use_for_query);
if (!keys_to_use.is_clear_all())
{
uchar buff[STACK_BUFF_ALLOC];
MEM_ROOT alloc;
SEL_TREE *tree= NULL;
KEY_PART *key_parts;
KEY *key_info;
PARAM param;
if (check_stack_overrun(thd, 2*STACK_MIN_SIZE + sizeof(PARAM), buff))
DBUG_RETURN(0); // Fatal error flag is set
/* set up parameter that is passed to all functions */
param.thd= thd;
param.baseflag= head->file->ha_table_flags();
param.prev_tables=prev_tables | const_tables;
param.read_tables=read_tables;
param.current_table= head->map;
param.table=head;
param.keys=0;
param.mem_root= &alloc;
param.old_root= thd->mem_root;
param.needed_reg= &needed_reg;
param.imerge_cost_buff_size= 0;
param.using_real_indexes= TRUE;
param.remove_jump_scans= TRUE;
param.force_default_mrr= ordered_output;
param.possible_keys.clear_all();
thd->no_errors=1; // Don't warn about NULL
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
if (!(param.key_parts=
(KEY_PART*) alloc_root(&alloc,
sizeof(KEY_PART) *
head->s->actual_n_key_parts(thd))) ||
fill_used_fields_bitmap(¶m))
{
thd->no_errors=0;
free_root(&alloc,MYF(0)); // Return memory & allocator
DBUG_RETURN(0); // Can't use range
}
key_parts= param.key_parts;
thd->mem_root= &alloc;
/*
Make an array with description of all key parts of all table keys.
This is used in get_mm_parts function.
*/
key_info= head->key_info;
for (idx=0 ; idx < head->s->keys ; idx++, key_info++)
{
KEY_PART_INFO *key_part_info;
uint n_key_parts= head->actual_n_key_parts(key_info);
if (!keys_to_use.is_set(idx))
continue;
if (key_info->flags & HA_FULLTEXT)
continue; // ToDo: ft-keys in non-ft ranges, if possible SerG
param.key[param.keys]=key_parts;
key_part_info= key_info->key_part;
for (uint part= 0 ; part < n_key_parts ;
part++, key_parts++, key_part_info++)
{
key_parts->key= param.keys;
key_parts->part= part;
key_parts->length= key_part_info->length;
key_parts->store_length= key_part_info->store_length;
key_parts->field= key_part_info->field;
key_parts->null_bit= key_part_info->null_bit;
key_parts->image_type =
(key_info->flags & HA_SPATIAL) ? Field::itMBR : Field::itRAW;
/* Only HA_PART_KEY_SEG is used */
key_parts->flag= (uint8) key_part_info->key_part_flag;
}
param.real_keynr[param.keys++]=idx;
}
param.key_parts_end=key_parts;
param.alloced_sel_args= 0;
/* Calculate cost of full index read for the shortest covering index */
if (!head->covering_keys.is_clear_all())
{
int key_for_use= find_shortest_key(head, &head->covering_keys);
double key_read_time= head->file->keyread_time(key_for_use, 1, records) +
(double) records / TIME_FOR_COMPARE;
DBUG_PRINT("info", ("'all'+'using index' scan will be using key %d, "
"read time %g", key_for_use, key_read_time));
if (key_read_time < read_time)
read_time= key_read_time;
}
TABLE_READ_PLAN *best_trp= NULL;
TRP_GROUP_MIN_MAX *group_trp;
double best_read_time= read_time;
if (cond)
{
if ((tree= get_mm_tree(¶m,cond)))
{
if (tree->type == SEL_TREE::IMPOSSIBLE)
{
records=0L; /* Return -1 from this function. */
read_time= (double) HA_POS_ERROR;
goto free_mem;
}
/*
If the tree can't be used for range scans, proceed anyway, as we
can construct a group-min-max quick select
*/
if (tree->type != SEL_TREE::KEY && tree->type != SEL_TREE::KEY_SMALLER)
tree= NULL;
}
}
/*
Try to construct a QUICK_GROUP_MIN_MAX_SELECT.
Notice that it can be constructed no matter if there is a range tree.
*/
group_trp= get_best_group_min_max(¶m, tree, best_read_time);
if (group_trp)
{
param.table->quick_condition_rows= MY_MIN(group_trp->records,
head->stat_records());
if (group_trp->read_cost < best_read_time)
{
best_trp= group_trp;
best_read_time= best_trp->read_cost;
}
}
if (tree)
{
/*
It is possible to use a range-based quick select (but it might be
slower than 'all' table scan).
*/
TRP_RANGE *range_trp;
TRP_ROR_INTERSECT *rori_trp;
TRP_INDEX_INTERSECT *intersect_trp;
bool can_build_covering= FALSE;
remove_nonrange_trees(¶m, tree);
/* Get best 'range' plan and prepare data for making other plans */
if ((range_trp= get_key_scans_params(¶m, tree, FALSE, TRUE,
best_read_time)))
{
best_trp= range_trp;
best_read_time= best_trp->read_cost;
}
/*
Simultaneous key scans and row deletes on several handler
objects are not allowed so don't use ROR-intersection for
table deletes.
*/
if ((thd->lex->sql_command != SQLCOM_DELETE) &&
optimizer_flag(thd, OPTIMIZER_SWITCH_INDEX_MERGE))
{
/*
Get best non-covering ROR-intersection plan and prepare data for
building covering ROR-intersection.
*/
if ((rori_trp= get_best_ror_intersect(¶m, tree, best_read_time,
&can_build_covering)))
{
best_trp= rori_trp;
best_read_time= best_trp->read_cost;
/*
Try constructing covering ROR-intersect only if it looks possible
and worth doing.
*/
if (!rori_trp->is_covering && can_build_covering &&
(rori_trp= get_best_covering_ror_intersect(¶m, tree,
best_read_time)))
best_trp= rori_trp;
}
}
/*
Do not look for an index intersection plan if there is a covering
index. The scan by this covering index will be always cheaper than
any index intersection.
*/
if (param.table->covering_keys.is_clear_all() &&
optimizer_flag(thd, OPTIMIZER_SWITCH_INDEX_MERGE) &&
optimizer_flag(thd, OPTIMIZER_SWITCH_INDEX_MERGE_SORT_INTERSECT))
{
if ((intersect_trp= get_best_index_intersect(¶m, tree,
best_read_time)))
{
best_trp= intersect_trp;
best_read_time= best_trp->read_cost;
set_if_smaller(param.table->quick_condition_rows,
intersect_trp->records);
}
}
if (optimizer_flag(thd, OPTIMIZER_SWITCH_INDEX_MERGE) &&
head->stat_records() != 0)
{
/* Try creating index_merge/ROR-union scan. */
SEL_IMERGE *imerge;
TABLE_READ_PLAN *best_conj_trp= NULL, *new_conj_trp;
LINT_INIT(new_conj_trp); /* no empty index_merge lists possible */
DBUG_PRINT("info",("No range reads possible,"
" trying to construct index_merge"));
List_iterator_fast<SEL_IMERGE> it(tree->merges);
while ((imerge= it++))
{
new_conj_trp= get_best_disjunct_quick(¶m, imerge, best_read_time);
if (new_conj_trp)
set_if_smaller(param.table->quick_condition_rows,
new_conj_trp->records);
if (new_conj_trp &&
(!best_conj_trp ||
new_conj_trp->read_cost < best_conj_trp->read_cost))
{
best_conj_trp= new_conj_trp;
best_read_time= best_conj_trp->read_cost;
}
}
if (best_conj_trp)
best_trp= best_conj_trp;
}
}
thd->mem_root= param.old_root;
/* If we got a read plan, create a quick select from it. */
if (best_trp)
{
records= best_trp->records;
if (!(quick= best_trp->make_quick(¶m, TRUE)) || quick->init())
{
delete quick;
quick= NULL;
}
}
possible_keys= param.possible_keys;
free_mem:
free_root(&alloc,MYF(0)); // Return memory & allocator
thd->mem_root= param.old_root;
thd->no_errors=0;
}
DBUG_EXECUTE("info", print_quick(quick, &needed_reg););
/*
Assume that if the user is using 'limit' we will only need to scan
limit rows if we are using a key
*/
DBUG_RETURN(records ? MY_TEST(quick) : -1);
}
/****************************************************************************
* Condition selectivity module
****************************************************************************/
/*
Build descriptors of pseudo-indexes over columns to perform range analysis
SYNOPSIS
create_key_parts_for_pseudo_indexes()
param IN/OUT data structure for the descriptors to be built
used_fields bitmap of columns for which the descriptors are to be built
DESCRIPTION
For each column marked in the bitmap used_fields the function builds
a descriptor of a single-component pseudo-index over this column that
can be used for the range analysis of the predicates over this columns.
The descriptors are created in the memory of param->mem_root.
RETURN
FALSE in the case of success
TRUE otherwise
*/
static
bool create_key_parts_for_pseudo_indexes(RANGE_OPT_PARAM *param,
MY_BITMAP *used_fields)
{
Field **field_ptr;
TABLE *table= param->table;
uint parts= 0;
for (field_ptr= table->field; *field_ptr; field_ptr++)
{
if (bitmap_is_set(used_fields, (*field_ptr)->field_index))
parts++;
}
KEY_PART *key_part;
uint keys= 0;
if (!(key_part= (KEY_PART *) alloc_root(param->mem_root,
sizeof(KEY_PART) * parts)))
return TRUE;
param->key_parts= key_part;
for (field_ptr= table->field; *field_ptr; field_ptr++)
{
if (bitmap_is_set(used_fields, (*field_ptr)->field_index))
{
Field *field= *field_ptr;
uint16 store_length;
key_part->key= keys;
key_part->part= 0;
key_part->length= (uint16) field->key_length();
store_length= key_part->length;
if (field->real_maybe_null())
store_length+= HA_KEY_NULL_LENGTH;
if (field->real_type() == MYSQL_TYPE_VARCHAR)
store_length+= HA_KEY_BLOB_LENGTH;
key_part->store_length= store_length;
key_part->field= field;
key_part->image_type= Field::itRAW;
key_part->flag= 0;
param->key[keys]= key_part;
keys++;
key_part++;
}
}
param->keys= keys;
param->key_parts_end= key_part;
return FALSE;
}
/*
Estimate the number of rows in all ranges built for a column
by the range optimizer
SYNOPSIS
records_in_column_ranges()
param the data structure to access descriptors of pseudo indexes
built over columns used in the condition of the processed query
idx the index of the descriptor of interest in param
tree the tree representing ranges built for the interesting column
DESCRIPTION
This function retrieves the ranges represented by the SEL_ARG 'tree' and
for each of them r it calls the function get_column_range_cardinality()
that estimates the number of expected rows in r. It is assumed that param
is the data structure containing the descriptors of pseudo-indexes that
has been built to perform range analysis of the range conditions imposed
on the columns used in the processed query, while idx is the index of the
descriptor created in 'param' exactly for the column for which 'tree'
has been built by the range optimizer.
RETURN
the number of rows in the retrieved ranges
*/
static
double records_in_column_ranges(PARAM *param, uint idx,
SEL_ARG *tree)
{
SEL_ARG_RANGE_SEQ seq;
KEY_MULTI_RANGE range;
range_seq_t seq_it;
double rows;
Field *field;
uint flags= 0;
double total_rows= 0;
RANGE_SEQ_IF seq_if = {NULL, sel_arg_range_seq_init,
sel_arg_range_seq_next, 0, 0};
/* Handle cases when we don't have a valid non-empty list of range */
if (!tree)
return HA_POS_ERROR;
if (tree->type == SEL_ARG::IMPOSSIBLE)
return (0L);
field= tree->field;
seq.keyno= idx;
seq.real_keyno= MAX_KEY;
seq.param= param;
seq.start= tree;
seq_it= seq_if.init((void *) &seq, 0, flags);
while (!seq_if.next(seq_it, &range))
{
key_range *min_endp, *max_endp;
min_endp= range.start_key.length? &range.start_key : NULL;
max_endp= range.end_key.length? &range.end_key : NULL;
rows= get_column_range_cardinality(field, min_endp, max_endp,
range.range_flag);
if (HA_POS_ERROR == rows)
{
total_rows= HA_POS_ERROR;
break;
}
total_rows += rows;
}
return total_rows;
}
/*
Calculate the selectivity of the condition imposed on the rows of a table
SYNOPSIS
calculate_cond_selectivity_for_table()
thd the context handle
table the table of interest
cond conditions imposed on the rows of the table
DESCRIPTION
This function calculates the selectivity of range conditions cond imposed
on the rows of 'table' in the processed query.
The calculated selectivity is assigned to the field table->cond_selectivity.
Selectivity is calculated as a product of selectivities imposed by:
1. possible range accesses. (if multiple range accesses use the same
restrictions on the same field, we make adjustments for that)
2. Sargable conditions on fields for which we have column statistics (if
a field is used in a possible range access, we assume that selectivity
is already provided by the range access' estimates)
3. Reading a few records from the table pages and checking the condition
selectivity (this is used for conditions like "column LIKE '%val%'"
where approaches #1 and #2 do not provide selectivity data).
NOTE
Currently the selectivities of range conditions over different columns are
considered independent.
RETURN
FALSE on success
TRUE otherwise
*/
bool calculate_cond_selectivity_for_table(THD *thd, TABLE *table, Item *cond)
{
uint keynr;
uint max_quick_key_parts= 0;
MY_BITMAP *used_fields= &table->cond_set;
double table_records= table->stat_records();
DBUG_ENTER("calculate_cond_selectivity_for_table");
table->cond_selectivity= 1.0;
if (!cond || table_records == 0)
DBUG_RETURN(FALSE);
if (table->pos_in_table_list->schema_table)
DBUG_RETURN(FALSE);
MY_BITMAP handled_columns;
my_bitmap_map* buf;
if (!(buf= (my_bitmap_map*)thd->alloc(table->s->column_bitmap_size)))
DBUG_RETURN(TRUE);
my_bitmap_init(&handled_columns, buf, table->s->fields, FALSE);
/*
Calculate the selectivity of the range conditions supported by indexes.
First, take into account possible range accesses.
range access estimates are the most precise, we prefer them to any other
estimate sources.
*/
for (keynr= 0; keynr < table->s->keys; keynr++)
{
if (table->quick_keys.is_set(keynr))
set_if_bigger(max_quick_key_parts, table->quick_key_parts[keynr]);
}
/*
Walk through all indexes, indexes where range access uses more keyparts
go first.
*/
for (uint quick_key_parts= max_quick_key_parts;
quick_key_parts; quick_key_parts--)
{
for (keynr= 0; keynr < table->s->keys; keynr++)
{
if (table->quick_keys.is_set(keynr) &&
table->quick_key_parts[keynr] == quick_key_parts)
{
uint i;
uint used_key_parts= table->quick_key_parts[keynr];
double quick_cond_selectivity= table->quick_rows[keynr] /
table_records;
KEY *key_info= table->key_info + keynr;
KEY_PART_INFO* key_part= key_info->key_part;
/*
Suppose, there are range conditions on two keys
KEY1 (col1, col2)
KEY2 (col3, col2)
we don't want to count selectivity of condition on col2 twice.
First, find the longest key prefix that's made of columns whose
selectivity wasn't already accounted for.
*/
for (i= 0; i < used_key_parts; i++, key_part++)
{
if (bitmap_is_set(&handled_columns, key_part->fieldnr-1))
break;
bitmap_set_bit(&handled_columns, key_part->fieldnr-1);
}
double selectivity_mult;
if (i)
{
/*
There is at least 1-column prefix of columns whose selectivity has
not yet been accounted for.
*/
table->cond_selectivity*= quick_cond_selectivity;
if (i != used_key_parts)
{
/*
Range access got us estimate for #used_key_parts.
We need estimate for #(i-1) key parts.
*/
double f1= key_info->actual_rec_per_key(i-1);
double f2= key_info->actual_rec_per_key(i);
if (f1 > 0 && f2 > 0)
selectivity_mult= f1 / f2;
else
{
/*
No statistics available, assume the selectivity is proportional
to the number of key parts.
(i=0 means 1 keypart, i=1 means 2 keyparts, so use i+1)
*/
selectivity_mult= ((double)(i+1)) / i;
}
table->cond_selectivity*= selectivity_mult;
}
/*
We need to set selectivity for fields supported by indexes.
For single-component indexes and for some first components
of other indexes we do it here. For the remaining fields
we do it later in this function, in the same way as for the
fields not used in any indexes.
*/
if (i == 1)
{
uint fieldnr= key_info->key_part[0].fieldnr;
table->field[fieldnr-1]->cond_selectivity= quick_cond_selectivity;
if (i != used_key_parts)
table->field[fieldnr-1]->cond_selectivity*= selectivity_mult;
bitmap_clear_bit(used_fields, fieldnr-1);
}
}
}
}
}
/*
Second step: calculate the selectivity of the range conditions not
supported by any index and selectivity of the range condition
over the fields whose selectivity has not been set yet.
*/
if (thd->variables.optimizer_use_condition_selectivity > 2 &&
!bitmap_is_clear_all(used_fields))
{
PARAM param;
MEM_ROOT alloc;
SEL_TREE *tree;
SEL_ARG **key, **end;
double rows;
uint idx= 0;
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
param.thd= thd;
param.mem_root= &alloc;
param.old_root= thd->mem_root;
param.table= table;
param.is_ror_scan= FALSE;
if (create_key_parts_for_pseudo_indexes(¶m, used_fields))
goto free_alloc;
param.prev_tables= param.read_tables= 0;
param.current_table= table->map;
param.using_real_indexes= FALSE;
param.real_keynr[0]= 0;
param.alloced_sel_args= 0;
thd->no_errors=1;
tree= get_mm_tree(¶m, cond);
if (!tree)
goto free_alloc;
table->reginfo.impossible_range= 0;
if (tree->type == SEL_TREE::IMPOSSIBLE)
{
rows= 0;
table->reginfo.impossible_range= 1;
goto free_alloc;
}
else if (tree->type == SEL_TREE::ALWAYS)
{
rows= table_records;
goto free_alloc;
}
else if (tree->type == SEL_TREE::MAYBE)
{
rows= table_records;
goto free_alloc;
}
for (key= tree->keys, end= key + param.keys; key != end; key++, idx++)
{
if (*key)
{
if ((*key)->type == SEL_ARG::IMPOSSIBLE)
{
rows= 0;
table->reginfo.impossible_range= 1;
goto free_alloc;
}
else
{
rows= records_in_column_ranges(¶m, idx, *key);
if (rows != HA_POS_ERROR)
(*key)->field->cond_selectivity= rows/table_records;
}
}
}
for (Field **field_ptr= table->field; *field_ptr; field_ptr++)
{
Field *table_field= *field_ptr;
if (bitmap_is_set(used_fields, table_field->field_index) &&
table_field->cond_selectivity < 1.0)
{
if (!bitmap_is_set(&handled_columns, table_field->field_index))
table->cond_selectivity*= table_field->cond_selectivity;
}
}
free_alloc:
thd->mem_root= param.old_root;
free_root(&alloc, MYF(0));
}
bitmap_union(used_fields, &handled_columns);
/* Check if we can improve selectivity estimates by using sampling */
ulong check_rows=
MY_MIN(thd->variables.optimizer_selectivity_sampling_limit,
(ulong) (table_records * SELECTIVITY_SAMPLING_SHARE));
if (cond && check_rows > SELECTIVITY_SAMPLING_THRESHOLD &&
thd->variables.optimizer_use_condition_selectivity > 4)
{
find_selective_predicates_list_processor_data *dt=
(find_selective_predicates_list_processor_data *)
alloc_root(thd->mem_root,
sizeof(find_selective_predicates_list_processor_data));
if (!dt)
DBUG_RETURN(TRUE);
dt->list.empty();
dt->table= table;
if (cond->walk(&Item::find_selective_predicates_list_processor, 0,
(uchar*) dt))
DBUG_RETURN(TRUE);
if (dt->list.elements > 0)
{
check_rows= check_selectivity(thd, check_rows, table, &dt->list);
if (check_rows > SELECTIVITY_SAMPLING_THRESHOLD)
{
COND_STATISTIC *stat;
List_iterator_fast<COND_STATISTIC> it(dt->list);
double examined_rows= check_rows;
while ((stat= it++))
{
if (!stat->positive)
{
DBUG_PRINT("info", ("To avoid 0 assigned 1 to the counter"));
stat->positive= 1; // avoid 0
}
DBUG_PRINT("info", ("The predicate selectivity : %g",
(double)stat->positive / examined_rows));
double selectivity= ((double)stat->positive) / examined_rows;
table->cond_selectivity*= selectivity;
/*
If a field is involved then we register its selectivity in case
there in an equality with the field.
For example in case
t1.a LIKE "%bla%" and t1.a = t2.b
the selectivity we have found could be used also for t2.
*/
if (stat->field_arg)
{
stat->field_arg->cond_selectivity*= selectivity;
if (stat->field_arg->next_equal_field)
{
for (Field *next_field= stat->field_arg->next_equal_field;
next_field != stat->field_arg;
next_field= next_field->next_equal_field)
{
next_field->cond_selectivity*= selectivity;
next_field->table->cond_selectivity*= selectivity;
}
}
}
}
}
/* This list and its elements put to mem_root so should not be freed */
table->cond_selectivity_sampling_explain= &dt->list;
}
}
DBUG_RETURN(FALSE);
}
/****************************************************************************
* Condition selectivity code ends
****************************************************************************/
/****************************************************************************
* Partition pruning module
****************************************************************************/
/*
Store field key image to table record
SYNOPSIS
store_key_image_to_rec()
field Field which key image should be stored
ptr Field value in key format
len Length of the value, in bytes
ATTENTION
len is the length of the value not counting the NULL-byte (at the same
time, ptr points to the key image, which starts with NULL-byte for
nullable columns)
DESCRIPTION
Copy the field value from its key image to the table record. The source
is the value in key image format, occupying len bytes in buffer pointed
by ptr. The destination is table record, in "field value in table record"
format.
*/
void store_key_image_to_rec(Field *field, uchar *ptr, uint len)
{
/* Do the same as print_key() does */
my_bitmap_map *old_map;
if (field->real_maybe_null())
{
if (*ptr)
{
field->set_null();
return;
}
field->set_notnull();
ptr++;
}
old_map= dbug_tmp_use_all_columns(field->table,
field->table->write_set);
field->set_key_image(ptr, len);
dbug_tmp_restore_column_map(field->table->write_set, old_map);
}
#ifdef WITH_PARTITION_STORAGE_ENGINE
/*
PartitionPruningModule
This part of the code does partition pruning. Partition pruning solves the
following problem: given a query over partitioned tables, find partitions
that we will not need to access (i.e. partitions that we can assume to be
empty) when executing the query.
The set of partitions to prune doesn't depend on which query execution
plan will be used to execute the query.
HOW IT WORKS
Partition pruning module makes use of RangeAnalysisModule. The following
examples show how the problem of partition pruning can be reduced to the
range analysis problem:
EXAMPLE 1
Consider a query:
SELECT * FROM t1 WHERE (t1.a < 5 OR t1.a = 10) AND t1.a > 3 AND t1.b='z'
where table t1 is partitioned using PARTITION BY RANGE(t1.a). An apparent
way to find the used (i.e. not pruned away) partitions is as follows:
1. analyze the WHERE clause and extract the list of intervals over t1.a
for the above query we will get this list: {(3 < t1.a < 5), (t1.a=10)}
2. for each interval I
{
find partitions that have non-empty intersection with I;
mark them as used;
}
EXAMPLE 2
Suppose the table is partitioned by HASH(part_func(t1.a, t1.b)). Then
we need to:
1. Analyze the WHERE clause and get a list of intervals over (t1.a, t1.b).
The list of intervals we'll obtain will look like this:
((t1.a, t1.b) = (1,'foo')),
((t1.a, t1.b) = (2,'bar')),
((t1,a, t1.b) > (10,'zz'))
2. for each interval I
{
if (the interval has form "(t1.a, t1.b) = (const1, const2)" )
{
calculate HASH(part_func(t1.a, t1.b));
find which partition has records with this hash value and mark
it as used;
}
else
{
mark all partitions as used;
break;
}
}
For both examples the step #1 is exactly what RangeAnalysisModule could
be used to do, if it was provided with appropriate index description
(array of KEY_PART structures).
In example #1, we need to provide it with description of index(t1.a),
in example #2, we need to provide it with description of index(t1.a, t1.b).
These index descriptions are further called "partitioning index
descriptions". Note that it doesn't matter if such indexes really exist,
as range analysis module only uses the description.
Putting it all together, partitioning module works as follows:
prune_partitions() {
call create_partition_index_description();
call get_mm_tree(); // invoke the RangeAnalysisModule
// analyze the obtained interval list and get used partitions
call find_used_partitions();
}
*/
struct st_part_prune_param;
struct st_part_opt_info;
typedef void (*mark_full_part_func)(partition_info*, uint32);
/*
Partition pruning operation context
*/
typedef struct st_part_prune_param
{
RANGE_OPT_PARAM range_param; /* Range analyzer parameters */
/***************************************************************
Following fields are filled in based solely on partitioning
definition and not modified after that:
**************************************************************/
partition_info *part_info; /* Copy of table->part_info */
/* Function to get partition id from partitioning fields only */
get_part_id_func get_top_partition_id_func;
/* Function to mark a partition as used (w/all subpartitions if they exist)*/
mark_full_part_func mark_full_partition_used;
/* Partitioning 'index' description, array of key parts */
KEY_PART *key;
/*
Number of fields in partitioning 'index' definition created for
partitioning (0 if partitioning 'index' doesn't include partitioning
fields)
*/
uint part_fields;
uint subpart_fields; /* Same as above for subpartitioning */
/*
Number of the last partitioning field keypart in the index, or -1 if
partitioning index definition doesn't include partitioning fields.
*/
int last_part_partno;
int last_subpart_partno; /* Same as above for supartitioning */
/*
is_part_keypart[i] == MY_TEST(keypart #i in partitioning index is a member
used in partitioning)
Used to maintain current values of cur_part_fields and cur_subpart_fields
*/
my_bool *is_part_keypart;
/* Same as above for subpartitioning */
my_bool *is_subpart_keypart;
my_bool ignore_part_fields; /* Ignore rest of partioning fields */
/***************************************************************
Following fields form find_used_partitions() recursion context:
**************************************************************/
SEL_ARG **arg_stack; /* "Stack" of SEL_ARGs */
SEL_ARG **arg_stack_end; /* Top of the stack */
/* Number of partitioning fields for which we have a SEL_ARG* in arg_stack */
uint cur_part_fields;
/* Same as cur_part_fields, but for subpartitioning */
uint cur_subpart_fields;
/* Iterator to be used to obtain the "current" set of used partitions */
PARTITION_ITERATOR part_iter;
/* Initialized bitmap of num_subparts size */
MY_BITMAP subparts_bitmap;
uchar *cur_min_key;
uchar *cur_max_key;
uint cur_min_flag, cur_max_flag;
} PART_PRUNE_PARAM;
static bool create_partition_index_description(PART_PRUNE_PARAM *prune_par);
static int find_used_partitions(PART_PRUNE_PARAM *ppar, SEL_ARG *key_tree);
static int find_used_partitions_imerge(PART_PRUNE_PARAM *ppar,
SEL_IMERGE *imerge);
static int find_used_partitions_imerge_list(PART_PRUNE_PARAM *ppar,
List<SEL_IMERGE> &merges);
static void mark_all_partitions_as_used(partition_info *part_info);
#ifndef DBUG_OFF
static void print_partitioning_index(KEY_PART *parts, KEY_PART *parts_end);
static void dbug_print_field(Field *field);
static void dbug_print_segment_range(SEL_ARG *arg, KEY_PART *part);
static void dbug_print_singlepoint_range(SEL_ARG **start, uint num);
#endif
/**
Perform partition pruning for a given table and condition.
@param thd Thread handle
@param table Table to perform partition pruning for
@param pprune_cond Condition to use for partition pruning
@note This function assumes that lock_partitions are setup when it
is invoked. The function analyzes the condition, finds partitions that
need to be used to retrieve the records that match the condition, and
marks them as used by setting appropriate bit in part_info->read_partitions
In the worst case all partitions are marked as used. If the table is not
yet locked, it will also unset bits in part_info->lock_partitions that is
not set in read_partitions.
This function returns promptly if called for non-partitioned table.
@return Operation status
@retval true Failure
@retval false Success
*/
bool prune_partitions(THD *thd, TABLE *table, Item *pprune_cond)
{
bool retval= FALSE;
partition_info *part_info = table->part_info;
DBUG_ENTER("prune_partitions");
if (!part_info)
DBUG_RETURN(FALSE); /* not a partitioned table */
if (!pprune_cond)
{
mark_all_partitions_as_used(part_info);
DBUG_RETURN(FALSE);
}
PART_PRUNE_PARAM prune_param;
MEM_ROOT alloc;
RANGE_OPT_PARAM *range_par= &prune_param.range_param;
my_bitmap_map *old_sets[2];
prune_param.part_info= part_info;
init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
range_par->mem_root= &alloc;
range_par->old_root= thd->mem_root;
if (create_partition_index_description(&prune_param))
{
mark_all_partitions_as_used(part_info);
free_root(&alloc,MYF(0)); // Return memory & allocator
DBUG_RETURN(FALSE);
}
dbug_tmp_use_all_columns(table, old_sets,
table->read_set, table->write_set);
range_par->thd= thd;
range_par->table= table;
/* range_par->cond doesn't need initialization */
range_par->prev_tables= range_par->read_tables= 0;
range_par->current_table= table->map;
range_par->keys= 1; // one index
range_par->using_real_indexes= FALSE;
range_par->remove_jump_scans= FALSE;
range_par->real_keynr[0]= 0;
range_par->alloced_sel_args= 0;
thd->no_errors=1; // Don't warn about NULL
thd->mem_root=&alloc;
bitmap_clear_all(&part_info->read_partitions);
prune_param.key= prune_param.range_param.key_parts;
SEL_TREE *tree;
int res;
tree= get_mm_tree(range_par, pprune_cond);
if (!tree)
goto all_used;
if (tree->type == SEL_TREE::IMPOSSIBLE)
{
retval= TRUE;
goto end;
}
if (tree->type != SEL_TREE::KEY && tree->type != SEL_TREE::KEY_SMALLER)
goto all_used;
if (tree->merges.is_empty())
{
/* Range analysis has produced a single list of intervals. */
prune_param.arg_stack_end= prune_param.arg_stack;
prune_param.cur_part_fields= 0;
prune_param.cur_subpart_fields= 0;
prune_param.cur_min_key= prune_param.range_param.min_key;
prune_param.cur_max_key= prune_param.range_param.max_key;
prune_param.cur_min_flag= prune_param.cur_max_flag= 0;
init_all_partitions_iterator(part_info, &prune_param.part_iter);
if (!tree->keys[0] || (-1 == (res= find_used_partitions(&prune_param,
tree->keys[0]))))
goto all_used;
}
else
{
if (tree->merges.elements == 1)
{
/*
Range analysis has produced a "merge" of several intervals lists, a
SEL_TREE that represents an expression in form
sel_imerge = (tree1 OR tree2 OR ... OR treeN)
that cannot be reduced to one tree. This can only happen when
partitioning index has several keyparts and the condition is OR of
conditions that refer to different key parts. For example, we'll get
here for "partitioning_field=const1 OR subpartitioning_field=const2"
*/
if (-1 == (res= find_used_partitions_imerge(&prune_param,
tree->merges.head())))
goto all_used;
}
else
{
/*
Range analysis has produced a list of several imerges, i.e. a
structure that represents a condition in form
imerge_list= (sel_imerge1 AND sel_imerge2 AND ... AND sel_imergeN)
This is produced for complicated WHERE clauses that range analyzer
can't really analyze properly.
*/
if (-1 == (res= find_used_partitions_imerge_list(&prune_param,
tree->merges)))
goto all_used;
}
}
/*
res == 0 => no used partitions => retval=TRUE
res == 1 => some used partitions => retval=FALSE
res == -1 - we jump over this line to all_used:
*/
retval= MY_TEST(!res);
goto end;
all_used:
retval= FALSE; // some partitions are used
mark_all_partitions_as_used(prune_param.part_info);
end:
dbug_tmp_restore_column_maps(table->read_set, table->write_set, old_sets);
thd->no_errors=0;
thd->mem_root= range_par->old_root;
free_root(&alloc,MYF(0)); // Return memory & allocator
/*
Must be a subset of the locked partitions.
lock_partitions contains the partitions marked by explicit partition
selection (... t PARTITION (pX) ...) and we must only use partitions
within that set.
*/
bitmap_intersect(&prune_param.part_info->read_partitions,
&prune_param.part_info->lock_partitions);
/*
If not yet locked, also prune partitions to lock if not UPDATEing
partition key fields. This will also prune lock_partitions if we are under
LOCK TABLES, so prune away calls to start_stmt().
TODO: enhance this prune locking to also allow pruning of
'UPDATE t SET part_key = const WHERE cond_is_prunable' so it adds
a lock for part_key partition.
*/
if (table->file->get_lock_type() == F_UNLCK &&
!partition_key_modified(table, table->write_set))
{
bitmap_copy(&prune_param.part_info->lock_partitions,
&prune_param.part_info->read_partitions);
}
if (bitmap_is_clear_all(&(prune_param.part_info->read_partitions)))
{
table->all_partitions_pruned_away= true;
retval= TRUE;
}
DBUG_RETURN(retval);
}
/*
For SEL_ARG* array, store sel_arg->min values into table record buffer
SYNOPSIS
store_selargs_to_rec()
ppar Partition pruning context
start Array of SEL_ARG* for which the minimum values should be stored
num Number of elements in the array
DESCRIPTION
For each SEL_ARG* interval in the specified array, store the left edge
field value (sel_arg->min, key image format) into the table record.
*/
static void store_selargs_to_rec(PART_PRUNE_PARAM *ppar, SEL_ARG **start,
int num)
{
KEY_PART *parts= ppar->range_param.key_parts;
for (SEL_ARG **end= start + num; start != end; start++)
{
SEL_ARG *sel_arg= (*start);
store_key_image_to_rec(sel_arg->field, sel_arg->min_value,
parts[sel_arg->part].length);
}
}
/* Mark a partition as used in the case when there are no subpartitions */
static void mark_full_partition_used_no_parts(partition_info* part_info,
uint32 part_id)
{
DBUG_ENTER("mark_full_partition_used_no_parts");
DBUG_PRINT("enter", ("Mark partition %u as used", part_id));
bitmap_set_bit(&part_info->read_partitions, part_id);
DBUG_VOID_RETURN;
}
/* Mark a partition as used in the case when there are subpartitions */
static void mark_full_partition_used_with_parts(partition_info *part_info,
uint32 part_id)
{
uint32 start= part_id * part_info->num_subparts;
uint32 end= start + part_info->num_subparts;
DBUG_ENTER("mark_full_partition_used_with_parts");
for (; start != end; start++)
{
DBUG_PRINT("info", ("1:Mark subpartition %u as used", start));
bitmap_set_bit(&part_info->read_partitions, start);
}
DBUG_VOID_RETURN;
}
/*
Find the set of used partitions for List<SEL_IMERGE>
SYNOPSIS
find_used_partitions_imerge_list
ppar Partition pruning context.
key_tree Intervals tree to perform pruning for.
DESCRIPTION
List<SEL_IMERGE> represents "imerge1 AND imerge2 AND ...".
The set of used partitions is an intersection of used partitions sets
for imerge_{i}.
We accumulate this intersection in a separate bitmap.
RETURN
See find_used_partitions()
*/
static int find_used_partitions_imerge_list(PART_PRUNE_PARAM *ppar,
List<SEL_IMERGE> &merges)
{
MY_BITMAP all_merges;
uint bitmap_bytes;
my_bitmap_map *bitmap_buf;
uint n_bits= ppar->part_info->read_partitions.n_bits;
bitmap_bytes= bitmap_buffer_size(n_bits);
if (!(bitmap_buf= (my_bitmap_map*) alloc_root(ppar->range_param.mem_root,
bitmap_bytes)))
{
/*
Fallback, process just the first SEL_IMERGE. This can leave us with more
partitions marked as used then actually needed.
*/
return find_used_partitions_imerge(ppar, merges.head());
}
my_bitmap_init(&all_merges, bitmap_buf, n_bits, FALSE);
bitmap_set_prefix(&all_merges, n_bits);
List_iterator<SEL_IMERGE> it(merges);
SEL_IMERGE *imerge;
while ((imerge=it++))
{
int res= find_used_partitions_imerge(ppar, imerge);
if (!res)
{
/* no used partitions on one ANDed imerge => no used partitions at all */
return 0;
}
if (res != -1)
bitmap_intersect(&all_merges, &ppar->part_info->read_partitions);
if (bitmap_is_clear_all(&all_merges))
return 0;
bitmap_clear_all(&ppar->part_info->read_partitions);
}
memcpy(ppar->part_info->read_partitions.bitmap, all_merges.bitmap,
bitmap_bytes);
return 1;
}
/*
Find the set of used partitions for SEL_IMERGE structure
SYNOPSIS
find_used_partitions_imerge()
ppar Partition pruning context.
key_tree Intervals tree to perform pruning for.
DESCRIPTION
SEL_IMERGE represents "tree1 OR tree2 OR ...". The implementation is
trivial - just use mark used partitions for each tree and bail out early
if for some tree_{i} all partitions are used.
RETURN
See find_used_partitions().
*/
static
int find_used_partitions_imerge(PART_PRUNE_PARAM *ppar, SEL_IMERGE *imerge)
{
int res= 0;
for (SEL_TREE **ptree= imerge->trees; ptree < imerge->trees_next; ptree++)
{
ppar->arg_stack_end= ppar->arg_stack;
ppar->cur_part_fields= 0;
ppar->cur_subpart_fields= 0;
ppar->cur_min_key= ppar->range_param.min_key;
ppar->cur_max_key= ppar->range_param.max_key;
ppar->cur_min_flag= ppar->cur_max_flag= 0;
init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
SEL_ARG *key_tree= (*ptree)->keys[0];
if (!key_tree || (-1 == (res |= find_used_partitions(ppar, key_tree))))
return -1;
}
return res;
}
/*
Collect partitioning ranges for the SEL_ARG tree and mark partitions as used
SYNOPSIS
find_used_partitions()
ppar Partition pruning context.
key_tree SEL_ARG range tree to perform pruning for
DESCRIPTION
This function
* recursively walks the SEL_ARG* tree collecting partitioning "intervals"
* finds the partitions one needs to use to get rows in these intervals
* marks these partitions as used.
The next session desribes the process in greater detail.
IMPLEMENTATION
TYPES OF RESTRICTIONS THAT WE CAN OBTAIN PARTITIONS FOR
We can find out which [sub]partitions to use if we obtain restrictions on
[sub]partitioning fields in the following form:
1. "partition_field1=const1 AND ... AND partition_fieldN=constN"
1.1 Same as (1) but for subpartition fields
If partitioning supports interval analysis (i.e. partitioning is a
function of a single table field, and partition_info::
get_part_iter_for_interval != NULL), then we can also use condition in
this form:
2. "const1 <=? partition_field <=? const2"
2.1 Same as (2) but for subpartition_field
INFERRING THE RESTRICTIONS FROM SEL_ARG TREE
The below is an example of what SEL_ARG tree may represent:
(start)
| $
| Partitioning keyparts $ subpartitioning keyparts
| $
| ... ... $
| | | $
| +---------+ +---------+ $ +-----------+ +-----------+
\-| par1=c1 |--| par2=c2 |-----| subpar1=c3|--| subpar2=c5|
+---------+ +---------+ $ +-----------+ +-----------+
| $ | |
| $ | +-----------+
| $ | | subpar2=c6|
| $ | +-----------+
| $ |
| $ +-----------+ +-----------+
| $ | subpar1=c4|--| subpar2=c8|
| $ +-----------+ +-----------+
| $
| $
+---------+ $ +------------+ +------------+
| par1=c2 |------------------| subpar1=c10|--| subpar2=c12|
+---------+ $ +------------+ +------------+
| $
... $
The up-down connections are connections via SEL_ARG::left and
SEL_ARG::right. A horizontal connection to the right is the
SEL_ARG::next_key_part connection.
find_used_partitions() traverses the entire tree via recursion on
* SEL_ARG::next_key_part (from left to right on the picture)
* SEL_ARG::left|right (up/down on the pic). Left-right recursion is
performed for each depth level.
Recursion descent on SEL_ARG::next_key_part is used to accumulate (in
ppar->arg_stack) constraints on partitioning and subpartitioning fields.
For the example in the above picture, one of stack states is:
in find_used_partitions(key_tree = "subpar2=c5") (***)
in find_used_partitions(key_tree = "subpar1=c3")
in find_used_partitions(key_tree = "par2=c2") (**)
in find_used_partitions(key_tree = "par1=c1")
in prune_partitions(...)
We apply partitioning limits as soon as possible, e.g. when we reach the
depth (**), we find which partition(s) correspond to "par1=c1 AND par2=c2",
and save them in ppar->part_iter.
When we reach the depth (***), we find which subpartition(s) correspond to
"subpar1=c3 AND subpar2=c5", and then mark appropriate subpartitions in
appropriate subpartitions as used.
It is possible that constraints on some partitioning fields are missing.
For the above example, consider this stack state:
in find_used_partitions(key_tree = "subpar2=c12") (***)
in find_used_partitions(key_tree = "subpar1=c10")
in find_used_partitions(key_tree = "par1=c2")
in prune_partitions(...)
Here we don't have constraints for all partitioning fields. Since we've
never set the ppar->part_iter to contain used set of partitions, we use
its default "all partitions" value. We get subpartition id for
"subpar1=c3 AND subpar2=c5", and mark that subpartition as used in every
partition.
The inverse is also possible: we may get constraints on partitioning
fields, but not constraints on subpartitioning fields. In that case,
calls to find_used_partitions() with depth below (**) will return -1,
and we will mark entire partition as used.
TODO
Replace recursion on SEL_ARG::left and SEL_ARG::right with a loop
RETURN
1 OK, one or more [sub]partitions are marked as used.
0 The passed condition doesn't match any partitions
-1 Couldn't infer any partition pruning "intervals" from the passed
SEL_ARG* tree (which means that all partitions should be marked as
used) Marking partitions as used is the responsibility of the caller.
*/
static
int find_used_partitions(PART_PRUNE_PARAM *ppar, SEL_ARG *key_tree)
{
int res, left_res=0, right_res=0;
int key_tree_part= (int)key_tree->part;
bool set_full_part_if_bad_ret= FALSE;
bool ignore_part_fields= ppar->ignore_part_fields;
bool did_set_ignore_part_fields= FALSE;
RANGE_OPT_PARAM *range_par= &(ppar->range_param);
if (check_stack_overrun(range_par->thd, 3*STACK_MIN_SIZE, NULL))
return -1;
if (key_tree->left != &null_element)
{
if (-1 == (left_res= find_used_partitions(ppar,key_tree->left)))
return -1;
}
/* Push SEL_ARG's to stack to enable looking backwards as well */
ppar->cur_part_fields+= ppar->is_part_keypart[key_tree_part];
ppar->cur_subpart_fields+= ppar->is_subpart_keypart[key_tree_part];
*(ppar->arg_stack_end++)= key_tree;
if (ignore_part_fields)
{
/*
We come here when a condition on the first partitioning
fields led to evaluating the partitioning condition
(due to finding a condition of the type a < const or
b > const). Thus we must ignore the rest of the
partitioning fields but we still want to analyse the
subpartitioning fields.
*/
if (key_tree->next_key_part)
res= find_used_partitions(ppar, key_tree->next_key_part);
else
res= -1;
goto pop_and_go_right;
}
if (key_tree->type == SEL_ARG::KEY_RANGE)
{
if (ppar->part_info->get_part_iter_for_interval &&
key_tree->part <= ppar->last_part_partno)
{
/* Collect left and right bound, their lengths and flags */
uchar *min_key= ppar->cur_min_key;
uchar *max_key= ppar->cur_max_key;
uchar *tmp_min_key= min_key;
uchar *tmp_max_key= max_key;
key_tree->store_min(ppar->key[key_tree->part].store_length,
&tmp_min_key, ppar->cur_min_flag);
key_tree->store_max(ppar->key[key_tree->part].store_length,
&tmp_max_key, ppar->cur_max_flag);
uint flag;
if (key_tree->next_key_part &&
key_tree->next_key_part->part == key_tree->part+1 &&
key_tree->next_key_part->part <= ppar->last_part_partno &&
key_tree->next_key_part->type == SEL_ARG::KEY_RANGE)
{
/*
There are more key parts for partition pruning to handle
This mainly happens when the condition is an equality
condition.
*/
if ((tmp_min_key - min_key) == (tmp_max_key - max_key) &&
(memcmp(min_key, max_key, (uint)(tmp_max_key - max_key)) == 0) &&
!key_tree->min_flag && !key_tree->max_flag)
{
/* Set 'parameters' */
ppar->cur_min_key= tmp_min_key;
ppar->cur_max_key= tmp_max_key;
uint save_min_flag= ppar->cur_min_flag;
uint save_max_flag= ppar->cur_max_flag;
ppar->cur_min_flag|= key_tree->min_flag;
ppar->cur_max_flag|= key_tree->max_flag;
res= find_used_partitions(ppar, key_tree->next_key_part);
/* Restore 'parameters' back */
ppar->cur_min_key= min_key;
ppar->cur_max_key= max_key;
ppar->cur_min_flag= save_min_flag;
ppar->cur_max_flag= save_max_flag;
goto pop_and_go_right;
}
/* We have arrived at the last field in the partition pruning */
uint tmp_min_flag= key_tree->min_flag,
tmp_max_flag= key_tree->max_flag;
if (!tmp_min_flag)
key_tree->next_key_part->store_min_key(ppar->key,
&tmp_min_key,
&tmp_min_flag,
ppar->last_part_partno);
if (!tmp_max_flag)
key_tree->next_key_part->store_max_key(ppar->key,
&tmp_max_key,
&tmp_max_flag,
ppar->last_part_partno);
flag= tmp_min_flag | tmp_max_flag;
}
else
flag= key_tree->min_flag | key_tree->max_flag;
if (tmp_min_key != range_par->min_key)
flag&= ~NO_MIN_RANGE;
else
flag|= NO_MIN_RANGE;
if (tmp_max_key != range_par->max_key)
flag&= ~NO_MAX_RANGE;
else
flag|= NO_MAX_RANGE;
/*
We need to call the interval mapper if we have a condition which
makes sense to prune on. In the example of COLUMNS on a and
b it makes sense if we have a condition on a, or conditions on
both a and b. If we only have conditions on b it might make sense
but this is a harder case we will solve later. For the harder case
this clause then turns into use of all partitions and thus we
simply set res= -1 as if the mapper had returned that.
TODO: What to do here is defined in WL#4065.
*/
if (ppar->arg_stack[0]->part == 0)
{
uint32 i;
uint32 store_length_array[MAX_KEY];
uint32 num_keys= ppar->part_fields;
for (i= 0; i < num_keys; i++)
store_length_array[i]= ppar->key[i].store_length;
res= ppar->part_info->
get_part_iter_for_interval(ppar->part_info,
FALSE,
store_length_array,
range_par->min_key,
range_par->max_key,
tmp_min_key - range_par->min_key,
tmp_max_key - range_par->max_key,
flag,
&ppar->part_iter);
if (!res)
goto pop_and_go_right; /* res==0 --> no satisfying partitions */
}
else
res= -1;
if (res == -1)
{
/* get a full range iterator */
init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
}
/*
Save our intent to mark full partition as used if we will not be able
to obtain further limits on subpartitions
*/
if (key_tree_part < ppar->last_part_partno)
{
/*
We need to ignore the rest of the partitioning fields in all
evaluations after this
*/
did_set_ignore_part_fields= TRUE;
ppar->ignore_part_fields= TRUE;
}
set_full_part_if_bad_ret= TRUE;
goto process_next_key_part;
}
if (key_tree_part == ppar->last_subpart_partno &&
(NULL != ppar->part_info->get_subpart_iter_for_interval))
{
PARTITION_ITERATOR subpart_iter;
DBUG_EXECUTE("info", dbug_print_segment_range(key_tree,
range_par->key_parts););
res= ppar->part_info->
get_subpart_iter_for_interval(ppar->part_info,
TRUE,
NULL, /* Currently not used here */
key_tree->min_value,
key_tree->max_value,
0, 0, /* Those are ignored here */
key_tree->min_flag |
key_tree->max_flag,
&subpart_iter);
DBUG_ASSERT(res); /* We can't get "no satisfying subpartitions" */
if (res == -1)
goto pop_and_go_right; /* all subpartitions satisfy */
uint32 subpart_id;
bitmap_clear_all(&ppar->subparts_bitmap);
while ((subpart_id= subpart_iter.get_next(&subpart_iter)) !=
NOT_A_PARTITION_ID)
bitmap_set_bit(&ppar->subparts_bitmap, subpart_id);
/* Mark each partition as used in each subpartition. */
uint32 part_id;
while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
NOT_A_PARTITION_ID)
{
for (uint i= 0; i < ppar->part_info->num_subparts; i++)
if (bitmap_is_set(&ppar->subparts_bitmap, i))
bitmap_set_bit(&ppar->part_info->read_partitions,
part_id * ppar->part_info->num_subparts + i);
}
goto pop_and_go_right;
}
if (key_tree->is_singlepoint())
{
if (key_tree_part == ppar->last_part_partno &&
ppar->cur_part_fields == ppar->part_fields &&
ppar->part_info->get_part_iter_for_interval == NULL)
{
/*
Ok, we've got "fieldN<=>constN"-type SEL_ARGs for all partitioning
fields. Save all constN constants into table record buffer.
*/
store_selargs_to_rec(ppar, ppar->arg_stack, ppar->part_fields);
DBUG_EXECUTE("info", dbug_print_singlepoint_range(ppar->arg_stack,
ppar->part_fields););
uint32 part_id;
longlong func_value;
/* Find in which partition the {const1, ...,constN} tuple goes */
if (ppar->get_top_partition_id_func(ppar->part_info, &part_id,
&func_value))
{
res= 0; /* No satisfying partitions */
goto pop_and_go_right;
}
/* Rembember the limit we got - single partition #part_id */
init_single_partition_iterator(part_id, &ppar->part_iter);
/*
If there are no subpartitions/we fail to get any limit for them,
then we'll mark full partition as used.
*/
set_full_part_if_bad_ret= TRUE;
goto process_next_key_part;
}
if (key_tree_part == ppar->last_subpart_partno &&
ppar->cur_subpart_fields == ppar->subpart_fields)
{
/*
Ok, we've got "fieldN<=>constN"-type SEL_ARGs for all subpartitioning
fields. Save all constN constants into table record buffer.
*/
store_selargs_to_rec(ppar, ppar->arg_stack_end - ppar->subpart_fields,
ppar->subpart_fields);
DBUG_EXECUTE("info", dbug_print_singlepoint_range(ppar->arg_stack_end-
ppar->subpart_fields,
ppar->subpart_fields););
/* Find the subpartition (it's HASH/KEY so we always have one) */
partition_info *part_info= ppar->part_info;
uint32 part_id, subpart_id;
if (part_info->get_subpartition_id(part_info, &subpart_id))
return 0;
/* Mark this partition as used in each subpartition. */
while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
NOT_A_PARTITION_ID)
{
bitmap_set_bit(&part_info->read_partitions,
part_id * part_info->num_subparts + subpart_id);
}
res= 1; /* Some partitions were marked as used */
goto pop_and_go_right;
}
}
else
{
/*
Can't handle condition on current key part. If we're that deep that
we're processing subpartititoning's key parts, this means we'll not be
able to infer any suitable condition, so bail out.
*/
if (key_tree_part >= ppar->last_part_partno)
{
res= -1;
goto pop_and_go_right;
}
/*
No meaning in continuing with rest of partitioning key parts.
Will try to continue with subpartitioning key parts.
*/
ppar->ignore_part_fields= true;
did_set_ignore_part_fields= true;
goto process_next_key_part;
}
}
process_next_key_part:
if (key_tree->next_key_part)
res= find_used_partitions(ppar, key_tree->next_key_part);
else
res= -1;
if (did_set_ignore_part_fields)
{
/*
We have returned from processing all key trees linked to our next
key part. We are ready to be moving down (using right pointers) and
this tree is a new evaluation requiring its own decision on whether
to ignore partitioning fields.
*/
ppar->ignore_part_fields= FALSE;
}
if (set_full_part_if_bad_ret)
{
if (res == -1)
{
/* Got "full range" for subpartitioning fields */
uint32 part_id;
bool found= FALSE;
while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
NOT_A_PARTITION_ID)
{
ppar->mark_full_partition_used(ppar->part_info, part_id);
found= TRUE;
}
res= MY_TEST(found);
}
/*
Restore the "used partitions iterator" to the default setting that
specifies iteration over all partitions.
*/
init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
}
pop_and_go_right:
/* Pop this key part info off the "stack" */
ppar->arg_stack_end--;
ppar->cur_part_fields-= ppar->is_part_keypart[key_tree_part];
ppar->cur_subpart_fields-= ppar->is_subpart_keypart[key_tree_part];
if (res == -1)
return -1;
if (key_tree->right != &null_element)
{
if (-1 == (right_res= find_used_partitions(ppar,key_tree->right)))
return -1;
}
return (left_res || right_res || res);
}
static void mark_all_partitions_as_used(partition_info *part_info)
{
bitmap_copy(&(part_info->read_partitions),
&(part_info->lock_partitions));
}
/*
Check if field types allow to construct partitioning index description
SYNOPSIS
fields_ok_for_partition_index()
pfield NULL-terminated array of pointers to fields.
DESCRIPTION
For an array of fields, check if we can use all of the fields to create
partitioning index description.
We can't process GEOMETRY fields - for these fields singlepoint intervals
cant be generated, and non-singlepoint are "special" kinds of intervals
to which our processing logic can't be applied.
It is not known if we could process ENUM fields, so they are disabled to be
on the safe side.
RETURN
TRUE Yes, fields can be used in partitioning index
FALSE Otherwise
*/
static bool fields_ok_for_partition_index(Field **pfield)
{
if (!pfield)
return FALSE;
for (; (*pfield); pfield++)
{
enum_field_types ftype= (*pfield)->real_type();
if (ftype == MYSQL_TYPE_ENUM || ftype == MYSQL_TYPE_GEOMETRY)
return FALSE;
}
return TRUE;
}
/*
Create partition index description and fill related info in the context
struct
SYNOPSIS
create_partition_index_description()
prune_par INOUT Partition pruning context
DESCRIPTION
Create partition index description. Partition index description is:
part_index(used_fields_list(part_expr), used_fields_list(subpart_expr))
If partitioning/sub-partitioning uses BLOB or Geometry fields, then
corresponding fields_list(...) is not included into index description
and we don't perform partition pruning for partitions/subpartitions.
RETURN
TRUE Out of memory or can't do partition pruning at all
FALSE OK
*/
static bool create_partition_index_description(PART_PRUNE_PARAM *ppar)
{
RANGE_OPT_PARAM *range_par= &(ppar->range_param);
partition_info *part_info= ppar->part_info;
uint used_part_fields, used_subpart_fields;
used_part_fields= fields_ok_for_partition_index(part_info->part_field_array) ?
part_info->num_part_fields : 0;
used_subpart_fields=
fields_ok_for_partition_index(part_info->subpart_field_array)?
part_info->num_subpart_fields : 0;
uint total_parts= used_part_fields + used_subpart_fields;
ppar->ignore_part_fields= FALSE;
ppar->part_fields= used_part_fields;
ppar->last_part_partno= (int)used_part_fields - 1;
ppar->subpart_fields= used_subpart_fields;
ppar->last_subpart_partno=
used_subpart_fields?(int)(used_part_fields + used_subpart_fields - 1): -1;
if (part_info->is_sub_partitioned())
{
ppar->mark_full_partition_used= mark_full_partition_used_with_parts;
ppar->get_top_partition_id_func= part_info->get_part_partition_id;
}
else
{
ppar->mark_full_partition_used= mark_full_partition_used_no_parts;
ppar->get_top_partition_id_func= part_info->get_partition_id;
}
KEY_PART *key_part;
MEM_ROOT *alloc= range_par->mem_root;
if (!total_parts ||
!(key_part= (KEY_PART*)alloc_root(alloc, sizeof(KEY_PART)*
total_parts)) ||
!(ppar->arg_stack= (SEL_ARG**)alloc_root(alloc, sizeof(SEL_ARG*)*
total_parts)) ||
!(ppar->is_part_keypart= (my_bool*)alloc_root(alloc, sizeof(my_bool)*
total_parts)) ||
!(ppar->is_subpart_keypart= (my_bool*)alloc_root(alloc, sizeof(my_bool)*
total_parts)))
return TRUE;
if (ppar->subpart_fields)
{
my_bitmap_map *buf;
uint32 bufsize= bitmap_buffer_size(ppar->part_info->num_subparts);
if (!(buf= (my_bitmap_map*) alloc_root(alloc, bufsize)))
return TRUE;
my_bitmap_init(&ppar->subparts_bitmap, buf, ppar->part_info->num_subparts,
FALSE);
}
range_par->key_parts= key_part;
Field **field= (ppar->part_fields)? part_info->part_field_array :
part_info->subpart_field_array;
bool in_subpart_fields= FALSE;
for (uint part= 0; part < total_parts; part++, key_part++)
{
key_part->key= 0;
key_part->part= part;
key_part->length= (uint16)(*field)->key_length();
key_part->store_length= (uint16)get_partition_field_store_length(*field);
DBUG_PRINT("info", ("part %u length %u store_length %u", part,
key_part->length, key_part->store_length));
key_part->field= (*field);
key_part->image_type = Field::itRAW;
/*
We set keypart flag to 0 here as the only HA_PART_KEY_SEG is checked
in the RangeAnalysisModule.
*/
key_part->flag= 0;
/* We don't set key_parts->null_bit as it will not be used */
ppar->is_part_keypart[part]= !in_subpart_fields;
ppar->is_subpart_keypart[part]= in_subpart_fields;
/*
Check if this was last field in this array, in this case we
switch to subpartitioning fields. (This will only happens if
there are subpartitioning fields to cater for).
*/
if (!*(++field))
{
field= part_info->subpart_field_array;
in_subpart_fields= TRUE;
}
}
range_par->key_parts_end= key_part;
DBUG_EXECUTE("info", print_partitioning_index(range_par->key_parts,
range_par->key_parts_end););
return FALSE;
}
#ifndef DBUG_OFF
static void print_partitioning_index(KEY_PART *parts, KEY_PART *parts_end)
{
DBUG_ENTER("print_partitioning_index");
DBUG_LOCK_FILE;
fprintf(DBUG_FILE, "partitioning INDEX(");
for (KEY_PART *p=parts; p != parts_end; p++)
{
fprintf(DBUG_FILE, "%s%s", p==parts?"":" ,", p->field->field_name);
}
fputs(");\n", DBUG_FILE);
DBUG_UNLOCK_FILE;
DBUG_VOID_RETURN;
}
/* Print field value into debug trace, in NULL-aware way. */
static void dbug_print_field(Field *field)
{
if (field->is_real_null())
fprintf(DBUG_FILE, "NULL");
else
{
char buf[256];
String str(buf, sizeof(buf), &my_charset_bin);
str.length(0);
String *pstr;
pstr= field->val_str(&str);
fprintf(DBUG_FILE, "'%s'", pstr->c_ptr_safe());
}
}
/* Print a "c1 < keypartX < c2" - type interval into debug trace. */
static void dbug_print_segment_range(SEL_ARG *arg, KEY_PART *part)
{
DBUG_ENTER("dbug_print_segment_range");
DBUG_LOCK_FILE;
if (!(arg->min_flag & NO_MIN_RANGE))
{
store_key_image_to_rec(part->field, arg->min_value, part->length);
dbug_print_field(part->field);
if (arg->min_flag & NEAR_MIN)
fputs(" < ", DBUG_FILE);
else
fputs(" <= ", DBUG_FILE);
}
fprintf(DBUG_FILE, "%s", part->field->field_name);
if (!(arg->max_flag & NO_MAX_RANGE))
{
if (arg->max_flag & NEAR_MAX)
fputs(" < ", DBUG_FILE);
else
fputs(" <= ", DBUG_FILE);
store_key_image_to_rec(part->field, arg->max_value, part->length);
dbug_print_field(part->field);
}
fputs("\n", DBUG_FILE);
DBUG_UNLOCK_FILE;
DBUG_VOID_RETURN;
}
/*
Print a singlepoint multi-keypart range interval to debug trace
SYNOPSIS
dbug_print_singlepoint_range()
start Array of SEL_ARG* ptrs representing conditions on key parts
num Number of elements in the array.
DESCRIPTION
This function prints a "keypartN=constN AND ... AND keypartK=constK"-type
interval to debug trace.
*/
static void dbug_print_singlepoint_range(SEL_ARG **start, uint num)
{
DBUG_ENTER("dbug_print_singlepoint_range");
DBUG_LOCK_FILE;
SEL_ARG **end= start + num;
for (SEL_ARG **arg= start; arg != end; arg++)
{
Field *field= (*arg)->field;
fprintf(DBUG_FILE, "%s%s=", (arg==start)?"":", ", field->field_name);
dbug_print_field(field);
}
fputs("\n", DBUG_FILE);
DBUG_UNLOCK_FILE;
DBUG_VOID_RETURN;
}
#endif
/****************************************************************************
* Partition pruning code ends
****************************************************************************/
#endif
/*
Get cost of 'sweep' full records retrieval.
SYNOPSIS
get_sweep_read_cost()
param Parameter from test_quick_select
records # of records to be retrieved
RETURN
cost of sweep
*/
double get_sweep_read_cost(const PARAM *param, ha_rows records)
{
double result;
DBUG_ENTER("get_sweep_read_cost");
if (param->table->file->primary_key_is_clustered())
{
/*
We are using the primary key to find the rows.
Calculate the cost for this.
*/
result= param->table->file->read_time(param->table->s->primary_key,
(uint)records, records);
}
else
{
/*
Rows will be retreived with rnd_pos(). Caluclate the expected
cost for this.
*/
double n_blocks=
ceil(ulonglong2double(param->table->file->stats.data_file_length) /
IO_SIZE);
double busy_blocks=
n_blocks * (1.0 - pow(1.0 - 1.0/n_blocks, rows2double(records)));
if (busy_blocks < 1.0)
busy_blocks= 1.0;
DBUG_PRINT("info",("sweep: nblocks: %g, busy_blocks: %g", n_blocks,
busy_blocks));
/*
Disabled: Bail out if # of blocks to read is bigger than # of blocks in
table data file.
if (max_cost != DBL_MAX && (busy_blocks+index_reads_cost) >= n_blocks)
return 1;
*/
JOIN *join= param->thd->lex->select_lex.join;
if (!join || join->table_count == 1)
{
/* No join, assume reading is done in one 'sweep' */
result= busy_blocks*(DISK_SEEK_BASE_COST +
DISK_SEEK_PROP_COST*n_blocks/busy_blocks);
}
else
{
/*
Possibly this is a join with source table being non-last table, so
assume that disk seeks are random here.
*/
result= busy_blocks;
}
}
DBUG_PRINT("return",("cost: %g", result));
DBUG_RETURN(result);
}
/*
Get best plan for a SEL_IMERGE disjunctive expression.
SYNOPSIS
get_best_disjunct_quick()
param Parameter from check_quick_select function
imerge Expression to use
read_time Don't create scans with cost > read_time
NOTES
index_merge cost is calculated as follows:
index_merge_cost =
cost(index_reads) + (see #1)
cost(rowid_to_row_scan) + (see #2)
cost(unique_use) (see #3)
1. cost(index_reads) =SUM_i(cost(index_read_i))
For non-CPK scans,
cost(index_read_i) = {cost of ordinary 'index only' scan}
For CPK scan,
cost(index_read_i) = {cost of non-'index only' scan}
2. cost(rowid_to_row_scan)
If table PK is clustered then
cost(rowid_to_row_scan) =
{cost of ordinary clustered PK scan with n_ranges=n_rows}
Otherwise, we use the following model to calculate costs:
We need to retrieve n_rows rows from file that occupies n_blocks blocks.
We assume that offsets of rows we need are independent variates with
uniform distribution in [0..max_file_offset] range.
We'll denote block as "busy" if it contains row(s) we need to retrieve
and "empty" if doesn't contain rows we need.
Probability that a block is empty is (1 - 1/n_blocks)^n_rows (this
applies to any block in file). Let x_i be a variate taking value 1 if
block #i is empty and 0 otherwise.
Then E(x_i) = (1 - 1/n_blocks)^n_rows;
E(n_empty_blocks) = E(sum(x_i)) = sum(E(x_i)) =
= n_blocks * ((1 - 1/n_blocks)^n_rows) =
~= n_blocks * exp(-n_rows/n_blocks).
E(n_busy_blocks) = n_blocks*(1 - (1 - 1/n_blocks)^n_rows) =
~= n_blocks * (1 - exp(-n_rows/n_blocks)).
Average size of "hole" between neighbor non-empty blocks is
E(hole_size) = n_blocks/E(n_busy_blocks).
The total cost of reading all needed blocks in one "sweep" is:
E(n_busy_blocks)*
(DISK_SEEK_BASE_COST + DISK_SEEK_PROP_COST*n_blocks/E(n_busy_blocks)).
3. Cost of Unique use is calculated in Unique::get_use_cost function.
ROR-union cost is calculated in the same way index_merge, but instead of
Unique a priority queue is used.
RETURN
Created read plan
NULL - Out of memory or no read scan could be built.
*/
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
double read_time)
{
SEL_TREE **ptree;
TRP_INDEX_MERGE *imerge_trp= NULL;
TRP_RANGE **range_scans;
TRP_RANGE **cur_child;
TRP_RANGE **cpk_scan= NULL;
bool imerge_too_expensive= FALSE;
double imerge_cost= 0.0;
ha_rows cpk_scan_records= 0;
ha_rows non_cpk_scan_records= 0;
bool pk_is_clustered= param->table->file->primary_key_is_clustered();
bool all_scans_ror_able= TRUE;
bool all_scans_rors= TRUE;
uint unique_calc_buff_size;
TABLE_READ_PLAN **roru_read_plans;
TABLE_READ_PLAN **cur_roru_plan;
double roru_index_costs;
ha_rows roru_total_records;
double roru_intersect_part= 1.0;
DBUG_ENTER("get_best_disjunct_quick");
DBUG_PRINT("info", ("Full table scan cost: %g", read_time));
/*
In every tree of imerge remove SEL_ARG trees that do not make ranges.
If after this removal some SEL_ARG tree becomes empty discard imerge.
*/
for (ptree= imerge->trees; ptree != imerge->trees_next; ptree++)
{
if (remove_nonrange_trees(param, *ptree))
{
imerge->trees_next= imerge->trees;
break;
}
}
uint n_child_scans= imerge->trees_next - imerge->trees;
if (!n_child_scans)
DBUG_RETURN(NULL);
if (!(range_scans= (TRP_RANGE**)alloc_root(param->mem_root,
sizeof(TRP_RANGE*)*
n_child_scans)))
DBUG_RETURN(NULL);
/*
Collect best 'range' scan for each of disjuncts, and, while doing so,
analyze possibility of ROR scans. Also calculate some values needed by
other parts of the code.
*/
for (ptree= imerge->trees, cur_child= range_scans;
ptree != imerge->trees_next;
ptree++, cur_child++)
{
DBUG_EXECUTE("info", print_sel_tree(param, *ptree, &(*ptree)->keys_map,
"tree in SEL_IMERGE"););
if (!(*cur_child= get_key_scans_params(param, *ptree, TRUE, FALSE, read_time)))
{
/*
One of index scans in this index_merge is more expensive than entire
table read for another available option. The entire index_merge (and
any possible ROR-union) will be more expensive then, too. We continue
here only to update SQL_SELECT members.
*/
imerge_too_expensive= TRUE;
}
if (imerge_too_expensive)
continue;
imerge_cost += (*cur_child)->read_cost;
all_scans_ror_able &= ((*ptree)->n_ror_scans > 0);
all_scans_rors &= (*cur_child)->is_ror;
if (pk_is_clustered &&
param->real_keynr[(*cur_child)->key_idx] ==
param->table->s->primary_key)
{
cpk_scan= cur_child;
cpk_scan_records= (*cur_child)->records;
}
else
non_cpk_scan_records += (*cur_child)->records;
}
DBUG_PRINT("info", ("index_merge scans cost %g", imerge_cost));
if (imerge_too_expensive || (imerge_cost > read_time) ||
((non_cpk_scan_records+cpk_scan_records >=
param->table->stat_records()) &&
read_time != DBL_MAX))
{
/*
Bail out if it is obvious that both index_merge and ROR-union will be
more expensive
*/
DBUG_PRINT("info", ("Sum of index_merge scans is more expensive than "
"full table scan, bailing out"));
DBUG_RETURN(NULL);
}
/*
If all scans happen to be ROR, proceed to generate a ROR-union plan (it's
guaranteed to be cheaper than non-ROR union), unless ROR-unions are
disabled in @@optimizer_switch
*/
if (all_scans_rors &&
optimizer_flag(param->thd, OPTIMIZER_SWITCH_INDEX_MERGE_UNION))
{
roru_read_plans= (TABLE_READ_PLAN**)range_scans;
goto skip_to_ror_scan;
}
if (cpk_scan)
{
/*
Add one ROWID comparison for each row retrieved on non-CPK scan. (it
is done in QUICK_RANGE_SELECT::row_in_ranges)
*/
imerge_cost += non_cpk_scan_records / TIME_FOR_COMPARE_ROWID;
}
/* Calculate cost(rowid_to_row_scan) */
imerge_cost += get_sweep_read_cost(param, non_cpk_scan_records);
DBUG_PRINT("info",("index_merge cost with rowid-to-row scan: %g",
imerge_cost));
if (imerge_cost > read_time ||
!optimizer_flag(param->thd, OPTIMIZER_SWITCH_INDEX_MERGE_SORT_UNION))
{
goto build_ror_index_merge;
}
/* Add Unique operations cost */
unique_calc_buff_size=
Unique::get_cost_calc_buff_size((ulong)non_cpk_scan_records,
param->table->file->ref_length,
param->thd->variables.sortbuff_size);
if (param->imerge_cost_buff_size < unique_calc_buff_size)
{
if (!(param->imerge_cost_buff= (uint*)alloc_root(param->mem_root,
unique_calc_buff_size)))
DBUG_RETURN(NULL);
param->imerge_cost_buff_size= unique_calc_buff_size;
}
imerge_cost +=
Unique::get_use_cost(param->imerge_cost_buff, (uint)non_cpk_scan_records,
param->table->file->ref_length,
param->thd->variables.sortbuff_size,
TIME_FOR_COMPARE_ROWID,
FALSE, NULL);
DBUG_PRINT("info",("index_merge total cost: %g (wanted: less then %g)",
imerge_cost, read_time));
if (imerge_cost < read_time)
{
if ((imerge_trp= new (param->mem_root)TRP_INDEX_MERGE))
{
imerge_trp->read_cost= imerge_cost;
imerge_trp->records= non_cpk_scan_records + cpk_scan_records;
imerge_trp->records= MY_MIN(imerge_trp->records,
param->table->stat_records());
imerge_trp->range_scans= range_scans;
imerge_trp->range_scans_end= range_scans + n_child_scans;
read_time= imerge_cost;
}
if (imerge_trp)
{
TABLE_READ_PLAN *trp= merge_same_index_scans(param, imerge, imerge_trp,
read_time);
if (trp != imerge_trp)
DBUG_RETURN(trp);
}
}
build_ror_index_merge:
if (!all_scans_ror_able ||
param->thd->lex->sql_command == SQLCOM_DELETE ||
!optimizer_flag(param->thd, OPTIMIZER_SWITCH_INDEX_MERGE_UNION))
DBUG_RETURN(imerge_trp);
/* Ok, it is possible to build a ROR-union, try it. */
bool dummy;
if (!(roru_read_plans=
(TABLE_READ_PLAN**)alloc_root(param->mem_root,
sizeof(TABLE_READ_PLAN*)*
n_child_scans)))
DBUG_RETURN(imerge_trp);
skip_to_ror_scan:
roru_index_costs= 0.0;
roru_total_records= 0;
cur_roru_plan= roru_read_plans;
/* Find 'best' ROR scan for each of trees in disjunction */
for (ptree= imerge->trees, cur_child= range_scans;
ptree != imerge->trees_next;
ptree++, cur_child++, cur_roru_plan++)
{
/*
Assume the best ROR scan is the one that has cheapest full-row-retrieval
scan cost.
Also accumulate index_only scan costs as we'll need them to calculate
overall index_intersection cost.
*/
double cost;
if ((*cur_child)->is_ror)
{
/* Ok, we have index_only cost, now get full rows scan cost */
cost= param->table->file->
read_time(param->real_keynr[(*cur_child)->key_idx], 1,
(*cur_child)->records) +
rows2double((*cur_child)->records) / TIME_FOR_COMPARE;
}
else
cost= read_time;
TABLE_READ_PLAN *prev_plan= *cur_child;
if (!(*cur_roru_plan= get_best_ror_intersect(param, *ptree, cost,
&dummy)))
{
if (prev_plan->is_ror)
*cur_roru_plan= prev_plan;
else
DBUG_RETURN(imerge_trp);
roru_index_costs += (*cur_roru_plan)->read_cost;
}
else
roru_index_costs +=
((TRP_ROR_INTERSECT*)(*cur_roru_plan))->index_scan_costs;
roru_total_records += (*cur_roru_plan)->records;
roru_intersect_part *= (*cur_roru_plan)->records /
param->table->stat_records();
}
/*
rows to retrieve=
SUM(rows_in_scan_i) - table_rows * PROD(rows_in_scan_i / table_rows).
This is valid because index_merge construction guarantees that conditions
in disjunction do not share key parts.
*/
roru_total_records -= (ha_rows)(roru_intersect_part*
param->table->stat_records());
/* ok, got a ROR read plan for each of the disjuncts
Calculate cost:
cost(index_union_scan(scan_1, ... scan_n)) =
SUM_i(cost_of_index_only_scan(scan_i)) +
queue_use_cost(rowid_len, n) +
cost_of_row_retrieval
See get_merge_buffers_cost function for queue_use_cost formula derivation.
*/
double roru_total_cost;
roru_total_cost= roru_index_costs +
rows2double(roru_total_records)*log((double)n_child_scans) /
(TIME_FOR_COMPARE_ROWID * M_LN2) +
get_sweep_read_cost(param, roru_total_records);
DBUG_PRINT("info", ("ROR-union: cost %g, %d members", roru_total_cost,
n_child_scans));
TRP_ROR_UNION* roru;
if (roru_total_cost < read_time)
{
if ((roru= new (param->mem_root) TRP_ROR_UNION))
{
roru->first_ror= roru_read_plans;
roru->last_ror= roru_read_plans + n_child_scans;
roru->read_cost= roru_total_cost;
roru->records= roru_total_records;
DBUG_RETURN(roru);
}
}
DBUG_RETURN(imerge_trp);
}
/*
Merge index scans for the same indexes in an index merge plan
SYNOPSIS
merge_same_index_scans()
param Context info for the operation
imerge IN/OUT SEL_IMERGE from which imerge_trp has been extracted
imerge_trp The index merge plan where index scans for the same
indexes are to be merges
read_time The upper bound for the cost of the plan to be evaluated
DESRIPTION
For the given index merge plan imerge_trp extracted from the SEL_MERGE
imerge the function looks for range scans with the same indexes and merges
them into SEL_ARG trees. Then for each such SEL_ARG tree r_i the function
creates a range tree rt_i that contains only r_i. All rt_i are joined
into one index merge that replaces the original index merge imerge.
The function calls get_best_disjunct_quick for the new index merge to
get a new index merge plan that contains index scans only for different
indexes.
If there are no index scans for the same index in the original index
merge plan the function does not change the original imerge and returns
imerge_trp as its result.
RETURN
The original or or improved index merge plan
*/
static
TABLE_READ_PLAN *merge_same_index_scans(PARAM *param, SEL_IMERGE *imerge,
TRP_INDEX_MERGE *imerge_trp,
double read_time)
{
uint16 first_scan_tree_idx[MAX_KEY];
SEL_TREE **tree;
TRP_RANGE **cur_child;
uint removed_cnt= 0;
DBUG_ENTER("merge_same_index_scans");
bzero(first_scan_tree_idx, sizeof(first_scan_tree_idx[0])*param->keys);
for (tree= imerge->trees, cur_child= imerge_trp->range_scans;
tree != imerge->trees_next;
tree++, cur_child++)
{
DBUG_ASSERT(tree);
uint key_idx= (*cur_child)->key_idx;
uint16 *tree_idx_ptr= &first_scan_tree_idx[key_idx];
if (!*tree_idx_ptr)
*tree_idx_ptr= (uint16) (tree-imerge->trees+1);
else
{
SEL_TREE **changed_tree= imerge->trees+(*tree_idx_ptr-1);
SEL_ARG *key= (*changed_tree)->keys[key_idx];
bzero((*changed_tree)->keys,
sizeof((*changed_tree)->keys[0])*param->keys);
(*changed_tree)->keys_map.clear_all();
if (key)
key->incr_refs();
if ((*tree)->keys[key_idx])
(*tree)->keys[key_idx]->incr_refs();
if (((*changed_tree)->keys[key_idx]=
key_or(param, key, (*tree)->keys[key_idx])))
(*changed_tree)->keys_map.set_bit(key_idx);
*tree= NULL;
removed_cnt++;
}
}
if (!removed_cnt)
DBUG_RETURN(imerge_trp);
TABLE_READ_PLAN *trp= NULL;
SEL_TREE **new_trees_next= imerge->trees;
for (tree= new_trees_next; tree != imerge->trees_next; tree++)
{
if (!*tree)
continue;
if (tree > new_trees_next)
*new_trees_next= *tree;
new_trees_next++;
}
imerge->trees_next= new_trees_next;
DBUG_ASSERT(imerge->trees_next>imerge->trees);
if (imerge->trees_next-imerge->trees > 1)
trp= get_best_disjunct_quick(param, imerge, read_time);
else
{
/*
This alternative theoretically can be reached when the cost
of the index merge for such a formula as
(key1 BETWEEN c1_1 AND c1_2) AND key2 > c2 OR
(key1 BETWEEN c1_3 AND c1_4) AND key3 > c3
is estimated as being cheaper than the cost of index scan for
the formula
(key1 BETWEEN c1_1 AND c1_2) OR (key1 BETWEEN c1_3 AND c1_4)
In the current code this may happen for two reasons:
1. for a single index range scan data records are accessed in
a random order
2. the functions that estimate the cost of a range scan and an
index merge retrievals are not well calibrated
*/
trp= get_key_scans_params(param, *imerge->trees, FALSE, TRUE,
read_time);
}
DBUG_RETURN(trp);
}
/*
This structure contains the info common for all steps of a partial
index intersection plan. Morever it contains also the info common
for index intersect plans. This info is filled in by the function
prepare_search_best just before searching for the best index
intersection plan.
*/
typedef struct st_common_index_intersect_info
{
PARAM *param; /* context info for range optimizations */
uint key_size; /* size of a ROWID element stored in Unique object */
uint compare_factor; /* 1/compare - cost to compare two ROWIDs */
ulonglong max_memory_size; /* maximum space allowed for Unique objects */
ha_rows table_cardinality; /* estimate of the number of records in table */
double cutoff_cost; /* discard index intersects with greater costs */
INDEX_SCAN_INFO *cpk_scan; /* clustered primary key used in intersection */
bool in_memory; /* unique object for intersection is completely in memory */
INDEX_SCAN_INFO **search_scans; /* scans possibly included in intersect */
uint n_search_scans; /* number of elements in search_scans */
bool best_uses_cpk; /* current best intersect uses clustered primary key */
double best_cost; /* cost of the current best index intersection */
/* estimate of the number of records in the current best intersection */
ha_rows best_records;
uint best_length; /* number of indexes in the current best intersection */
INDEX_SCAN_INFO **best_intersect; /* the current best index intersection */
/* scans from the best intersect to be filtrered by cpk conditions */
key_map filtered_scans;
uint *buff_elems; /* buffer to calculate cost of index intersection */
} COMMON_INDEX_INTERSECT_INFO;
/*
This structure contains the info specific for one step of an index
intersection plan. The structure is filled in by the function
check_index_intersect_extension.
*/
typedef struct st_partial_index_intersect_info
{
COMMON_INDEX_INTERSECT_INFO *common_info; /* shared by index intersects */
uint length; /* number of index scans in the partial intersection */
ha_rows records; /* estimate of the number of records in intersection */
double cost; /* cost of the partial index intersection */
/* estimate of total number of records of all scans of the partial index
intersect sent to the Unique object used for the intersection */
ha_rows records_sent_to_unique;
/* total cost of the scans of indexes from the partial index intersection */
double index_read_cost;
bool use_cpk_filter; /* cpk filter is to be used for this scan */
bool in_memory; /* uses unique object in memory */
double in_memory_cost; /* cost of using unique object in memory */
key_map filtered_scans; /* scans to be filtered by cpk conditions */
MY_BITMAP *intersect_fields; /* bitmap of fields used in intersection */
} PARTIAL_INDEX_INTERSECT_INFO;
/* Check whether two indexes have the same first n components */
static
bool same_index_prefix(KEY *key1, KEY *key2, uint used_parts)
{
KEY_PART_INFO *part1= key1->key_part;
KEY_PART_INFO *part2= key2->key_part;
for(uint i= 0; i < used_parts; i++, part1++, part2++)
{
if (part1->fieldnr != part2->fieldnr)
return FALSE;
}
return TRUE;
}
/* Create a bitmap for all fields of a table */
static
bool create_fields_bitmap(PARAM *param, MY_BITMAP *fields_bitmap)
{
my_bitmap_map *bitmap_buf;
if (!(bitmap_buf= (my_bitmap_map *) alloc_root(param->mem_root,
param->fields_bitmap_size)))
return TRUE;
if (my_bitmap_init(fields_bitmap, bitmap_buf, param->table->s->fields, FALSE))
return TRUE;
return FALSE;
}
/* Compare two indexes scans for sort before search for the best intersection */
static
int cmp_intersect_index_scan(INDEX_SCAN_INFO **a, INDEX_SCAN_INFO **b)
{
return (*a)->records < (*b)->records ?
-1 : (*a)->records == (*b)->records ? 0 : 1;
}
static inline
void set_field_bitmap_for_index_prefix(MY_BITMAP *field_bitmap,
KEY_PART_INFO *key_part,
uint used_key_parts)
{
bitmap_clear_all(field_bitmap);
for (KEY_PART_INFO *key_part_end= key_part+used_key_parts;
key_part < key_part_end; key_part++)
{
bitmap_set_bit(field_bitmap, key_part->fieldnr-1);
}
}
/*
Round up table cardinality read from statistics provided by engine.
This function should go away when mysql test will allow to handle
more or less easily in the test suites deviations of InnoDB
statistical data.
*/
static inline
ha_rows get_table_cardinality_for_index_intersect(TABLE *table)
{
if (table->file->ha_table_flags() & HA_STATS_RECORDS_IS_EXACT)
return table->stat_records();
else
{
ha_rows d;
double q;
for (q= (double)table->stat_records(), d= 1 ; q >= 10; q/= 10, d*= 10 ) ;
return (ha_rows) (floor(q+0.5) * d);
}
}
static
ha_rows records_in_index_intersect_extension(PARTIAL_INDEX_INTERSECT_INFO *curr,
INDEX_SCAN_INFO *ext_index_scan);
/*
Prepare to search for the best index intersection
SYNOPSIS
prepare_search_best_index_intersect()
param common info about index ranges
tree tree of ranges for indexes than can be intersected
common OUT info needed for search to be filled by the function
init OUT info for an initial pseudo step of the intersection plans
cutoff_cost cut off cost of the interesting index intersection
DESCRIPTION
The function initializes all fields of the structure 'common' to be used
when searching for the best intersection plan. It also allocates
memory to store the most cheap index intersection.
NOTES
When selecting candidates for index intersection we always take only
one representative out of any set of indexes that share the same range
conditions. These indexes always have the same prefixes and the
components of this prefixes are exactly those used in these range
conditions.
Range conditions over clustered primary key (cpk) is always used only
as the condition that filters out some rowids retrieved by the scans
for secondary indexes. The cpk index will be handled in special way by
the function that search for the best index intersection.
RETURN
FALSE in the case of success
TRUE otherwise
*/
static
bool prepare_search_best_index_intersect(PARAM *param,
SEL_TREE *tree,
COMMON_INDEX_INTERSECT_INFO *common,
PARTIAL_INDEX_INTERSECT_INFO *init,
double cutoff_cost)
{
uint i;
uint n_search_scans;
double cost;
INDEX_SCAN_INFO **index_scan;
INDEX_SCAN_INFO **scan_ptr;
INDEX_SCAN_INFO *cpk_scan= NULL;
TABLE *table= param->table;
uint n_index_scans= tree->index_scans_end - tree->index_scans;
if (!n_index_scans)
return 1;
bzero(init, sizeof(*init));
init->common_info= common;
init->cost= cutoff_cost;
common->param= param;
common->key_size= table->file->ref_length;
common->compare_factor= TIME_FOR_COMPARE_ROWID;
common->max_memory_size= param->thd->variables.sortbuff_size;
common->cutoff_cost= cutoff_cost;
common->cpk_scan= NULL;
common->table_cardinality=
get_table_cardinality_for_index_intersect(table);
if (n_index_scans <= 1)
return TRUE;
if (table->file->primary_key_is_clustered())
{
INDEX_SCAN_INFO **index_scan_end;
index_scan= tree->index_scans;
index_scan_end= index_scan+n_index_scans;
for ( ; index_scan < index_scan_end; index_scan++)
{
if ((*index_scan)->keynr == table->s->primary_key)
{
common->cpk_scan= cpk_scan= *index_scan;
break;
}
}
}
i= n_index_scans - MY_TEST(cpk_scan != NULL) + 1;
if (!(common->search_scans =
(INDEX_SCAN_INFO **) alloc_root (param->mem_root,
sizeof(INDEX_SCAN_INFO *) * i)))
return TRUE;
bzero(common->search_scans, sizeof(INDEX_SCAN_INFO *) * i);
INDEX_SCAN_INFO **selected_index_scans= common->search_scans;
for (i=0, index_scan= tree->index_scans; i < n_index_scans; i++, index_scan++)
{
uint used_key_parts= (*index_scan)->used_key_parts;
KEY *key_info= (*index_scan)->key_info;
if (*index_scan == cpk_scan)
continue;
if (cpk_scan && cpk_scan->used_key_parts >= used_key_parts &&
same_index_prefix(cpk_scan->key_info, key_info, used_key_parts))
continue;
cost= table->file->keyread_time((*index_scan)->keynr,
(*index_scan)->range_count,
(*index_scan)->records);
if (cost >= cutoff_cost)
continue;
for (scan_ptr= selected_index_scans; *scan_ptr ; scan_ptr++)
{
/*
When we have range conditions for two different indexes with the same
beginning it does not make sense to consider both of them for index
intersection if the range conditions are covered by common initial
components of the indexes. Actually in this case the indexes are
guaranteed to have the same range conditions.
*/
if ((*scan_ptr)->used_key_parts == used_key_parts &&
same_index_prefix((*scan_ptr)->key_info, key_info, used_key_parts))
break;
}
if (!*scan_ptr || cost < (*scan_ptr)->index_read_cost)
{
*scan_ptr= *index_scan;
(*scan_ptr)->index_read_cost= cost;
}
}
ha_rows records_in_scans= 0;
for (scan_ptr=selected_index_scans, i= 0; *scan_ptr; scan_ptr++, i++)
{
if (create_fields_bitmap(param, &(*scan_ptr)->used_fields))
return TRUE;
records_in_scans+= (*scan_ptr)->records;
}
n_search_scans= i;
if (cpk_scan && create_fields_bitmap(param, &cpk_scan->used_fields))
return TRUE;
if (!(common->n_search_scans= n_search_scans))
return TRUE;
common->best_uses_cpk= FALSE;
common->best_cost= cutoff_cost + COST_EPS;
common->best_length= 0;
if (!(common->best_intersect=
(INDEX_SCAN_INFO **) alloc_root (param->mem_root,
sizeof(INDEX_SCAN_INFO *) *
(i + MY_TEST(cpk_scan != NULL)))))
return TRUE;
size_t calc_cost_buff_size=
Unique::get_cost_calc_buff_size((size_t)records_in_scans,
common->key_size,
common->max_memory_size);
if (!(common->buff_elems= (uint *) alloc_root(param->mem_root,
calc_cost_buff_size)))
return TRUE;
my_qsort(selected_index_scans, n_search_scans, sizeof(INDEX_SCAN_INFO *),
(qsort_cmp) cmp_intersect_index_scan);
if (cpk_scan)
{
PARTIAL_INDEX_INTERSECT_INFO curr;
set_field_bitmap_for_index_prefix(&cpk_scan->used_fields,
cpk_scan->key_info->key_part,
cpk_scan->used_key_parts);
curr.common_info= common;
curr.intersect_fields= &cpk_scan->used_fields;
curr.records= cpk_scan->records;
curr.length= 1;
for (scan_ptr=selected_index_scans; *scan_ptr; scan_ptr++)
{
ha_rows scan_records= (*scan_ptr)->records;
ha_rows records= records_in_index_intersect_extension(&curr, *scan_ptr);
(*scan_ptr)->filtered_out= records >= scan_records ?
0 : scan_records-records;
}
}
else
{
for (scan_ptr=selected_index_scans; *scan_ptr; scan_ptr++)
(*scan_ptr)->filtered_out= 0;
}
return FALSE;
}
/*
On Estimation of the Number of Records in an Index Intersection
===============================================================
Consider query Q over table t. Let C be the WHERE condition of this query,
and, idx1(a1_1,...,a1_k1) and idx2(a2_1,...,a2_k2) be some indexes defined
on table t.
Let rt1 and rt2 be the range trees extracted by the range optimizer from C
for idx1 and idx2 respectively.
Let #t be the estimate of the number of records in table t provided for the
optimizer.
Let #r1 and #r2 be the estimates of the number of records in the range trees
rt1 and rt2, respectively, obtained by the range optimizer.
We need to get an estimate for the number of records in the index
intersection of rt1 and rt2. In other words, we need to estimate the
cardinality of the set of records that are in both trees. Let's designate
this number by #r.
If we do not make any assumptions then we can only state that
#r<=MY_MIN(#r1,#r2).
With this estimate we can't say that the index intersection scan will be
cheaper than the cheapest index scan.
Let Rt1 and Rt2 be AND/OR conditions representing rt and rt2 respectively.
The probability that a record belongs to rt1 is sel(Rt1)=#r1/#t.
The probability that a record belongs to rt2 is sel(Rt2)=#r2/#t.
If we assume that the values in columns of idx1 and idx2 are independent
then #r/#t=sel(Rt1&Rt2)=sel(Rt1)*sel(Rt2)=(#r1/#t)*(#r2/#t).
So in this case we have: #r=#r1*#r2/#t.
The above assumption of independence of the columns in idx1 and idx2 means
that:
- all columns are different
- values from one column do not correlate with values from any other column.
We can't help with the case when column correlate with each other.
Yet, if they are assumed to be uncorrelated the value of #r theoretically can
be evaluated . Unfortunately this evaluation, in general, is rather complex.
Let's consider two indexes idx1:(dept, manager), idx2:(dept, building)
over table 'employee' and two range conditions over these indexes:
Rt1: dept=10 AND manager LIKE 'S%'
Rt2: dept=10 AND building LIKE 'L%'.
We can state that:
sel(Rt1&Rt2)=sel(dept=10)*sel(manager LIKE 'S%')*sel(building LIKE 'L%')
=sel(Rt1)*sel(Rt2)/sel(dept=10).
sel(Rt1/2_0:dept=10) can be estimated if we know the cardinality #r1_0 of
the range for sub-index idx1_0 (dept) of the index idx1 or the cardinality
#rt2_0 of the same range for sub-index idx2_0(dept) of the index idx2.
The current code does not make an estimate either for #rt1_0, or for #rt2_0,
but it can be adjusted to provide those numbers.
Alternatively, MY_MIN(rec_per_key) for (dept) could be used to get an upper
bound for the value of sel(Rt1&Rt2). Yet this statistics is not provided
now.
Let's consider two other indexes idx1:(dept, last_name),
idx2:(first_name, last_name) and two range conditions over these indexes:
Rt1: dept=5 AND last_name='Sm%'
Rt2: first_name='Robert' AND last_name='Sm%'.
sel(Rt1&Rt2)=sel(dept=5)*sel(last_name='Sm5')*sel(first_name='Robert')
=sel(Rt2)*sel(dept=5)
Here MY_MAX(rec_per_key) for (dept) could be used to get an upper bound for
the value of sel(Rt1&Rt2).
When the intersected indexes have different major columns, but some
minor column are common the picture may be more complicated.
Let's consider the following range conditions for the same indexes as in
the previous example:
Rt1: (Rt11: dept=5 AND last_name='So%')
OR
(Rt12: dept=7 AND last_name='Saw%')
Rt2: (Rt21: first_name='Robert' AND last_name='Saw%')
OR
(Rt22: first_name='Bob' AND last_name='So%')
Here we have:
sel(Rt1&Rt2)= sel(Rt11)*sel(Rt21)+sel(Rt22)*sel(dept=5) +
sel(Rt21)*sel(dept=7)+sel(Rt12)*sel(Rt22)
Now consider the range condition:
Rt1_0: (dept=5 OR dept=7)
For this condition we can state that:
sel(Rt1_0&Rt2)=(sel(dept=5)+sel(dept=7))*(sel(Rt21)+sel(Rt22))=
sel(dept=5)*sel(Rt21)+sel(dept=7)*sel(Rt21)+
sel(dept=5)*sel(Rt22)+sel(dept=7)*sel(Rt22)=
sel(dept=5)*sel(Rt21)+sel(Rt21)*sel(dept=7)+
sel(Rt22)*sel(dept=5)+sel(dept=7)*sel(Rt22) >
sel(Rt11)*sel(Rt21)+sel(Rt22)*sel(dept=5)+
sel(Rt21)*sel(dept=7)+sel(Rt12)*sel(Rt22) >
sel(Rt1 & Rt2)
We've just demonstrated for an example what is intuitively almost obvious
in general. We can remove the ending parts fromrange trees getting less
selective range conditions for sub-indexes.
So if not a most major component with the number k of an index idx is
encountered in the index with which we intersect we can use the sub-index
idx_k-1 that includes the components of idx up to the i-th component and
the range tree for idx_k-1 to make an upper bound estimate for the number
of records in the index intersection.
The range tree for idx_k-1 we use here is the subtree of the original range
tree for idx that contains only parts from the first k-1 components.
As it was mentioned above the range optimizer currently does not provide
an estimate for the number of records in the ranges for sub-indexes.
However, some reasonable upper bound estimate can be obtained.
Let's consider the following range tree:
Rt: (first_name='Robert' AND last_name='Saw%')
OR
(first_name='Bob' AND last_name='So%')
Let #r be the number of records in Rt. Let f_1 be the fan-out of column
last_name:
f_1 = rec_per_key[first_name]/rec_per_key[last_name].
The the number of records in the range tree:
Rt_0: (first_name='Robert' OR first_name='Bob')
for the sub-index (first_name) is not greater than MY_MAX(#r*f_1, #t).
Strictly speaking, we can state only that it's not greater than
MY_MAX(#r*max_f_1, #t), where
max_f_1= max_rec_per_key[first_name]/min_rec_per_key[last_name].
Yet, if #r/#t is big enough (and this is the case of an index intersection,
because using this index range with a single index scan is cheaper than
the cost of the intersection when #r/#t is small) then almost safely we
can use here f_1 instead of max_f_1.
The above considerations can be used in future development. Now, they are
used partly in the function that provides a rough upper bound estimate for
the number of records in an index intersection that follow below.
*/
/*
Estimate the number of records selected by an extension a partial intersection
SYNOPSIS
records_in_index_intersect_extension()
curr partial intersection plan to be extended
ext_index_scan the evaluated extension of this partial plan
DESCRIPTION
The function provides an estimate for the number of records in the
intersection of the partial index intersection curr with the index
ext_index_scan. If all intersected indexes does not have common columns
then the function returns an exact estimate (assuming there are no
correlations between values in the columns). If the intersected indexes
have common columns the function returns an upper bound for the number
of records in the intersection provided that the intersection of curr
with ext_index_scan can is expected to have less records than the expected
number of records in the partial intersection curr. In this case the
function also assigns the bitmap of the columns in the extended
intersection to ext_index_scan->used_fields.
If the function cannot expect that the number of records in the extended
intersection is less that the expected number of records #r in curr then
the function returns a number bigger than #r.
NOTES
See the comment before the desription of the function that explains the
reasoning used by this function.
RETURN
The expected number of rows in the extended index intersection
*/
static
ha_rows records_in_index_intersect_extension(PARTIAL_INDEX_INTERSECT_INFO *curr,
INDEX_SCAN_INFO *ext_index_scan)
{
KEY *key_info= ext_index_scan->key_info;
KEY_PART_INFO* key_part= key_info->key_part;
uint used_key_parts= ext_index_scan->used_key_parts;
MY_BITMAP *used_fields= &ext_index_scan->used_fields;
if (!curr->length)
{
/*
If this the first index in the intersection just mark the
fields in the used_fields bitmap and return the expected
number of records in the range scan for the index provided
by the range optimizer.
*/
set_field_bitmap_for_index_prefix(used_fields, key_part, used_key_parts);
return ext_index_scan->records;
}
uint i;
bool better_selectivity= FALSE;
ha_rows records= curr->records;
MY_BITMAP *curr_intersect_fields= curr->intersect_fields;
for (i= 0; i < used_key_parts; i++, key_part++)
{
if (bitmap_is_set(curr_intersect_fields, key_part->fieldnr-1))
break;
}
if (i)
{
ha_rows table_cardinality= curr->common_info->table_cardinality;
ha_rows ext_records= ext_index_scan->records;
if (i < used_key_parts)
{
double f1= key_info->actual_rec_per_key(i-1);
double f2= key_info->actual_rec_per_key(i);
ext_records= (ha_rows) ((double) ext_records / f2 * f1);
}
if (ext_records < table_cardinality)
{
better_selectivity= TRUE;
records= (ha_rows) ((double) records / table_cardinality *
ext_records);
bitmap_copy(used_fields, curr_intersect_fields);
key_part= key_info->key_part;
for (uint j= 0; j < used_key_parts; j++, key_part++)
bitmap_set_bit(used_fields, key_part->fieldnr-1);
}
}
return !better_selectivity ? records+1 :
!records ? 1 : records;
}
/*
Estimate the cost a binary search within disjoint cpk range intervals
Number of comparisons to check whether a cpk value satisfies
the cpk range condition = log2(cpk_scan->range_count).
*/
static inline
double get_cpk_filter_cost(ha_rows filtered_records,
INDEX_SCAN_INFO *cpk_scan,
double compare_factor)
{
return log((double) (cpk_scan->range_count+1)) / (compare_factor * M_LN2) *
filtered_records;
}
/*
Check whether a patial index intersection plan can be extended
SYNOPSIS
check_index_intersect_extension()
curr partial intersection plan to be extended
ext_index_scan a possible extension of this plan to be checked
next OUT the structure to be filled for the extended plan
DESCRIPTION
The function checks whether it makes sense to extend the index
intersection plan adding the index ext_index_scan, and, if this
the case, the function fills in the structure for the extended plan.
RETURN
TRUE if it makes sense to extend the given plan
FALSE otherwise
*/
static
bool check_index_intersect_extension(PARTIAL_INDEX_INTERSECT_INFO *curr,
INDEX_SCAN_INFO *ext_index_scan,
PARTIAL_INDEX_INTERSECT_INFO *next)
{
ha_rows records;
ha_rows records_sent_to_unique;
double cost;
ha_rows ext_index_scan_records= ext_index_scan->records;
ha_rows records_filtered_out_by_cpk= ext_index_scan->filtered_out;
COMMON_INDEX_INTERSECT_INFO *common_info= curr->common_info;
double cutoff_cost= common_info->cutoff_cost;
uint idx= curr->length;
next->index_read_cost= curr->index_read_cost+ext_index_scan->index_read_cost;
if (next->index_read_cost > cutoff_cost)
return FALSE;
if ((next->in_memory= curr->in_memory))
next->in_memory_cost= curr->in_memory_cost;
next->intersect_fields= &ext_index_scan->used_fields;
next->filtered_scans= curr->filtered_scans;
records_sent_to_unique= curr->records_sent_to_unique;
next->use_cpk_filter= FALSE;
/* Calculate the cost of using a Unique object for index intersection */
if (idx && next->in_memory)
{
/*
All rowids received from the first scan are expected in one unique tree
*/
ha_rows elems_in_tree= common_info->search_scans[0]->records-
common_info->search_scans[0]->filtered_out ;
next->in_memory_cost+= Unique::get_search_cost(elems_in_tree,
common_info->compare_factor)*
ext_index_scan_records;
cost= next->in_memory_cost;
}
else
{
uint *buff_elems= common_info->buff_elems;
uint key_size= common_info->key_size;
uint compare_factor= common_info->compare_factor;
ulonglong max_memory_size= common_info->max_memory_size;
records_sent_to_unique+= ext_index_scan_records;
cost= Unique::get_use_cost(buff_elems, (size_t) records_sent_to_unique, key_size,
max_memory_size, compare_factor, TRUE,
&next->in_memory);
if (records_filtered_out_by_cpk)
{
/* Check whether using cpk filter for this scan is beneficial */
double cost2;
bool in_memory2;
ha_rows records2= records_sent_to_unique-records_filtered_out_by_cpk;
cost2= Unique::get_use_cost(buff_elems, (size_t) records2, key_size,
max_memory_size, compare_factor, TRUE,
&in_memory2);
cost2+= get_cpk_filter_cost(ext_index_scan_records, common_info->cpk_scan,
compare_factor);
if (cost > cost2 + COST_EPS)
{
cost= cost2;
next->in_memory= in_memory2;
next->use_cpk_filter= TRUE;
records_sent_to_unique= records2;
}
}
if (next->in_memory)
next->in_memory_cost= cost;
}
if (next->use_cpk_filter)
{
next->filtered_scans.set_bit(ext_index_scan->keynr);
bitmap_union(&ext_index_scan->used_fields,
&common_info->cpk_scan->used_fields);
}
next->records_sent_to_unique= records_sent_to_unique;
records= records_in_index_intersect_extension(curr, ext_index_scan);
if (idx && records > curr->records)
return FALSE;
if (next->use_cpk_filter && curr->filtered_scans.is_clear_all())
records-= records_filtered_out_by_cpk;
next->records= records;
cost+= next->index_read_cost;
if (cost >= cutoff_cost)
return FALSE;
cost+= get_sweep_read_cost(common_info->param, records);
next->cost= cost;
next->length= curr->length+1;
return TRUE;
}
/*
Search for the cheapest extensions of range scans used to access a table
SYNOPSIS
find_index_intersect_best_extension()
curr partial intersection to evaluate all possible extension for
DESCRIPTION
The function tries to extend the partial plan curr in all possible ways
to look for a cheapest index intersection whose cost less than the
cut off value set in curr->common_info.cutoff_cost.
*/
static
void find_index_intersect_best_extension(PARTIAL_INDEX_INTERSECT_INFO *curr)
{
PARTIAL_INDEX_INTERSECT_INFO next;
COMMON_INDEX_INTERSECT_INFO *common_info= curr->common_info;
INDEX_SCAN_INFO **index_scans= common_info->search_scans;
uint idx= curr->length;
INDEX_SCAN_INFO **rem_first_index_scan_ptr= &index_scans[idx];
double cost= curr->cost;
if (cost + COST_EPS < common_info->best_cost)
{
common_info->best_cost= cost;
common_info->best_length= curr->length;
common_info->best_records= curr->records;
common_info->filtered_scans= curr->filtered_scans;
/* common_info->best_uses_cpk <=> at least one scan uses a cpk filter */
common_info->best_uses_cpk= !curr->filtered_scans.is_clear_all();
uint sz= sizeof(INDEX_SCAN_INFO *) * curr->length;
memcpy(common_info->best_intersect, common_info->search_scans, sz);
common_info->cutoff_cost= cost;
}
if (!(*rem_first_index_scan_ptr))
return;
next.common_info= common_info;
INDEX_SCAN_INFO *rem_first_index_scan= *rem_first_index_scan_ptr;
for (INDEX_SCAN_INFO **index_scan_ptr= rem_first_index_scan_ptr;
*index_scan_ptr; index_scan_ptr++)
{
*rem_first_index_scan_ptr= *index_scan_ptr;
*index_scan_ptr= rem_first_index_scan;
if (check_index_intersect_extension(curr, *rem_first_index_scan_ptr, &next))
find_index_intersect_best_extension(&next);
*index_scan_ptr= *rem_first_index_scan_ptr;
*rem_first_index_scan_ptr= rem_first_index_scan;
}
}
/*
Get the plan of the best intersection of range scans used to access a table
SYNOPSIS
get_best_index_intersect()
param common info about index ranges
tree tree of ranges for indexes than can be intersected
read_time cut off value for the evaluated plans
DESCRIPTION
The function looks for the cheapest index intersection of the range
scans to access a table. The info about the ranges for all indexes
is provided by the range optimizer and is passed through the
parameters param and tree. Any plan whose cost is greater than read_time
is rejected.
After the best index intersection is found the function constructs
the structure that manages the execution by the chosen plan.
RETURN
Pointer to the generated execution structure if a success,
0 - otherwise.
*/
static
TRP_INDEX_INTERSECT *get_best_index_intersect(PARAM *param, SEL_TREE *tree,
double read_time)
{
uint i;
uint count;
TRP_RANGE **cur_range;
TRP_RANGE **range_scans;
INDEX_SCAN_INFO *index_scan;
COMMON_INDEX_INTERSECT_INFO common;
PARTIAL_INDEX_INTERSECT_INFO init;
TRP_INDEX_INTERSECT *intersect_trp= NULL;
TABLE *table= param->table;
DBUG_ENTER("get_best_index_intersect");
if (prepare_search_best_index_intersect(param, tree, &common, &init,
read_time))
DBUG_RETURN(NULL);
find_index_intersect_best_extension(&init);
if (common.best_length <= 1 && !common.best_uses_cpk)
DBUG_RETURN(NULL);
if (common.best_uses_cpk)
{
memmove((char *) (common.best_intersect+1), (char *) common.best_intersect,
sizeof(INDEX_SCAN_INFO *) * common.best_length);
common.best_intersect[0]= common.cpk_scan;
common.best_length++;
}
count= common.best_length;
if (!(range_scans= (TRP_RANGE**)alloc_root(param->mem_root,
sizeof(TRP_RANGE *)*
count)))
DBUG_RETURN(NULL);
for (i= 0, cur_range= range_scans; i < count; i++)
{
index_scan= common.best_intersect[i];
if ((*cur_range= new (param->mem_root) TRP_RANGE(index_scan->sel_arg,
index_scan->idx, 0)))
{
TRP_RANGE *trp= *cur_range;
trp->read_cost= index_scan->index_read_cost;
trp->records= index_scan->records;
trp->is_ror= FALSE;
trp->mrr_buf_size= 0;
table->intersect_keys.set_bit(index_scan->keynr);
cur_range++;
}
}
count= tree->index_scans_end - tree->index_scans;
for (i= 0; i < count; i++)
{
index_scan= tree->index_scans[i];
if (!table->intersect_keys.is_set(index_scan->keynr))
{
for (uint j= 0; j < common.best_length; j++)
{
INDEX_SCAN_INFO *scan= common.best_intersect[j];
if (same_index_prefix(index_scan->key_info, scan->key_info,
scan->used_key_parts))
{
table->intersect_keys.set_bit(index_scan->keynr);
break;
}
}
}
}
if ((intersect_trp= new (param->mem_root)TRP_INDEX_INTERSECT))
{
intersect_trp->read_cost= common.best_cost;
intersect_trp->records= common.best_records;
intersect_trp->range_scans= range_scans;
intersect_trp->range_scans_end= cur_range;
intersect_trp->filtered_scans= common.filtered_scans;
}
DBUG_RETURN(intersect_trp);
}
typedef struct st_ror_scan_info : INDEX_SCAN_INFO
{
} ROR_SCAN_INFO;
/*
Create ROR_SCAN_INFO* structure with a single ROR scan on index idx using
sel_arg set of intervals.
SYNOPSIS
make_ror_scan()
param Parameter from test_quick_select function
idx Index of key in param->keys
sel_arg Set of intervals for a given key
RETURN
NULL - out of memory
ROR scan structure containing a scan for {idx, sel_arg}
*/
static
ROR_SCAN_INFO *make_ror_scan(const PARAM *param, int idx, SEL_ARG *sel_arg)
{
ROR_SCAN_INFO *ror_scan;
my_bitmap_map *bitmap_buf;
uint keynr;
DBUG_ENTER("make_ror_scan");
if (!(ror_scan= (ROR_SCAN_INFO*)alloc_root(param->mem_root,
sizeof(ROR_SCAN_INFO))))
DBUG_RETURN(NULL);
ror_scan->idx= idx;
ror_scan->keynr= keynr= param->real_keynr[idx];
ror_scan->key_rec_length= (param->table->key_info[keynr].key_length +
param->table->file->ref_length);
ror_scan->sel_arg= sel_arg;
ror_scan->records= param->quick_rows[keynr];
if (!(bitmap_buf= (my_bitmap_map*) alloc_root(param->mem_root,
param->fields_bitmap_size)))
DBUG_RETURN(NULL);
if (my_bitmap_init(&ror_scan->covered_fields, bitmap_buf,
param->table->s->fields, FALSE))
DBUG_RETURN(NULL);
bitmap_clear_all(&ror_scan->covered_fields);
KEY_PART_INFO *key_part= param->table->key_info[keynr].key_part;
KEY_PART_INFO *key_part_end= key_part +
param->table->key_info[keynr].user_defined_key_parts;
for (;key_part != key_part_end; ++key_part)
{
if (bitmap_is_set(¶m->needed_fields, key_part->fieldnr-1))
bitmap_set_bit(&ror_scan->covered_fields, key_part->fieldnr-1);
}
ror_scan->index_read_cost=
param->table->file->keyread_time(ror_scan->keynr, 1, ror_scan->records);
DBUG_RETURN(ror_scan);
}
/*
Compare two ROR_SCAN_INFO** by E(#records_matched) * key_record_length.
SYNOPSIS
cmp_ror_scan_info()
a ptr to first compared value
b ptr to second compared value
RETURN
-1 a < b
0 a = b
1 a > b
*/
static int cmp_ror_scan_info(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
{
double val1= rows2double((*a)->records) * (*a)->key_rec_length;
double val2= rows2double((*b)->records) * (*b)->key_rec_length;
return (val1 < val2)? -1: (val1 == val2)? 0 : 1;
}
/*
Compare two ROR_SCAN_INFO** by
(#covered fields in F desc,
#components asc,
number of first not covered component asc)
SYNOPSIS
cmp_ror_scan_info_covering()
a ptr to first compared value
b ptr to second compared value
RETURN
-1 a < b
0 a = b
1 a > b
*/
static int cmp_ror_scan_info_covering(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
{
if ((*a)->used_fields_covered > (*b)->used_fields_covered)
return -1;
if ((*a)->used_fields_covered < (*b)->used_fields_covered)
return 1;
if ((*a)->key_components < (*b)->key_components)
return -1;
if ((*a)->key_components > (*b)->key_components)
return 1;
if ((*a)->first_uncovered_field < (*b)->first_uncovered_field)
return -1;
if ((*a)->first_uncovered_field > (*b)->first_uncovered_field)
return 1;
return 0;
}
/* Auxiliary structure for incremental ROR-intersection creation */
typedef struct
{
const PARAM *param;
MY_BITMAP covered_fields; /* union of fields covered by all scans */
/*
Fraction of table records that satisfies conditions of all scans.
This is the number of full records that will be retrieved if a
non-index_only index intersection will be employed.
*/
double out_rows;
/* TRUE if covered_fields is a superset of needed_fields */
bool is_covering;
ha_rows index_records; /* sum(#records to look in indexes) */
double index_scan_costs; /* SUM(cost of 'index-only' scans) */
double total_cost;
} ROR_INTERSECT_INFO;
/*
Allocate a ROR_INTERSECT_INFO and initialize it to contain zero scans.
SYNOPSIS
ror_intersect_init()
param Parameter from test_quick_select
RETURN
allocated structure
NULL on error
*/
static
ROR_INTERSECT_INFO* ror_intersect_init(const PARAM *param)
{
ROR_INTERSECT_INFO *info;
my_bitmap_map* buf;
if (!(info= (ROR_INTERSECT_INFO*)alloc_root(param->mem_root,
sizeof(ROR_INTERSECT_INFO))))
return NULL;
info->param= param;
if (!(buf= (my_bitmap_map*) alloc_root(param->mem_root,
param->fields_bitmap_size)))
return NULL;
if (my_bitmap_init(&info->covered_fields, buf, param->table->s->fields,
FALSE))
return NULL;
info->is_covering= FALSE;
info->index_scan_costs= 0.0;
info->index_records= 0;
info->out_rows= (double) param->table->stat_records();
bitmap_clear_all(&info->covered_fields);
return info;
}
void ror_intersect_cpy(ROR_INTERSECT_INFO *dst, const ROR_INTERSECT_INFO *src)
{
dst->param= src->param;
memcpy(dst->covered_fields.bitmap, src->covered_fields.bitmap,
no_bytes_in_map(&src->covered_fields));
dst->out_rows= src->out_rows;
dst->is_covering= src->is_covering;
dst->index_records= src->index_records;
dst->index_scan_costs= src->index_scan_costs;
dst->total_cost= src->total_cost;
}
/*
Get selectivity of a ROR scan wrt ROR-intersection.
SYNOPSIS
ror_scan_selectivity()
info ROR-interection
scan ROR scan
NOTES
Suppose we have a condition on several keys
cond=k_11=c_11 AND k_12=c_12 AND ... // parts of first key
k_21=c_21 AND k_22=c_22 AND ... // parts of second key
...
k_n1=c_n1 AND k_n3=c_n3 AND ... (1) //parts of the key used by *scan
where k_ij may be the same as any k_pq (i.e. keys may have common parts).
A full row is retrieved if entire condition holds.
The recursive procedure for finding P(cond) is as follows:
First step:
Pick 1st part of 1st key and break conjunction (1) into two parts:
cond= (k_11=c_11 AND R)
Here R may still contain condition(s) equivalent to k_11=c_11.
Nevertheless, the following holds:
P(k_11=c_11 AND R) = P(k_11=c_11) * P(R | k_11=c_11).
Mark k_11 as fixed field (and satisfied condition) F, save P(F),
save R to be cond and proceed to recursion step.
Recursion step:
We have a set of fixed fields/satisfied conditions) F, probability P(F),
and remaining conjunction R
Pick next key part on current key and its condition "k_ij=c_ij".
We will add "k_ij=c_ij" into F and update P(F).
Lets denote k_ij as t, R = t AND R1, where R1 may still contain t. Then
P((t AND R1)|F) = P(t|F) * P(R1|t|F) = P(t|F) * P(R1|(t AND F)) (2)
(where '|' mean conditional probability, not "or")
Consider the first multiplier in (2). One of the following holds:
a) F contains condition on field used in t (i.e. t AND F = F).
Then P(t|F) = 1
b) F doesn't contain condition on field used in t. Then F and t are
considered independent.
P(t|F) = P(t|(fields_before_t_in_key AND other_fields)) =
= P(t|fields_before_t_in_key).
P(t|fields_before_t_in_key) = #records(fields_before_t_in_key) /
#records(fields_before_t_in_key, t)
The second multiplier is calculated by applying this step recursively.
IMPLEMENTATION
This function calculates the result of application of the "recursion step"
described above for all fixed key members of a single key, accumulating set
of covered fields, selectivity, etc.
The calculation is conducted as follows:
Lets denote #records(keypart1, ... keypartK) as n_k. We need to calculate
n_{k1} n_{k2}
--------- * --------- * .... (3)
n_{k1-1} n_{k2-1}
where k1,k2,... are key parts which fields were not yet marked as fixed
( this is result of application of option b) of the recursion step for
parts of a single key).
Since it is reasonable to expect that most of the fields are not marked
as fixed, we calculate (3) as
n_{i1} n_{i2}
(3) = n_{max_key_part} / ( --------- * --------- * .... )
n_{i1-1} n_{i2-1}
where i1,i2, .. are key parts that were already marked as fixed.
In order to minimize number of expensive records_in_range calls we group
and reduce adjacent fractions.
RETURN
Selectivity of given ROR scan.
*/
static double ror_scan_selectivity(const ROR_INTERSECT_INFO *info,
const ROR_SCAN_INFO *scan)
{
double selectivity_mult= 1.0;
KEY_PART_INFO *key_part= info->param->table->key_info[scan->keynr].key_part;
uchar key_val[MAX_KEY_LENGTH+MAX_FIELD_WIDTH]; /* key values tuple */
uchar *key_ptr= key_val;
SEL_ARG *sel_arg, *tuple_arg= NULL;
key_part_map keypart_map= 0;
bool cur_covered;
bool prev_covered= MY_TEST(bitmap_is_set(&info->covered_fields,
key_part->fieldnr - 1));
key_range min_range;
key_range max_range;
min_range.key= key_val;
min_range.flag= HA_READ_KEY_EXACT;
max_range.key= key_val;
max_range.flag= HA_READ_AFTER_KEY;
ha_rows prev_records= info->param->table->stat_records();
DBUG_ENTER("ror_scan_selectivity");
for (sel_arg= scan->sel_arg; sel_arg;
sel_arg= sel_arg->next_key_part)
{
DBUG_PRINT("info",("sel_arg step"));
cur_covered= MY_TEST(bitmap_is_set(&info->covered_fields,
key_part[sel_arg->part].fieldnr - 1));
if (cur_covered != prev_covered)
{
/* create (part1val, ..., part{n-1}val) tuple. */
ha_rows records;
if (!tuple_arg)
{
tuple_arg= scan->sel_arg;
/* Here we use the length of the first key part */
tuple_arg->store_min(key_part->store_length, &key_ptr, 0);
keypart_map= 1;
}
while (tuple_arg->next_key_part != sel_arg)
{
tuple_arg= tuple_arg->next_key_part;
tuple_arg->store_min(key_part[tuple_arg->part].store_length,
&key_ptr, 0);
keypart_map= (keypart_map << 1) | 1;
}
min_range.length= max_range.length= (size_t) (key_ptr - key_val);
min_range.keypart_map= max_range.keypart_map= keypart_map;
records= (info->param->table->file->
records_in_range(scan->keynr, &min_range, &max_range));
if (cur_covered)
{
/* uncovered -> covered */
double tmp= rows2double(records)/rows2double(prev_records);
DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
selectivity_mult *= tmp;
prev_records= HA_POS_ERROR;
}
else
{
/* covered -> uncovered */
prev_records= records;
}
}
prev_covered= cur_covered;
}
if (!prev_covered)
{
double tmp= rows2double(info->param->quick_rows[scan->keynr]) /
rows2double(prev_records);
DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
selectivity_mult *= tmp;
}
DBUG_PRINT("info", ("Returning multiplier: %g", selectivity_mult));
DBUG_RETURN(selectivity_mult);
}
/*
Check if adding a ROR scan to a ROR-intersection reduces its cost of
ROR-intersection and if yes, update parameters of ROR-intersection,
including its cost.
SYNOPSIS
ror_intersect_add()
param Parameter from test_quick_select
info ROR-intersection structure to add the scan to.
ror_scan ROR scan info to add.
is_cpk_scan If TRUE, add the scan as CPK scan (this can be inferred
from other parameters and is passed separately only to
avoid duplicating the inference code)
NOTES
Adding a ROR scan to ROR-intersect "makes sense" iff the cost of ROR-
intersection decreases. The cost of ROR-intersection is calculated as
follows:
cost= SUM_i(key_scan_cost_i) + cost_of_full_rows_retrieval
When we add a scan the first increases and the second decreases.
cost_of_full_rows_retrieval=
(union of indexes used covers all needed fields) ?
cost_of_sweep_read(E(rows_to_retrieve), rows_in_table) :
0
E(rows_to_retrieve) = #rows_in_table * ror_scan_selectivity(null, scan1) *
ror_scan_selectivity({scan1}, scan2) * ... *
ror_scan_selectivity({scan1,...}, scanN).
RETURN
TRUE ROR scan added to ROR-intersection, cost updated.
FALSE It doesn't make sense to add this ROR scan to this ROR-intersection.
*/
static bool ror_intersect_add(ROR_INTERSECT_INFO *info,
ROR_SCAN_INFO* ror_scan, bool is_cpk_scan)
{
double selectivity_mult= 1.0;
DBUG_ENTER("ror_intersect_add");
DBUG_PRINT("info", ("Current out_rows= %g", info->out_rows));
DBUG_PRINT("info", ("Adding scan on %s",
info->param->table->key_info[ror_scan->keynr].name));
DBUG_PRINT("info", ("is_cpk_scan: %d",is_cpk_scan));
selectivity_mult = ror_scan_selectivity(info, ror_scan);
if (selectivity_mult == 1.0)
{
/* Don't add this scan if it doesn't improve selectivity. */
DBUG_PRINT("info", ("The scan doesn't improve selectivity."));
DBUG_RETURN(FALSE);
}
info->out_rows *= selectivity_mult;
if (is_cpk_scan)
{
/*
CPK scan is used to filter out rows. We apply filtering for
each record of every scan. Assuming 1/TIME_FOR_COMPARE_ROWID
per check this gives us:
*/
info->index_scan_costs += rows2double(info->index_records) /
TIME_FOR_COMPARE_ROWID;
}
else
{
info->index_records += info->param->quick_rows[ror_scan->keynr];
info->index_scan_costs += ror_scan->index_read_cost;
bitmap_union(&info->covered_fields, &ror_scan->covered_fields);
if (!info->is_covering && bitmap_is_subset(&info->param->needed_fields,
&info->covered_fields))
{
DBUG_PRINT("info", ("ROR-intersect is covering now"));
info->is_covering= TRUE;
}
}
info->total_cost= info->index_scan_costs;
DBUG_PRINT("info", ("info->total_cost: %g", info->total_cost));
if (!info->is_covering)
{
info->total_cost +=
get_sweep_read_cost(info->param, double2rows(info->out_rows));
DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
}
DBUG_PRINT("info", ("New out_rows: %g", info->out_rows));
DBUG_PRINT("info", ("New cost: %g, %scovering", info->total_cost,
info->is_covering?"" : "non-"));
DBUG_RETURN(TRUE);
}
/*
Get best ROR-intersection plan using non-covering ROR-intersection search
algorithm. The returned plan may be covering.
SYNOPSIS
get_best_ror_intersect()
param Parameter from test_quick_select function.
tree Transformed restriction condition to be used to look
for ROR scans.
read_time Do not return read plans with cost > read_time.
are_all_covering [out] set to TRUE if union of all scans covers all
fields needed by the query (and it is possible to build
a covering ROR-intersection)
NOTES
get_key_scans_params must be called before this function can be called.
When this function is called by ROR-union construction algorithm it
assumes it is building an uncovered ROR-intersection (and thus # of full
records to be retrieved is wrong here). This is a hack.
IMPLEMENTATION
The approximate best non-covering plan search algorithm is as follows:
find_min_ror_intersection_scan()
{
R= select all ROR scans;
order R by (E(#records_matched) * key_record_length).
S= first(R); -- set of scans that will be used for ROR-intersection
R= R-first(S);
min_cost= cost(S);
min_scan= make_scan(S);
while (R is not empty)
{
firstR= R - first(R);
if (!selectivity(S + firstR < selectivity(S)))
continue;
S= S + first(R);
if (cost(S) < min_cost)
{
min_cost= cost(S);
min_scan= make_scan(S);
}
}
return min_scan;
}
See ror_intersect_add function for ROR intersection costs.
Special handling for Clustered PK scans
Clustered PK contains all table fields, so using it as a regular scan in
index intersection doesn't make sense: a range scan on CPK will be less
expensive in this case.
Clustered PK scan has special handling in ROR-intersection: it is not used
to retrieve rows, instead its condition is used to filter row references
we get from scans on other keys.
RETURN
ROR-intersection table read plan
NULL if out of memory or no suitable plan found.
*/
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
double read_time,
bool *are_all_covering)
{
uint idx;
double min_cost= DBL_MAX;
DBUG_ENTER("get_best_ror_intersect");
if ((tree->n_ror_scans < 2) || !param->table->stat_records() ||
!optimizer_flag(param->thd, OPTIMIZER_SWITCH_INDEX_MERGE_INTERSECT))
DBUG_RETURN(NULL);
/*
Step1: Collect ROR-able SEL_ARGs and create ROR_SCAN_INFO for each of
them. Also find and save clustered PK scan if there is one.
*/
ROR_SCAN_INFO **cur_ror_scan;
ROR_SCAN_INFO *cpk_scan= NULL;
uint cpk_no;
if (!(tree->ror_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
sizeof(ROR_SCAN_INFO*)*
param->keys)))
return NULL;
cpk_no= ((param->table->file->primary_key_is_clustered()) ?
param->table->s->primary_key : MAX_KEY);
for (idx= 0, cur_ror_scan= tree->ror_scans; idx < param->keys; idx++)
{
ROR_SCAN_INFO *scan;
uint key_no;
if (!tree->ror_scans_map.is_set(idx))
continue;
key_no= param->real_keynr[idx];
if (key_no != cpk_no &&
param->table->file->index_flags(key_no,0,0) & HA_CLUSTERED_INDEX)
{
/* Ignore clustering keys */
tree->n_ror_scans--;
continue;
}
if (!(scan= make_ror_scan(param, idx, tree->keys[idx])))
return NULL;
if (key_no == cpk_no)
{
cpk_scan= scan;
tree->n_ror_scans--;
}
else
*(cur_ror_scan++)= scan;
}
tree->ror_scans_end= cur_ror_scan;
DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "original",
tree->ror_scans,
tree->ror_scans_end););
/*
Ok, [ror_scans, ror_scans_end) is array of ptrs to initialized
ROR_SCAN_INFO's.
Step 2: Get best ROR-intersection using an approximate algorithm.
*/
my_qsort(tree->ror_scans, tree->n_ror_scans, sizeof(ROR_SCAN_INFO*),
(qsort_cmp)cmp_ror_scan_info);
DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "ordered",
tree->ror_scans,
tree->ror_scans_end););
ROR_SCAN_INFO **intersect_scans; /* ROR scans used in index intersection */
ROR_SCAN_INFO **intersect_scans_end;
if (!(intersect_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
sizeof(ROR_SCAN_INFO*)*
tree->n_ror_scans)))
return NULL;
intersect_scans_end= intersect_scans;
/* Create and incrementally update ROR intersection. */
ROR_INTERSECT_INFO *intersect, *intersect_best;
if (!(intersect= ror_intersect_init(param)) ||
!(intersect_best= ror_intersect_init(param)))
return NULL;
/* [intersect_scans,intersect_scans_best) will hold the best intersection */
ROR_SCAN_INFO **intersect_scans_best;
cur_ror_scan= tree->ror_scans;
intersect_scans_best= intersect_scans;
while (cur_ror_scan != tree->ror_scans_end && !intersect->is_covering)
{
/* S= S + first(R); R= R - first(R); */
if (!ror_intersect_add(intersect, *cur_ror_scan, FALSE))
{
cur_ror_scan++;
continue;
}
*(intersect_scans_end++)= *(cur_ror_scan++);
if (intersect->total_cost < min_cost)
{
/* Local minimum found, save it */
ror_intersect_cpy(intersect_best, intersect);
intersect_scans_best= intersect_scans_end;
min_cost = intersect->total_cost;
}
}
if (intersect_scans_best == intersect_scans)
{
DBUG_PRINT("info", ("None of scans increase selectivity"));
DBUG_RETURN(NULL);
}
DBUG_EXECUTE("info",print_ror_scans_arr(param->table,
"best ROR-intersection",
intersect_scans,
intersect_scans_best););
*are_all_covering= intersect->is_covering;
uint best_num= intersect_scans_best - intersect_scans;
ror_intersect_cpy(intersect, intersect_best);
/*
Ok, found the best ROR-intersection of non-CPK key scans.
Check if we should add a CPK scan. If the obtained ROR-intersection is
covering, it doesn't make sense to add CPK scan.
*/
if (cpk_scan && !intersect->is_covering)
{
if (ror_intersect_add(intersect, cpk_scan, TRUE) &&
(intersect->total_cost < min_cost))
intersect_best= intersect; //just set pointer here
}
else
cpk_scan= 0; // Don't use cpk_scan
/* Ok, return ROR-intersect plan if we have found one */
TRP_ROR_INTERSECT *trp= NULL;
if (min_cost < read_time && (cpk_scan || best_num > 1))
{
if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
DBUG_RETURN(trp);
if (!(trp->first_scan=
(ROR_SCAN_INFO**)alloc_root(param->mem_root,
sizeof(ROR_SCAN_INFO*)*best_num)))
DBUG_RETURN(NULL);
memcpy(trp->first_scan, intersect_scans, best_num*sizeof(ROR_SCAN_INFO*));
trp->last_scan= trp->first_scan + best_num;
trp->is_covering= intersect_best->is_covering;
trp->read_cost= intersect_best->total_cost;
/* Prevent divisons by zero */
ha_rows best_rows = double2rows(intersect_best->out_rows);
if (!best_rows)
best_rows= 1;
set_if_smaller(param->table->quick_condition_rows, best_rows);
trp->records= best_rows;
trp->index_scan_costs= intersect_best->index_scan_costs;
trp->cpk_scan= cpk_scan;
DBUG_PRINT("info", ("Returning non-covering ROR-intersect plan:"
"cost %g, records %lu",
trp->read_cost, (ulong) trp->records));
}
DBUG_RETURN(trp);
}
/*
Get best covering ROR-intersection.
SYNOPSIS
get_best_ntersectcovering_ror_intersect()
param Parameter from test_quick_select function.
tree SEL_TREE with sets of intervals for different keys.
read_time Don't return table read plans with cost > read_time.
RETURN
Best covering ROR-intersection plan
NULL if no plan found.
NOTES
get_best_ror_intersect must be called for a tree before calling this
function for it.
This function invalidates tree->ror_scans member values.
The following approximate algorithm is used:
I=set of all covering indexes
F=set of all fields to cover
S={}
do
{
Order I by (#covered fields in F desc,
#components asc,
number of first not covered component asc);
F=F-covered by first(I);
S=S+first(I);
I=I-first(I);
} while F is not empty.
*/
static
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
SEL_TREE *tree,
double read_time)
{
ROR_SCAN_INFO **ror_scan_mark;
ROR_SCAN_INFO **ror_scans_end= tree->ror_scans_end;
DBUG_ENTER("get_best_covering_ror_intersect");
if (!optimizer_flag(param->thd, OPTIMIZER_SWITCH_INDEX_MERGE_INTERSECT))
DBUG_RETURN(NULL);
for (ROR_SCAN_INFO **scan= tree->ror_scans; scan != ror_scans_end; ++scan)
(*scan)->key_components=
param->table->key_info[(*scan)->keynr].user_defined_key_parts;
/*
Run covering-ROR-search algorithm.
Assume set I is [ror_scan .. ror_scans_end)
*/
/*I=set of all covering indexes */
ror_scan_mark= tree->ror_scans;
MY_BITMAP *covered_fields= ¶m->tmp_covered_fields;
if (!covered_fields->bitmap)
covered_fields->bitmap= (my_bitmap_map*)alloc_root(param->mem_root,
param->fields_bitmap_size);
if (!covered_fields->bitmap ||
my_bitmap_init(covered_fields, covered_fields->bitmap,
param->table->s->fields, FALSE))
DBUG_RETURN(0);
bitmap_clear_all(covered_fields);
double total_cost= 0.0f;
ha_rows records=0;
bool all_covered;
DBUG_PRINT("info", ("Building covering ROR-intersection"));
DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
"building covering ROR-I",
ror_scan_mark, ror_scans_end););
do
{
/*
Update changed sorting info:
#covered fields,
number of first not covered component
Calculate and save these values for each of remaining scans.
*/
for (ROR_SCAN_INFO **scan= ror_scan_mark; scan != ror_scans_end; ++scan)
{
bitmap_subtract(&(*scan)->covered_fields, covered_fields);
(*scan)->used_fields_covered=
bitmap_bits_set(&(*scan)->covered_fields);
(*scan)->first_uncovered_field=
bitmap_get_first(&(*scan)->covered_fields);
}
my_qsort(ror_scan_mark, ror_scans_end-ror_scan_mark, sizeof(ROR_SCAN_INFO*),
(qsort_cmp)cmp_ror_scan_info_covering);
DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
"remaining scans",
ror_scan_mark, ror_scans_end););
/* I=I-first(I) */
total_cost += (*ror_scan_mark)->index_read_cost;
records += (*ror_scan_mark)->records;
DBUG_PRINT("info", ("Adding scan on %s",
param->table->key_info[(*ror_scan_mark)->keynr].name));
if (total_cost > read_time)
DBUG_RETURN(NULL);
/* F=F-covered by first(I) */
bitmap_union(covered_fields, &(*ror_scan_mark)->covered_fields);
all_covered= bitmap_is_subset(¶m->needed_fields, covered_fields);
} while ((++ror_scan_mark < ror_scans_end) && !all_covered);
if (!all_covered || (ror_scan_mark - tree->ror_scans) == 1)
DBUG_RETURN(NULL);
/*
Ok, [tree->ror_scans .. ror_scan) holds covering index_intersection with
cost total_cost.
*/
DBUG_PRINT("info", ("Covering ROR-intersect scans cost: %g", total_cost));
DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
"creating covering ROR-intersect",
tree->ror_scans, ror_scan_mark););
/* Add priority queue use cost. */
total_cost += rows2double(records)*
log((double)(ror_scan_mark - tree->ror_scans)) /
(TIME_FOR_COMPARE_ROWID * M_LN2);
DBUG_PRINT("info", ("Covering ROR-intersect full cost: %g", total_cost));
if (total_cost > read_time)
DBUG_RETURN(NULL);
TRP_ROR_INTERSECT *trp;
if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
DBUG_RETURN(trp);
uint best_num= (ror_scan_mark - tree->ror_scans);
if (!(trp->first_scan= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
sizeof(ROR_SCAN_INFO*)*
best_num)))
DBUG_RETURN(NULL);
memcpy(trp->first_scan, tree->ror_scans, best_num*sizeof(ROR_SCAN_INFO*));
trp->last_scan= trp->first_scan + best_num;
trp->is_covering= TRUE;
trp->read_cost= total_cost;
trp->records= records;
trp->cpk_scan= NULL;
set_if_smaller(param->table->quick_condition_rows, records);
DBUG_PRINT("info",
("Returning covering ROR-intersect plan: cost %g, records %lu",
trp->read_cost, (ulong) trp->records));
DBUG_RETURN(trp);
}
/*
Get best "range" table read plan for given SEL_TREE.
Also update PARAM members and store ROR scans info in the SEL_TREE.
SYNOPSIS
get_key_scans_params
param parameters from test_quick_select
tree make range select for this SEL_TREE
index_read_must_be_used if TRUE, assume 'index only' option will be set
(except for clustered PK indexes)
read_time don't create read plans with cost > read_time.
RETURN
Best range read plan
NULL if no plan found or error occurred
*/
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
bool index_read_must_be_used,
bool update_tbl_stats,
double read_time)
{
uint idx;
SEL_ARG **key,**end, **key_to_read= NULL;
ha_rows UNINIT_VAR(best_records); /* protected by key_to_read */
uint UNINIT_VAR(best_mrr_flags), /* protected by key_to_read */
UNINIT_VAR(best_buf_size); /* protected by key_to_read */
TRP_RANGE* read_plan= NULL;
DBUG_ENTER("get_key_scans_params");
/*
Note that there may be trees that have type SEL_TREE::KEY but contain no
key reads at all, e.g. tree for expression "key1 is not null" where key1
is defined as "not null".
*/
DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->keys_map,
"tree scans"););
tree->ror_scans_map.clear_all();
tree->n_ror_scans= 0;
tree->index_scans= 0;
if (!tree->keys_map.is_clear_all())
{
tree->index_scans=
(INDEX_SCAN_INFO **) alloc_root(param->mem_root,
sizeof(INDEX_SCAN_INFO *) * param->keys);
}
tree->index_scans_end= tree->index_scans;
for (idx= 0,key=tree->keys, end=key+param->keys; key != end; key++,idx++)
{
if (*key)
{
ha_rows found_records;
Cost_estimate cost;
double found_read_time;
uint mrr_flags, buf_size;
INDEX_SCAN_INFO *index_scan;
uint keynr= param->real_keynr[idx];
if ((*key)->type == SEL_ARG::MAYBE_KEY ||
(*key)->maybe_flag)
param->needed_reg->set_bit(keynr);
bool read_index_only= index_read_must_be_used ? TRUE :
(bool) param->table->covering_keys.is_set(keynr);
found_records= check_quick_select(param, idx, read_index_only, *key,
update_tbl_stats, &mrr_flags,
&buf_size, &cost);
if (found_records != HA_POS_ERROR && tree->index_scans &&
(index_scan= (INDEX_SCAN_INFO *)alloc_root(param->mem_root,
sizeof(INDEX_SCAN_INFO))))
{
index_scan->idx= idx;
index_scan->keynr= keynr;
index_scan->key_info= ¶m->table->key_info[keynr];
index_scan->used_key_parts= param->max_key_part+1;
index_scan->range_count= param->range_count;
index_scan->records= found_records;
index_scan->sel_arg= *key;
*tree->index_scans_end++= index_scan;
}
if ((found_records != HA_POS_ERROR) && param->is_ror_scan)
{
tree->n_ror_scans++;
tree->ror_scans_map.set_bit(idx);
}
if (found_records != HA_POS_ERROR &&
read_time > (found_read_time= cost.total_cost()))
{
read_time= found_read_time;
best_records= found_records;
key_to_read= key;
best_mrr_flags= mrr_flags;
best_buf_size= buf_size;
}
}
}
DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->ror_scans_map,
"ROR scans"););
if (key_to_read)
{
idx= key_to_read - tree->keys;
if ((read_plan= new (param->mem_root) TRP_RANGE(*key_to_read, idx,
best_mrr_flags)))
{
read_plan->records= best_records;
read_plan->is_ror= tree->ror_scans_map.is_set(idx);
read_plan->read_cost= read_time;
read_plan->mrr_buf_size= best_buf_size;
DBUG_PRINT("info",
("Returning range plan for key %s, cost %g, records %lu",
param->table->key_info[param->real_keynr[idx]].name,
read_plan->read_cost, (ulong) read_plan->records));
}
}
else
DBUG_PRINT("info", ("No 'range' table read plan found"));
DBUG_RETURN(read_plan);
}
QUICK_SELECT_I *TRP_INDEX_MERGE::make_quick(PARAM *param,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
QUICK_INDEX_MERGE_SELECT *quick_imerge;
QUICK_RANGE_SELECT *quick;
/* index_merge always retrieves full rows, ignore retrieve_full_rows */
if (!(quick_imerge= new QUICK_INDEX_MERGE_SELECT(param->thd, param->table)))
return NULL;
quick_imerge->records= records;
quick_imerge->read_time= read_cost;
for (TRP_RANGE **range_scan= range_scans; range_scan != range_scans_end;
range_scan++)
{
if (!(quick= (QUICK_RANGE_SELECT*)
((*range_scan)->make_quick(param, FALSE, &quick_imerge->alloc)))||
quick_imerge->push_quick_back(quick))
{
delete quick;
delete quick_imerge;
return NULL;
}
}
return quick_imerge;
}
QUICK_SELECT_I *TRP_INDEX_INTERSECT::make_quick(PARAM *param,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
QUICK_INDEX_INTERSECT_SELECT *quick_intersect;
QUICK_RANGE_SELECT *quick;
/* index_merge always retrieves full rows, ignore retrieve_full_rows */
if (!(quick_intersect= new QUICK_INDEX_INTERSECT_SELECT(param->thd, param->table)))
return NULL;
quick_intersect->records= records;
quick_intersect->read_time= read_cost;
quick_intersect->filtered_scans= filtered_scans;
for (TRP_RANGE **range_scan= range_scans; range_scan != range_scans_end;
range_scan++)
{
if (!(quick= (QUICK_RANGE_SELECT*)
((*range_scan)->make_quick(param, FALSE, &quick_intersect->alloc)))||
quick_intersect->push_quick_back(quick))
{
delete quick;
delete quick_intersect;
return NULL;
}
}
return quick_intersect;
}
QUICK_SELECT_I *TRP_ROR_INTERSECT::make_quick(PARAM *param,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
QUICK_ROR_INTERSECT_SELECT *quick_intrsect;
QUICK_RANGE_SELECT *quick;
DBUG_ENTER("TRP_ROR_INTERSECT::make_quick");
MEM_ROOT *alloc;
if ((quick_intrsect=
new QUICK_ROR_INTERSECT_SELECT(param->thd, param->table,
(retrieve_full_rows? (!is_covering) :
FALSE),
parent_alloc)))
{
DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
"creating ROR-intersect",
first_scan, last_scan););
alloc= parent_alloc? parent_alloc: &quick_intrsect->alloc;
for (; first_scan != last_scan;++first_scan)
{
if (!(quick= get_quick_select(param, (*first_scan)->idx,
(*first_scan)->sel_arg,
HA_MRR_USE_DEFAULT_IMPL | HA_MRR_SORTED,
0, alloc)) ||
quick_intrsect->push_quick_back(alloc, quick))
{
delete quick_intrsect;
DBUG_RETURN(NULL);
}
}
if (cpk_scan)
{
if (!(quick= get_quick_select(param, cpk_scan->idx,
cpk_scan->sel_arg,
HA_MRR_USE_DEFAULT_IMPL | HA_MRR_SORTED,
0, alloc)))
{
delete quick_intrsect;
DBUG_RETURN(NULL);
}
quick->file= NULL;
quick_intrsect->cpk_quick= quick;
}
quick_intrsect->records= records;
quick_intrsect->read_time= read_cost;
}
DBUG_RETURN(quick_intrsect);
}
QUICK_SELECT_I *TRP_ROR_UNION::make_quick(PARAM *param,
bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
QUICK_ROR_UNION_SELECT *quick_roru;
TABLE_READ_PLAN **scan;
QUICK_SELECT_I *quick;
DBUG_ENTER("TRP_ROR_UNION::make_quick");
/*
It is impossible to construct a ROR-union that will not retrieve full
rows, ignore retrieve_full_rows parameter.
*/
if ((quick_roru= new QUICK_ROR_UNION_SELECT(param->thd, param->table)))
{
for (scan= first_ror; scan != last_ror; scan++)
{
if (!(quick= (*scan)->make_quick(param, FALSE, &quick_roru->alloc)) ||
quick_roru->push_quick_back(quick))
DBUG_RETURN(NULL);
}
quick_roru->records= records;
quick_roru->read_time= read_cost;
}
DBUG_RETURN(quick_roru);
}
/*
Build a SEL_TREE for <> or NOT BETWEEN predicate
SYNOPSIS
get_ne_mm_tree()
param PARAM from SQL_SELECT::test_quick_select
cond_func item for the predicate
field field in the predicate
lt_value constant that field should be smaller
gt_value constant that field should be greaterr
cmp_type compare type for the field
RETURN
# Pointer to tree built tree
0 on error
*/
static SEL_TREE *get_ne_mm_tree(RANGE_OPT_PARAM *param, Item_func *cond_func,
Field *field,
Item *lt_value, Item *gt_value,
Item_result cmp_type)
{
SEL_TREE *tree;
tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
lt_value, cmp_type);
if (tree)
{
tree= tree_or(param, tree, get_mm_parts(param, cond_func, field,
Item_func::GT_FUNC,
gt_value, cmp_type));
}
return tree;
}
/*
Build a SEL_TREE for a simple predicate
SYNOPSIS
get_func_mm_tree()
param PARAM from SQL_SELECT::test_quick_select
cond_func item for the predicate
field field in the predicate
value constant in the predicate
cmp_type compare type for the field
inv TRUE <> NOT cond_func is considered
(makes sense only when cond_func is BETWEEN or IN)
RETURN
Pointer to the tree built tree
*/
static SEL_TREE *get_func_mm_tree(RANGE_OPT_PARAM *param, Item_func *cond_func,
Field *field, Item *value,
Item_result cmp_type, bool inv)
{
SEL_TREE *tree= 0;
DBUG_ENTER("get_func_mm_tree");
switch (cond_func->functype()) {
case Item_func::NE_FUNC:
tree= get_ne_mm_tree(param, cond_func, field, value, value, cmp_type);
break;
case Item_func::BETWEEN:
{
if (!value)
{
if (inv)
{
tree= get_ne_mm_tree(param, cond_func, field, cond_func->arguments()[1],
cond_func->arguments()[2], cmp_type);
}
else
{
tree= get_mm_parts(param, cond_func, field, Item_func::GE_FUNC,
cond_func->arguments()[1],cmp_type);
if (tree)
{
tree= tree_and(param, tree, get_mm_parts(param, cond_func, field,
Item_func::LE_FUNC,
cond_func->arguments()[2],
cmp_type));
}
}
}
else
tree= get_mm_parts(param, cond_func, field,
(inv ?
(value == (Item*)1 ? Item_func::GT_FUNC :
Item_func::LT_FUNC):
(value == (Item*)1 ? Item_func::LE_FUNC :
Item_func::GE_FUNC)),
cond_func->arguments()[0], cmp_type);
break;
}
case Item_func::IN_FUNC:
{
Item_func_in *func=(Item_func_in*) cond_func;
/*
Array for IN() is constructed when all values have the same result
type. Tree won't be built for values with different result types,
so we check it here to avoid unnecessary work.
*/
if (!func->arg_types_compatible)
break;
if (inv)
{
if (func->array && func->array->result_type() != ROW_RESULT)
{
/*
We get here for conditions in form "t.key NOT IN (c1, c2, ...)",
where c{i} are constants. Our goal is to produce a SEL_TREE that
represents intervals:
($MIN<t.key<c1) OR (c1<t.key<c2) OR (c2<t.key<c3) OR ... (*)
where $MIN is either "-inf" or NULL.
The most straightforward way to produce it is to convert NOT IN
into "(t.key != c1) AND (t.key != c2) AND ... " and let the range
analyzer to build SEL_TREE from that. The problem is that the
range analyzer will use O(N^2) memory (which is probably a bug),
and people do use big NOT IN lists (e.g. see BUG#15872, BUG#21282),
will run out of memory.
Another problem with big lists like (*) is that a big list is
unlikely to produce a good "range" access, while considering that
range access will require expensive CPU calculations (and for
MyISAM even index accesses). In short, big NOT IN lists are rarely
worth analyzing.
Considering the above, we'll handle NOT IN as follows:
* if the number of entries in the NOT IN list is less than
NOT_IN_IGNORE_THRESHOLD, construct the SEL_TREE (*) manually.
* Otherwise, don't produce a SEL_TREE.
*/
#define NOT_IN_IGNORE_THRESHOLD 1000
MEM_ROOT *tmp_root= param->mem_root;
param->thd->mem_root= param->old_root;
/*
Create one Item_type constant object. We'll need it as
get_mm_parts only accepts constant values wrapped in Item_Type
objects.
We create the Item on param->mem_root which points to
per-statement mem_root (while thd->mem_root is currently pointing
to mem_root local to range optimizer).
*/
Item *value_item= func->array->create_item();
param->thd->mem_root= tmp_root;
if (func->array->count > NOT_IN_IGNORE_THRESHOLD || !value_item)
break;
/* Get a SEL_TREE for "(-inf|NULL) < X < c_0" interval. */
uint i=0;
do
{
func->array->value_to_item(i, value_item);
tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
value_item, cmp_type);
if (!tree)
break;
i++;
} while (i < func->array->count && tree->type == SEL_TREE::IMPOSSIBLE);
if (!tree || tree->type == SEL_TREE::IMPOSSIBLE)
{
/* We get here in cases like "t.unsigned NOT IN (-1,-2,-3) */
tree= NULL;
break;
}
SEL_TREE *tree2;
for (; i < func->array->count; i++)
{
if (func->array->compare_elems(i, i-1))
{
/* Get a SEL_TREE for "-inf < X < c_i" interval */
func->array->value_to_item(i, value_item);
tree2= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
value_item, cmp_type);
if (!tree2)
{
tree= NULL;
break;
}
/* Change all intervals to be "c_{i-1} < X < c_i" */
for (uint idx= 0; idx < param->keys; idx++)
{
SEL_ARG *new_interval, *last_val;
if (((new_interval= tree2->keys[idx])) &&
(tree->keys[idx]) &&
((last_val= tree->keys[idx]->last())))
{
new_interval->min_value= last_val->max_value;
new_interval->min_flag= NEAR_MIN;
/*
If the interval is over a partial keypart, the
interval must be "c_{i-1} <= X < c_i" instead of
"c_{i-1} < X < c_i". Reason:
Consider a table with a column "my_col VARCHAR(3)",
and an index with definition
"INDEX my_idx my_col(1)". If the table contains rows
with my_col values "f" and "foo", the index will not
distinguish the two rows.
Note that tree_or() below will effectively merge
this range with the range created for c_{i-1} and
we'll eventually end up with only one range:
"NULL < X".
Partitioning indexes are never partial.
*/
if (param->using_real_indexes)
{
const KEY key=
param->table->key_info[param->real_keynr[idx]];
const KEY_PART_INFO *kpi= key.key_part + new_interval->part;
if (kpi->key_part_flag & HA_PART_KEY_SEG)
new_interval->min_flag= 0;
}
}
}
/*
The following doesn't try to allocate memory so no need to
check for NULL.
*/
tree= tree_or(param, tree, tree2);
}
}
if (tree && tree->type != SEL_TREE::IMPOSSIBLE)
{
/*
Get the SEL_TREE for the last "c_last < X < +inf" interval
(value_item cotains c_last already)
*/
tree2= get_mm_parts(param, cond_func, field, Item_func::GT_FUNC,
value_item, cmp_type);
tree= tree_or(param, tree, tree2);
}
}
else
{
tree= get_ne_mm_tree(param, cond_func, field,
func->arguments()[1], func->arguments()[1],
cmp_type);
if (tree)
{
Item **arg, **end;
for (arg= func->arguments()+2, end= arg+func->argument_count()-2;
arg < end ; arg++)
{
tree= tree_and(param, tree, get_ne_mm_tree(param, cond_func, field,
*arg, *arg, cmp_type));
}
}
}
}
else
{
tree= get_mm_parts(param, cond_func, field, Item_func::EQ_FUNC,
func->arguments()[1], cmp_type);
if (tree)
{
Item **arg, **end;
for (arg= func->arguments()+2, end= arg+func->argument_count()-2;
arg < end ; arg++)
{
tree= tree_or(param, tree, get_mm_parts(param, cond_func, field,
Item_func::EQ_FUNC,
*arg, cmp_type));
}
}
}
break;
}
default:
{
/*
Here the function for the following predicates are processed:
<, <=, =, >=, >, LIKE, IS NULL, IS NOT NULL.
If the predicate is of the form (value op field) it is handled
as the equivalent predicate (field rev_op value), e.g.
2 <= a is handled as a >= 2.
*/
Item_func::Functype func_type=
(value != cond_func->arguments()[0]) ? cond_func->functype() :
((Item_bool_func2*) cond_func)->rev_functype();
tree= get_mm_parts(param, cond_func, field, func_type, value, cmp_type);
}
}
DBUG_RETURN(tree);
}
/*
Build conjunction of all SEL_TREEs for a simple predicate applying equalities
SYNOPSIS
get_full_func_mm_tree()
param PARAM from SQL_SELECT::test_quick_select
cond_func item for the predicate
field_item field in the predicate
value constant in the predicate (or a field already read from
a table in the case of dynamic range access)
(for BETWEEN it contains the number of the field argument,
for IN it's always 0)
inv TRUE <> NOT cond_func is considered
(makes sense only when cond_func is BETWEEN or IN)
DESCRIPTION
For a simple SARGable predicate of the form (f op c), where f is a field and
c is a constant, the function builds a conjunction of all SEL_TREES that can
be obtained by the substitution of f for all different fields equal to f.
NOTES
If the WHERE condition contains a predicate (fi op c),
then not only SELL_TREE for this predicate is built, but
the trees for the results of substitution of fi for
each fj belonging to the same multiple equality as fi
are built as well.
E.g. for WHERE t1.a=t2.a AND t2.a > 10
a SEL_TREE for t2.a > 10 will be built for quick select from t2
and
a SEL_TREE for t1.a > 10 will be built for quick select from t1.
A BETWEEN predicate of the form (fi [NOT] BETWEEN c1 AND c2) is treated
in a similar way: we build a conjuction of trees for the results
of all substitutions of fi for equal fj.
Yet a predicate of the form (c BETWEEN f1i AND f2i) is processed
differently. It is considered as a conjuction of two SARGable
predicates (f1i <= c) and (f2i <=c) and the function get_full_func_mm_tree
is called for each of them separately producing trees for
AND j (f1j <=c ) and AND j (f2j <= c)
After this these two trees are united in one conjunctive tree.
It's easy to see that the same tree is obtained for
AND j,k (f1j <=c AND f2k<=c)
which is equivalent to
AND j,k (c BETWEEN f1j AND f2k).
The validity of the processing of the predicate (c NOT BETWEEN f1i AND f2i)
which equivalent to (f1i > c OR f2i < c) is not so obvious. Here the
function get_full_func_mm_tree is called for (f1i > c) and (f2i < c)
producing trees for AND j (f1j > c) and AND j (f2j < c). Then this two
trees are united in one OR-tree. The expression
(AND j (f1j > c) OR AND j (f2j < c)
is equivalent to the expression
AND j,k (f1j > c OR f2k < c)
which is just a translation of
AND j,k (c NOT BETWEEN f1j AND f2k)
In the cases when one of the items f1, f2 is a constant c1 we do not create
a tree for it at all. It works for BETWEEN predicates but does not
work for NOT BETWEEN predicates as we have to evaluate the expression
with it. If it is TRUE then the other tree can be completely ignored.
We do not do it now and no trees are built in these cases for
NOT BETWEEN predicates.
As to IN predicates only ones of the form (f IN (c1,...,cn)),
where f1 is a field and c1,...,cn are constant, are considered as
SARGable. We never try to narrow the index scan using predicates of
the form (c IN (c1,...,f,...,cn)).
RETURN
Pointer to the tree representing the built conjunction of SEL_TREEs
*/
static SEL_TREE *get_full_func_mm_tree(RANGE_OPT_PARAM *param,
Item_func *cond_func,
Item_field *field_item, Item *value,
bool inv)
{
SEL_TREE *tree= 0;
SEL_TREE *ftree= 0;
table_map ref_tables= 0;
table_map param_comp= ~(param->prev_tables | param->read_tables |
param->current_table);
DBUG_ENTER("get_full_func_mm_tree");
#ifdef HAVE_SPATIAL
if (field_item->field->type() == MYSQL_TYPE_GEOMETRY)
{
/* We have to be able to store all sorts of spatial features here */
((Field_geom*) field_item->field)->geom_type= Field::GEOM_GEOMETRY;
}
#endif /*HAVE_SPATIAL*/
for (uint i= 0; i < cond_func->arg_count; i++)
{
Item *arg= cond_func->arguments()[i]->real_item();
if (arg != field_item)
ref_tables|= arg->used_tables();
}
Field *field= field_item->field;
Item_result cmp_type= field->cmp_type();
if (!((ref_tables | field->table->map) & param_comp))
ftree= get_func_mm_tree(param, cond_func, field, value, cmp_type, inv);
Item_equal *item_equal= field_item->item_equal;
if (item_equal)
{
Item_equal_fields_iterator it(*item_equal);
while (it++)
{
Field *f= it.get_curr_field();
if (field->eq(f))
continue;
if (!((ref_tables | f->table->map) & param_comp))
{
tree= get_func_mm_tree(param, cond_func, f, value, cmp_type, inv);
ftree= !ftree ? tree : tree_and(param, ftree, tree);
}
}
}
DBUG_RETURN(ftree);
}
/* make a select tree of all keys in condition */
static SEL_TREE *get_mm_tree(RANGE_OPT_PARAM *param,COND *cond)
{
SEL_TREE *tree=0;
SEL_TREE *ftree= 0;
Item_field *field_item= 0;
bool inv= FALSE;
Item *value= 0;
DBUG_ENTER("get_mm_tree");
if (cond->type() == Item::COND_ITEM)
{
List_iterator<Item> li(*((Item_cond*) cond)->argument_list());
if (((Item_cond*) cond)->functype() == Item_func::COND_AND_FUNC)
{
tree= NULL;
Item *item;
while ((item=li++))
{
SEL_TREE *new_tree= get_mm_tree(param,item);
if (param->statement_should_be_aborted())
DBUG_RETURN(NULL);
tree= tree_and(param,tree,new_tree);
if (tree && tree->type == SEL_TREE::IMPOSSIBLE)
break;
}
}
else
{ // COND OR
tree= get_mm_tree(param,li++);
if (param->statement_should_be_aborted())
DBUG_RETURN(NULL);
if (tree)
{
Item *item;
while ((item=li++))
{
SEL_TREE *new_tree=get_mm_tree(param,item);
if (new_tree == NULL || param->statement_should_be_aborted())
DBUG_RETURN(NULL);
tree= tree_or(param,tree,new_tree);
if (tree == NULL || tree->type == SEL_TREE::ALWAYS)
break;
}
}
}
DBUG_RETURN(tree);
}
/* Here when simple cond */
if (cond->const_item())
{
if (cond->is_expensive())
DBUG_RETURN(0);
/*
During the cond->val_int() evaluation we can come across a subselect
item which may allocate memory on the thd->mem_root and assumes
all the memory allocated has the same life span as the subselect
item itself. So we have to restore the thread's mem_root here.
*/
MEM_ROOT *tmp_root= param->mem_root;
param->thd->mem_root= param->old_root;
tree= cond->val_int() ? new(tmp_root) SEL_TREE(SEL_TREE::ALWAYS) :
new(tmp_root) SEL_TREE(SEL_TREE::IMPOSSIBLE);
param->thd->mem_root= tmp_root;
DBUG_RETURN(tree);
}
table_map ref_tables= 0;
table_map param_comp= ~(param->prev_tables | param->read_tables |
param->current_table);
if (cond->type() != Item::FUNC_ITEM)
{ // Should be a field
ref_tables= cond->used_tables();
if ((ref_tables & param->current_table) ||
(ref_tables & ~(param->prev_tables | param->read_tables)))
DBUG_RETURN(0);
DBUG_RETURN(new SEL_TREE(SEL_TREE::MAYBE));
}
Item_func *cond_func= (Item_func*) cond;
if (cond_func->functype() == Item_func::BETWEEN ||
cond_func->functype() == Item_func::IN_FUNC)
inv= ((Item_func_opt_neg *) cond_func)->negated;
else if (cond_func->select_optimize() == Item_func::OPTIMIZE_NONE)
DBUG_RETURN(0);
param->cond= cond;
switch (cond_func->functype()) {
case Item_func::BETWEEN:
if (cond_func->arguments()[0]->real_item()->type() == Item::FIELD_ITEM)
{
field_item= (Item_field*) (cond_func->arguments()[0]->real_item());
ftree= get_full_func_mm_tree(param, cond_func, field_item, NULL, inv);
}
/*
Concerning the code below see the NOTES section in
the comments for the function get_full_func_mm_tree()
*/
for (uint i= 1 ; i < cond_func->arg_count ; i++)
{
if (cond_func->arguments()[i]->real_item()->type() == Item::FIELD_ITEM)
{
field_item= (Item_field*) (cond_func->arguments()[i]->real_item());
SEL_TREE *tmp= get_full_func_mm_tree(param, cond_func,
field_item, (Item*)(intptr)i, inv);
if (inv)
{
tree= !tree ? tmp : tree_or(param, tree, tmp);
if (tree == NULL)
break;
}
else
tree= tree_and(param, tree, tmp);
}
else if (inv)
{
tree= 0;
break;
}
}
ftree = tree_and(param, ftree, tree);
break;
case Item_func::IN_FUNC:
{
Item_func_in *func=(Item_func_in*) cond_func;
if (func->key_item()->real_item()->type() != Item::FIELD_ITEM)
DBUG_RETURN(0);
field_item= (Item_field*) (func->key_item()->real_item());
ftree= get_full_func_mm_tree(param, cond_func, field_item, NULL, inv);
break;
}
case Item_func::MULT_EQUAL_FUNC:
{
Item_equal *item_equal= (Item_equal *) cond;
if (!(value= item_equal->get_const()) || value->is_expensive())
DBUG_RETURN(0);
Item_equal_fields_iterator it(*item_equal);
ref_tables= value->used_tables();
while (it++)
{
Field *field= it.get_curr_field();
Item_result cmp_type= field->cmp_type();
if (!((ref_tables | field->table->map) & param_comp))
{
tree= get_mm_parts(param, cond, field, Item_func::EQ_FUNC,
value,cmp_type);
ftree= !ftree ? tree : tree_and(param, ftree, tree);
}
}
DBUG_RETURN(ftree);
}
default:
DBUG_ASSERT (!ftree);
if (cond_func->arguments()[0]->real_item()->type() == Item::FIELD_ITEM)
{
field_item= (Item_field*) (cond_func->arguments()[0]->real_item());
value= cond_func->arg_count > 1 ? cond_func->arguments()[1] : NULL;
if (value && value->is_expensive())
DBUG_RETURN(0);
if (!cond_func->arguments()[0]->real_item()->const_item())
ftree= get_full_func_mm_tree(param, cond_func, field_item, value, inv);
}
/*
Even if get_full_func_mm_tree() was executed above and did not
return a range predicate it may still be possible to create one
by reversing the order of the operands. Note that this only
applies to predicates where both operands are fields. Example: A
query of the form
WHERE t1.a OP t2.b
In this case, arguments()[0] == t1.a and arguments()[1] == t2.b.
When creating range predicates for t2, get_full_func_mm_tree()
above will return NULL because 'field' belongs to t1 and only
predicates that applies to t2 are of interest. In this case a
call to get_full_func_mm_tree() with reversed operands (see
below) may succeed.
*/
if (!ftree && cond_func->have_rev_func() &&
cond_func->arguments()[1]->real_item()->type() == Item::FIELD_ITEM)
{
field_item= (Item_field*) (cond_func->arguments()[1]->real_item());
value= cond_func->arguments()[0];
if (value && value->is_expensive())
DBUG_RETURN(0);
if (!cond_func->arguments()[1]->real_item()->const_item())
ftree= get_full_func_mm_tree(param, cond_func, field_item, value, inv);
}
}
DBUG_RETURN(ftree);
}
static SEL_TREE *
get_mm_parts(RANGE_OPT_PARAM *param, COND *cond_func, Field *field,
Item_func::Functype type,
Item *value, Item_result cmp_type)
{
DBUG_ENTER("get_mm_parts");
if (field->table != param->table)
DBUG_RETURN(0);
KEY_PART *key_part = param->key_parts;
KEY_PART *end = param->key_parts_end;
SEL_TREE *tree=0;
if (value &&
value->used_tables() & ~(param->prev_tables | param->read_tables))
DBUG_RETURN(0);
for (; key_part != end ; key_part++)
{
if (field->eq(key_part->field))
{
SEL_ARG *sel_arg=0;
if (!tree && !(tree=new SEL_TREE()))
DBUG_RETURN(0); // OOM
if (!value || !(value->used_tables() & ~param->read_tables))
{
sel_arg=get_mm_leaf(param,cond_func,
key_part->field,key_part,type,value);
if (!sel_arg)
continue;
if (sel_arg->type == SEL_ARG::IMPOSSIBLE)
{
tree->type=SEL_TREE::IMPOSSIBLE;
DBUG_RETURN(tree);
}
}
else
{
// This key may be used later
if (!(sel_arg= new SEL_ARG(SEL_ARG::MAYBE_KEY)))
DBUG_RETURN(0); // OOM
}
sel_arg->part=(uchar) key_part->part;
sel_arg->max_part_no= sel_arg->part+1;
tree->keys[key_part->key]=sel_add(tree->keys[key_part->key],sel_arg);
tree->keys_map.set_bit(key_part->key);
}
}
if (tree && tree->merges.is_empty() && tree->keys_map.is_clear_all())
tree= NULL;
DBUG_RETURN(tree);
}
static SEL_ARG *
get_mm_leaf(RANGE_OPT_PARAM *param, COND *conf_func, Field *field,
KEY_PART *key_part, Item_func::Functype type,Item *value)
{
uint maybe_null=(uint) field->real_maybe_null();
bool optimize_range;
SEL_ARG *tree= 0;
MEM_ROOT *alloc= param->mem_root;
uchar *str;
int err;
DBUG_ENTER("get_mm_leaf");
/*
We need to restore the runtime mem_root of the thread in this
function because it evaluates the value of its argument, while
the argument can be any, e.g. a subselect. The subselect
items, in turn, assume that all the memory allocated during
the evaluation has the same life span as the item itself.
TODO: opt_range.cc should not reset thd->mem_root at all.
*/
param->thd->mem_root= param->old_root;
if (!value) // IS NULL or IS NOT NULL
{
if (field->table->maybe_null) // Can't use a key on this
goto end;
if (!maybe_null) // Not null field
{
if (type == Item_func::ISNULL_FUNC)
tree= &null_element;
goto end;
}
if (!(tree= new (alloc) SEL_ARG(field,is_null_string,is_null_string)))
goto end; // out of memory
if (type == Item_func::ISNOTNULL_FUNC)
{
tree->min_flag=NEAR_MIN; /* IS NOT NULL -> X > NULL */
tree->max_flag=NO_MAX_RANGE;
}
goto end;
}
/*
1. Usually we can't use an index if the column collation
differ from the operation collation.
2. However, we can reuse a case insensitive index for
the binary searches:
WHERE latin1_swedish_ci_column = 'a' COLLATE lati1_bin;
WHERE latin1_swedish_ci_colimn = BINARY 'a '
*/
if (field->result_type() == STRING_RESULT &&
field->match_collation_to_optimize_range() &&
value->result_type() == STRING_RESULT &&
key_part->image_type == Field::itRAW &&
field->charset() != conf_func->compare_collation() &&
!(conf_func->compare_collation()->state & MY_CS_BINSORT &&
(type == Item_func::EQUAL_FUNC || type == Item_func::EQ_FUNC)))
goto end;
if (value->cmp_type() == TIME_RESULT && field->cmp_type() != TIME_RESULT)
goto end;
if (key_part->image_type == Field::itMBR)
{
// @todo: use is_spatial_operator() instead?
switch (type) {
case Item_func::SP_EQUALS_FUNC:
case Item_func::SP_DISJOINT_FUNC:
case Item_func::SP_INTERSECTS_FUNC:
case Item_func::SP_TOUCHES_FUNC:
case Item_func::SP_CROSSES_FUNC:
case Item_func::SP_WITHIN_FUNC:
case Item_func::SP_CONTAINS_FUNC:
case Item_func::SP_OVERLAPS_FUNC:
break;
default:
/*
We cannot involve spatial indexes for queries that
don't use MBREQUALS(), MBRDISJOINT(), etc. functions.
*/
goto end;
}
}
if (param->using_real_indexes)
optimize_range= field->optimize_range(param->real_keynr[key_part->key],
key_part->part);
else
optimize_range= TRUE;
if (type == Item_func::LIKE_FUNC)
{
bool like_error;
char buff1[MAX_FIELD_WIDTH];
uchar *min_str,*max_str;
String tmp(buff1,sizeof(buff1),value->collation.collation),*res;
size_t length, offset, min_length, max_length;
uint field_length= field->pack_length()+maybe_null;
if (!optimize_range)
goto end;
if (!(res= value->val_str(&tmp)))
{
tree= &null_element;
goto end;
}
/*
TODO:
Check if this was a function. This should have be optimized away
in the sql_select.cc
*/
if (res != &tmp)
{
tmp.copy(*res); // Get own copy
res= &tmp;
}
if (field->cmp_type() != STRING_RESULT)
goto end; // Can only optimize strings
offset=maybe_null;
length=key_part->store_length;
if (length != key_part->length + maybe_null)
{
/* key packed with length prefix */
offset+= HA_KEY_BLOB_LENGTH;
field_length= length - HA_KEY_BLOB_LENGTH;
}
else
{
if (unlikely(length < field_length))
{
/*
This can only happen in a table created with UNIREG where one key
overlaps many fields
*/
length= field_length;
}
else
field_length= length;
}
length+=offset;
if (!(min_str= (uchar*) alloc_root(alloc, length*2)))
goto end;
max_str=min_str+length;
if (maybe_null)
max_str[0]= min_str[0]=0;
field_length-= maybe_null;
like_error= my_like_range(field->charset(),
res->ptr(), res->length(),
((Item_func_like*)(param->cond))->escape,
wild_one, wild_many,
field_length,
(char*) min_str+offset, (char*) max_str+offset,
&min_length, &max_length);
if (like_error) // Can't optimize with LIKE
goto end;
if (offset != maybe_null) // BLOB or VARCHAR
{
int2store(min_str+maybe_null,min_length);
int2store(max_str+maybe_null,max_length);
}
tree= new (alloc) SEL_ARG(field, min_str, max_str);
goto end;
}
if (!optimize_range &&
type != Item_func::EQ_FUNC &&
type != Item_func::EQUAL_FUNC)
goto end; // Can't optimize this
/*
We can't always use indexes when comparing a string index to a number
cmp_type() is checked to allow compare of dates to numbers
*/
if (field->cmp_type() == STRING_RESULT && value->cmp_type() != STRING_RESULT)
goto end;
err= value->save_in_field_no_warnings(field, 1);
if (err == 2 && field->cmp_type() == STRING_RESULT)
{
if (type == Item_func::EQ_FUNC)
{
tree= new (alloc) SEL_ARG(field, 0, 0);
tree->type= SEL_ARG::IMPOSSIBLE;
}
else
tree= NULL; /* Cannot infer anything */
goto end;
}
if (err > 0)
{
if (field->cmp_type() != value->result_type())
{
if ((type == Item_func::EQ_FUNC || type == Item_func::EQUAL_FUNC) &&
value->result_type() == item_cmp_type(field->result_type(),
value->result_type()))
{
tree= new (alloc) SEL_ARG(field, 0, 0);
tree->type= SEL_ARG::IMPOSSIBLE;
goto end;
}
else
{
/*
TODO: We should return trees of the type SEL_ARG::IMPOSSIBLE
for the cases like int_field > 999999999999999999999999 as well.
*/
tree= 0;
if (err == 3 && field->type() == FIELD_TYPE_DATE &&
(type == Item_func::GT_FUNC || type == Item_func::GE_FUNC ||
type == Item_func::LT_FUNC || type == Item_func::LE_FUNC) )
{
/*
We were saving DATETIME into a DATE column, the conversion went ok
but a non-zero time part was cut off.
In MySQL's SQL dialect, DATE and DATETIME are compared as datetime
values. Index over a DATE column uses DATE comparison. Changing
from one comparison to the other is possible:
datetime(date_col)< '2007-12-10 12:34:55' -> date_col<='2007-12-10'
datetime(date_col)<='2007-12-10 12:34:55' -> date_col<='2007-12-10'
datetime(date_col)> '2007-12-10 12:34:55' -> date_col>='2007-12-10'
datetime(date_col)>='2007-12-10 12:34:55' -> date_col>='2007-12-10'
but we'll need to convert '>' to '>=' and '<' to '<='. This will
be done together with other types at the end of this function
(grep for stored_field_cmp_to_item)
*/
}
else
goto end;
}
}
/*
guaranteed at this point: err > 0; field and const of same type
If an integer got bounded (e.g. to within 0..255 / -128..127)
for < or >, set flags as for <= or >= (no NEAR_MAX / NEAR_MIN)
*/
else if (err == 1 && field->result_type() == INT_RESULT)
{
if (type == Item_func::LT_FUNC && (value->val_int() > 0))
type = Item_func::LE_FUNC;
else if (type == Item_func::GT_FUNC &&
(field->type() != FIELD_TYPE_BIT) &&
!((Field_num*)field)->unsigned_flag &&
!((Item_int*)value)->unsigned_flag &&
(value->val_int() < 0))
type = Item_func::GE_FUNC;
}
}
else if (err < 0)
{
/* This happens when we try to insert a NULL field in a not null column */
tree= &null_element; // cmp with NULL is never TRUE
goto end;
}
/*
Any sargable predicate except "<=>" involving NULL as a constant is always
FALSE
*/
if (type != Item_func::EQUAL_FUNC && field->is_real_null())
{
tree= &null_element;
goto end;
}
str= (uchar*) alloc_root(alloc, key_part->store_length+1);
if (!str)
goto end;
if (maybe_null)
*str= (uchar) field->is_real_null(); // Set to 1 if null
field->get_key_image(str+maybe_null, key_part->length,
key_part->image_type);
if (!(tree= new (alloc) SEL_ARG(field, str, str)))
goto end; // out of memory
/*
Check if we are comparing an UNSIGNED integer with a negative constant.
In this case we know that:
(a) (unsigned_int [< | <=] negative_constant) == FALSE
(b) (unsigned_int [> | >=] negative_constant) == TRUE
In case (a) the condition is false for all values, and in case (b) it
is true for all values, so we can avoid unnecessary retrieval and condition
testing, and we also get correct comparison of unsinged integers with
negative integers (which otherwise fails because at query execution time
negative integers are cast to unsigned if compared with unsigned).
*/
if (field->result_type() == INT_RESULT &&
value->result_type() == INT_RESULT &&
((field->type() == FIELD_TYPE_BIT ||
((Field_num *) field)->unsigned_flag) &&
!((Item_int*) value)->unsigned_flag))
{
longlong item_val= value->val_int();
if (item_val < 0)
{
if (type == Item_func::LT_FUNC || type == Item_func::LE_FUNC)
{
tree->type= SEL_ARG::IMPOSSIBLE;
goto end;
}
if (type == Item_func::GT_FUNC || type == Item_func::GE_FUNC)
{
tree= 0;
goto end;
}
}
}
switch (type) {
case Item_func::LT_FUNC:
if (stored_field_cmp_to_item(param->thd, field, value) == 0)
tree->max_flag=NEAR_MAX;
/* fall through */
case Item_func::LE_FUNC:
if (!maybe_null)
tree->min_flag=NO_MIN_RANGE; /* From start */
else
{ // > NULL
tree->min_value=is_null_string;
tree->min_flag=NEAR_MIN;
}
break;
case Item_func::GT_FUNC:
/* Don't use open ranges for partial key_segments */
if ((!(key_part->flag & HA_PART_KEY_SEG)) &&
(stored_field_cmp_to_item(param->thd, field, value) <= 0))
tree->min_flag=NEAR_MIN;
tree->max_flag= NO_MAX_RANGE;
break;
case Item_func::GE_FUNC:
/* Don't use open ranges for partial key_segments */
if ((!(key_part->flag & HA_PART_KEY_SEG)) &&
(stored_field_cmp_to_item(param->thd, field, value) < 0))
tree->min_flag= NEAR_MIN;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_EQUALS_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_EQUAL;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_DISJOINT_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_DISJOINT;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_INTERSECTS_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_TOUCHES_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_CROSSES_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_WITHIN_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_WITHIN;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_CONTAINS_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_CONTAIN;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
case Item_func::SP_OVERLAPS_FUNC:
tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
tree->max_flag=NO_MAX_RANGE;
break;
default:
break;
}
end:
param->thd->mem_root= alloc;
DBUG_RETURN(tree);
}
/******************************************************************************
** Tree manipulation functions
** If tree is 0 it means that the condition can't be tested. It refers
** to a non existent table or to a field in current table with isn't a key.
** The different tree flags:
** IMPOSSIBLE: Condition is never TRUE
** ALWAYS: Condition is always TRUE
** MAYBE: Condition may exists when tables are read
** MAYBE_KEY: Condition refers to a key that may be used in join loop
** KEY_RANGE: Condition uses a key
******************************************************************************/
/*
Add a new key test to a key when scanning through all keys
This will never be called for same key parts.
*/
static SEL_ARG *
sel_add(SEL_ARG *key1,SEL_ARG *key2)
{
SEL_ARG *root,**key_link;
if (!key1)
return key2;
if (!key2)
return key1;
key_link= &root;
while (key1 && key2)
{
if (key1->part < key2->part)
{
*key_link= key1;
key_link= &key1->next_key_part;
key1=key1->next_key_part;
}
else
{
*key_link= key2;
key_link= &key2->next_key_part;
key2=key2->next_key_part;
}
}
*key_link=key1 ? key1 : key2;
return root;
}
/*
Build a range tree for the conjunction of the range parts of two trees
SYNOPSIS
and_range_trees()
param Context info for the operation
tree1 SEL_TREE for the first conjunct
tree2 SEL_TREE for the second conjunct
result SEL_TREE for the result
DESCRIPTION
This function takes range parts of two trees tree1 and tree2 and builds
a range tree for the conjunction of the formulas that these two range parts
represent.
More exactly:
if the range part of tree1 represents the normalized formula
R1_1 AND ... AND R1_k,
and the range part of tree2 represents the normalized formula
R2_1 AND ... AND R2_k,
then the range part of the result represents the formula:
RT = R_1 AND ... AND R_k, where R_i=(R1_i AND R2_i) for each i from [1..k]
The function assumes that tree1 is never equal to tree2. At the same
time the tree result can be the same as tree1 (but never as tree2).
If result==tree1 then rt replaces the range part of tree1 leaving
imerges as they are.
if result!=tree1 than it is assumed that the SEL_ARG trees in tree1 and
tree2 should be preserved. Otherwise they can be destroyed.
RETURN
1 if the type the result tree is SEL_TREE::IMPOSSIBLE
0 otherwise
*/
static
int and_range_trees(RANGE_OPT_PARAM *param, SEL_TREE *tree1, SEL_TREE *tree2,
SEL_TREE *result)
{
DBUG_ENTER("and_ranges");
key_map result_keys;
result_keys.clear_all();
key_map anded_keys= tree1->keys_map;
anded_keys.merge(tree2->keys_map);
int key_no;
key_map::Iterator it(anded_keys);
while ((key_no= it++) != key_map::Iterator::BITMAP_END)
{
uint flag=0;
SEL_ARG *key1= tree1->keys[key_no];
SEL_ARG *key2= tree2->keys[key_no];
if (key1 && !key1->simple_key())
flag|= CLONE_KEY1_MAYBE;
if (key2 && !key2->simple_key())
flag|=CLONE_KEY2_MAYBE;
if (result != tree1)
{
if (key1)
key1->incr_refs();
if (key2)
key2->incr_refs();
}
SEL_ARG *key;
if ((result->keys[key_no]= key =key_and(param, key1, key2, flag)))
{
if (key && key->type == SEL_ARG::IMPOSSIBLE)
{
result->type= SEL_TREE::IMPOSSIBLE;
DBUG_RETURN(1);
}
result_keys.set_bit(key_no);
#ifdef EXTRA_DEBUG
if (param->alloced_sel_args < SEL_ARG::MAX_SEL_ARGS)
key->test_use_count(key);
#endif
}
}
result->keys_map= result_keys;
DBUG_RETURN(0);
}
/*
Build a SEL_TREE for a conjunction out of such trees for the conjuncts
SYNOPSIS
tree_and()
param Context info for the operation
tree1 SEL_TREE for the first conjunct
tree2 SEL_TREE for the second conjunct
DESCRIPTION
This function builds a tree for the formula (A AND B) out of the trees
tree1 and tree2 that has been built for the formulas A and B respectively.
In a general case
tree1 represents the formula RT1 AND MT1,
where RT1 = R1_1 AND ... AND R1_k1, MT1=M1_1 AND ... AND M1_l1;
tree2 represents the formula RT2 AND MT2
where RT2 = R2_1 AND ... AND R2_k2, MT2=M2_1 AND ... AND M2_l2.
The result tree will represent the formula of the the following structure:
RT AND RT1MT2 AND RT2MT1, such that
rt is a tree obtained by range intersection of trees tree1 and tree2,
RT1MT2 = RT1M2_1 AND ... AND RT1M2_l2,
RT2MT1 = RT2M1_1 AND ... AND RT2M1_l1,
where rt1m2_i (i=1,...,l2) is the result of the pushdown operation
of range tree rt1 into imerge m2_i, while rt2m1_j (j=1,...,l1) is the
result of the pushdown operation of range tree rt2 into imerge m1_j.
RT1MT2/RT2MT is empty if MT2/MT1 is empty.
The range intersection of two range trees is produced by the function
and_range_trees. The pushdown of a range tree to a imerge is performed
by the function imerge_list_and_tree. This function may produce imerges
containing only one range tree. Such trees are intersected with rt and
the result of intersection is returned as the range part of the result
tree, while the corresponding imerges are removed altogether from its
imerge part.
NOTE
The pushdown operation of range trees into imerges is needed to be able
to construct valid imerges for the condition like this:
key1_p1=c1 AND (key1_p2 BETWEEN c21 AND c22 OR key2 < c2)
NOTE
Currently we do not support intersection between indexes and index merges.
When this will be supported the list of imerges for the result tree
should include also imerges from M1 and M2. That's why an extra parameter
is added to the function imerge_list_and_tree. If we call the function
with the last parameter equal to FALSE then MT1 and MT2 will be preserved
in the imerge list of the result tree. This can lead to the exponential
growth of the imerge list though.
Currently the last parameter of imerge_list_and_tree calls is always
TRUE.
RETURN
The result tree, if a success
0 - otherwise.
*/
static
SEL_TREE *tree_and(RANGE_OPT_PARAM *param, SEL_TREE *tree1, SEL_TREE *tree2)
{
DBUG_ENTER("tree_and");
if (!tree1)
DBUG_RETURN(tree2);
if (!tree2)
DBUG_RETURN(tree1);
if (tree1->type == SEL_TREE::IMPOSSIBLE || tree2->type == SEL_TREE::ALWAYS)
DBUG_RETURN(tree1);
if (tree2->type == SEL_TREE::IMPOSSIBLE || tree1->type == SEL_TREE::ALWAYS)
DBUG_RETURN(tree2);
if (tree1->type == SEL_TREE::MAYBE)
{
if (tree2->type == SEL_TREE::KEY)
tree2->type=SEL_TREE::KEY_SMALLER;
DBUG_RETURN(tree2);
}
if (tree2->type == SEL_TREE::MAYBE)
{
tree1->type=SEL_TREE::KEY_SMALLER;
DBUG_RETURN(tree1);
}
if (!tree1->merges.is_empty())
imerge_list_and_tree(param, &tree1->merges, tree2, TRUE);
if (!tree2->merges.is_empty())
imerge_list_and_tree(param, &tree2->merges, tree1, TRUE);
if (and_range_trees(param, tree1, tree2, tree1))
DBUG_RETURN(tree1);
imerge_list_and_list(&tree1->merges, &tree2->merges);
eliminate_single_tree_imerges(param, tree1);
DBUG_RETURN(tree1);
}
/*
Eliminate single tree imerges in a SEL_TREE objects
SYNOPSIS
eliminate_single_tree_imerges()
param Context info for the function
tree SEL_TREE where single tree imerges are to be eliminated
DESCRIPTION
For each imerge in 'tree' that contains only one disjunct tree, i.e.
for any imerge of the form m=rt, the function performs and operation
the range part of tree, replaces rt the with the result of anding and
removes imerge m from the the merge part of 'tree'.
RETURN VALUE
none
*/
static
void eliminate_single_tree_imerges(RANGE_OPT_PARAM *param, SEL_TREE *tree)
{
SEL_IMERGE *imerge;
List<SEL_IMERGE> merges= tree->merges;
List_iterator<SEL_IMERGE> it(merges);
tree->merges.empty();
while ((imerge= it++))
{
if (imerge->trees+1 == imerge->trees_next)
{
tree= tree_and(param, tree, *imerge->trees);
it.remove();
}
}
tree->merges= merges;
}
/*
For two trees check that there are indexes with ranges in both of them
SYNOPSIS
sel_trees_have_common_keys()
tree1 SEL_TREE for the first tree
tree2 SEL_TREE for the second tree
common_keys OUT bitmap of all indexes with ranges in both trees
DESCRIPTION
For two trees tree1 and tree1 the function checks if there are indexes
in their range parts such that SEL_ARG trees are defined for them in the
range parts of both trees. The function returns the bitmap of such
indexes in the parameter common_keys.
RETURN
TRUE if there are such indexes (common_keys is nor empty)
FALSE otherwise
*/
static
bool sel_trees_have_common_keys(SEL_TREE *tree1, SEL_TREE *tree2,
key_map *common_keys)
{
*common_keys= tree1->keys_map;
common_keys->intersect(tree2->keys_map);
return !common_keys->is_clear_all();
}
/*
Check whether range parts of two trees can be ored for some indexes
SYNOPSIS
sel_trees_can_be_ored()
param Context info for the function
tree1 SEL_TREE for the first tree
tree2 SEL_TREE for the second tree
common_keys IN/OUT IN: bitmap of all indexes with SEL_ARG in both trees
OUT: bitmap of all indexes that can be ored
DESCRIPTION
For two trees tree1 and tree2 and the bitmap common_keys containing
bits for indexes that have SEL_ARG trees in range parts of both trees
the function checks if there are indexes for which SEL_ARG trees can
be ored. Two SEL_ARG trees for the same index can be ored if the most
major components of the index used in these trees coincide. If the
SEL_ARG trees for an index cannot be ored the function clears the bit
for this index in the bitmap common_keys.
The function does not verify that indexes marked in common_keys really
have SEL_ARG trees in both tree1 and tree2. It assumes that this is true.
NOTE
The function sel_trees_can_be_ored is usually used in pair with the
function sel_trees_have_common_keys.
RETURN
TRUE if there are indexes for which SEL_ARG trees can be ored
FALSE otherwise
*/
static
bool sel_trees_can_be_ored(RANGE_OPT_PARAM* param,
SEL_TREE *tree1, SEL_TREE *tree2,
key_map *common_keys)
{
DBUG_ENTER("sel_trees_can_be_ored");
if (!sel_trees_have_common_keys(tree1, tree2, common_keys))
DBUG_RETURN(FALSE);
int key_no;
key_map::Iterator it(*common_keys);
while ((key_no= it++) != key_map::Iterator::BITMAP_END)
{
DBUG_ASSERT(tree1->keys[key_no] && tree2->keys[key_no]);
/* Trees have a common key, check if they refer to the same key part */
if (tree1->keys[key_no]->part != tree2->keys[key_no]->part)
common_keys->clear_bit(key_no);
}
DBUG_RETURN(!common_keys->is_clear_all());
}
/*
Check whether range parts of two trees must be ored for some indexes
SYNOPSIS
sel_trees_must_be_ored()
param Context info for the function
tree1 SEL_TREE for the first tree
tree2 SEL_TREE for the second tree
ordable_keys bitmap of SEL_ARG trees that can be ored
DESCRIPTION
For two trees tree1 and tree2 the function checks whether they must be
ored. The function assumes that the bitmap ordable_keys contains bits for
those corresponding pairs of SEL_ARG trees from tree1 and tree2 that can
be ored.
We believe that tree1 and tree2 must be ored if any pair of SEL_ARG trees
r1 and r2, such that r1 is from tree1 and r2 is from tree2 and both
of them are marked in ordable_keys, can be merged.
NOTE
The function sel_trees_must_be_ored as a rule is used in pair with the
function sel_trees_can_be_ored.
RETURN
TRUE if there are indexes for which SEL_ARG trees must be ored
FALSE otherwise
*/
static
bool sel_trees_must_be_ored(RANGE_OPT_PARAM* param,
SEL_TREE *tree1, SEL_TREE *tree2,
key_map oredable_keys)
{
key_map tmp;
DBUG_ENTER("sel_trees_must_be_ored");
tmp= tree1->keys_map;
tmp.merge(tree2->keys_map);
tmp.subtract(oredable_keys);
if (!tmp.is_clear_all())
DBUG_RETURN(FALSE);
int idx1, idx2;
key_map::Iterator it1(oredable_keys);
while ((idx1= it1++) != key_map::Iterator::BITMAP_END)
{
KEY_PART *key1_init= param->key[idx1]+tree1->keys[idx1]->part;
KEY_PART *key1_end= param->key[idx1]+tree1->keys[idx1]->max_part_no;
key_map::Iterator it2(oredable_keys);
while ((idx2= it2++) != key_map::Iterator::BITMAP_END)
{
if (idx2 <= idx1)
continue;
KEY_PART *key2_init= param->key[idx2]+tree2->keys[idx2]->part;
KEY_PART *key2_end= param->key[idx2]+tree2->keys[idx2]->max_part_no;
KEY_PART *part1, *part2;
for (part1= key1_init, part2= key2_init;
part1 < key1_end && part2 < key2_end;
part1++, part2++)
{
if (!part1->field->eq(part2->field))
DBUG_RETURN(FALSE);
}
}
}
DBUG_RETURN(TRUE);
}
/*
Remove the trees that are not suitable for record retrieval
SYNOPSIS
remove_nonrange_trees()
param Context info for the function
tree Tree to be processed, tree->type is KEY or KEY_SMALLER
DESCRIPTION
This function walks through tree->keys[] and removes the SEL_ARG* trees
that are not "maybe" trees (*) and cannot be used to construct quick range
selects.
(*) - have type MAYBE or MAYBE_KEY. Perhaps we should remove trees of
these types here as well.
A SEL_ARG* tree cannot be used to construct quick select if it has
tree->part != 0. (e.g. it could represent "keypart2 < const").
Normally we allow construction of SEL_TREE objects that have SEL_ARG
trees that do not allow quick range select construction.
For example:
for " keypart1=1 AND keypart2=2 " the execution will proceed as follows:
tree1= SEL_TREE { SEL_ARG{keypart1=1} }
tree2= SEL_TREE { SEL_ARG{keypart2=2} } -- can't make quick range select
from this
call tree_and(tree1, tree2) -- this joins SEL_ARGs into a usable SEL_ARG
tree.
Another example:
tree3= SEL_TREE { SEL_ARG{key1part1 = 1} }
tree4= SEL_TREE { SEL_ARG{key2part2 = 2} } -- can't make quick range select
from this
call tree_or(tree3, tree4) -- creates a SEL_MERGE ot of which no index
merge can be constructed, but it is potentially useful, as anding it with
tree5= SEL_TREE { SEL_ARG{key2part1 = 3} } creates an index merge that
represents the formula
key1part1=1 AND key2part1=3 OR key2part1=3 AND key2part2=2
for which an index merge can be built.
Any final SEL_TREE may contain SEL_ARG trees for which no quick select
can be built. Such SEL_ARG trees should be removed from the range part
before different range scans are evaluated. Such SEL_ARG trees also should
be removed from all range trees of each index merge before different
possible index merge plans are evaluated. If after this removal one
of the range trees in the index merge becomes empty the whole index merge
must be discarded.
RETURN
0 Ok, some suitable trees left
1 No tree->keys[] left.
*/
static bool remove_nonrange_trees(RANGE_OPT_PARAM *param, SEL_TREE *tree)
{
bool res= FALSE;
for (uint i=0; i < param->keys; i++)
{
if (tree->keys[i])
{
if (tree->keys[i]->part)
{
tree->keys[i]= NULL;
tree->keys_map.clear_bit(i);
}
else
res= TRUE;
}
}
return !res;
}
/*
Build a SEL_TREE for a disjunction out of such trees for the disjuncts
SYNOPSIS
tree_or()
param Context info for the operation
tree1 SEL_TREE for the first disjunct
tree2 SEL_TREE for the second disjunct
DESCRIPTION
This function builds a tree for the formula (A OR B) out of the trees
tree1 and tree2 that has been built for the formulas A and B respectively.
In a general case
tree1 represents the formula RT1 AND MT1,
where RT1=R1_1 AND ... AND R1_k1, MT1=M1_1 AND ... AND M1_l1;
tree2 represents the formula RT2 AND MT2
where RT2=R2_1 AND ... AND R2_k2, MT2=M2_1 and ... and M2_l2.
The function constructs the result tree according the formula
(RT1 OR RT2) AND (MT1 OR RT1) AND (MT2 OR RT2) AND (MT1 OR MT2)
that is equivalent to the formula (RT1 AND MT1) OR (RT2 AND MT2).
To limit the number of produced imerges the function considers
a weaker formula than the original one:
(RT1 AND M1_1) OR (RT2 AND M2_1)
that is equivalent to:
(RT1 OR RT2) (1)
AND
(M1_1 OR M2_1) (2)
AND
(M1_1 OR RT2) (3)
AND
(M2_1 OR RT1) (4)
For the first conjunct (1) the function builds a tree with a range part
and, possibly, one imerge. For the other conjuncts (2-4)the function
produces sets of imerges. All constructed imerges are included into the
result tree.
For the formula (1) the function produces the tree representing a formula
of the structure RT [AND M], such that:
- the range tree rt contains the result of oring SEL_ARG trees from rt1
and rt2
- the imerge m consists of two range trees rt1 and rt2.
The imerge m is added if it's not true that rt1 and rt2 must be ored
If rt1 and rt2 can't be ored rt is empty and only m is produced for (1).
To produce imerges for the formula (2) the function calls the function
imerge_list_or_list passing it the merge parts of tree1 and tree2 as
parameters.
To produce imerges for the formula (3) the function calls the function
imerge_list_or_tree passing it the imerge m1_1 and the range tree rt2 as
parameters. Similarly, to produce imerges for the formula (4) the function
calls the function imerge_list_or_tree passing it the imerge m2_1 and the
range tree rt1.
If rt1 is empty then the trees for (1) and (4) are empty.
If rt2 is empty then the trees for (1) and (3) are empty.
If mt1 is empty then the trees for (2) and (3) are empty.
If mt2 is empty then the trees for (2) and (4) are empty.
RETURN
The result tree for the operation if a success
0 - otherwise
*/
static SEL_TREE *
tree_or(RANGE_OPT_PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
{
DBUG_ENTER("tree_or");
if (!tree1 || !tree2)
DBUG_RETURN(0);
if (tree1->type == SEL_TREE::IMPOSSIBLE || tree2->type == SEL_TREE::ALWAYS)
DBUG_RETURN(tree2);
if (tree2->type == SEL_TREE::IMPOSSIBLE || tree1->type == SEL_TREE::ALWAYS)
DBUG_RETURN(tree1);
if (tree1->type == SEL_TREE::MAYBE)
DBUG_RETURN(tree1); // Can't use this
if (tree2->type == SEL_TREE::MAYBE)
DBUG_RETURN(tree2);
SEL_TREE *result= NULL;
key_map result_keys;
key_map ored_keys;
SEL_TREE *rtree[2]= {NULL,NULL};
SEL_IMERGE *imerge[2]= {NULL, NULL};
bool no_ranges1= tree1->without_ranges();
bool no_ranges2= tree2->without_ranges();
bool no_merges1= tree1->without_imerges();
bool no_merges2= tree2->without_imerges();
if (!no_ranges1 && !no_merges2)
{
rtree[0]= new SEL_TREE(tree1, TRUE, param);
imerge[1]= new SEL_IMERGE(tree2->merges.head(), 0, param);
}
if (!no_ranges2 && !no_merges1)
{
rtree[1]= new SEL_TREE(tree2, TRUE, param);
imerge[0]= new SEL_IMERGE(tree1->merges.head(), 0, param);
}
bool no_imerge_from_ranges= FALSE;
if (!(result= new SEL_TREE()))
DBUG_RETURN(result);
/* Build the range part of the tree for the formula (1) */
if (sel_trees_can_be_ored(param, tree1, tree2, &ored_keys))
{
bool must_be_ored= sel_trees_must_be_ored(param, tree1, tree2, ored_keys);
no_imerge_from_ranges= must_be_ored;
key_map::Iterator it(ored_keys);
int key_no;
while ((key_no= it++) != key_map::Iterator::BITMAP_END)
{
SEL_ARG *key1= tree1->keys[key_no];
SEL_ARG *key2= tree2->keys[key_no];
if (!must_be_ored)
{
key1->incr_refs();
key2->incr_refs();
}
if ((result->keys[key_no]= key_or(param, key1, key2)))
result->keys_map.set_bit(key_no);
}
result->type= tree1->type;
}
if (no_imerge_from_ranges && no_merges1 && no_merges2)
{
if (result->keys_map.is_clear_all())
result->type= SEL_TREE::ALWAYS;
DBUG_RETURN(result);
}
SEL_IMERGE *imerge_from_ranges;
if (!(imerge_from_ranges= new SEL_IMERGE()))
result= NULL;
else if (!no_ranges1 && !no_ranges2 && !no_imerge_from_ranges)
{
/* Build the imerge part of the tree for the formula (1) */
SEL_TREE *rt1= tree1;
SEL_TREE *rt2= tree2;
if (no_merges1)
rt1= new SEL_TREE(tree1, TRUE, param);
if (no_merges2)
rt2= new SEL_TREE(tree2, TRUE, param);
if (!rt1 || !rt2 ||
result->merges.push_back(imerge_from_ranges) ||
imerge_from_ranges->or_sel_tree(param, rt1) ||
imerge_from_ranges->or_sel_tree(param, rt2))
result= NULL;
}
if (!result)
DBUG_RETURN(result);
result->type= tree1->type;
if (!no_merges1 && !no_merges2 &&
!imerge_list_or_list(param, &tree1->merges, &tree2->merges))
{
/* Build the imerges for the formula (2) */
imerge_list_and_list(&result->merges, &tree1->merges);
}
/* Build the imerges for the formulas (3) and (4) */
for (uint i=0; i < 2; i++)
{
List<SEL_IMERGE> merges;
SEL_TREE *rt= rtree[i];
SEL_IMERGE *im= imerge[1-i];
if (rt && im && !merges.push_back(im) &&
!imerge_list_or_tree(param, &merges, rt))
imerge_list_and_list(&result->merges, &merges);
}
DBUG_RETURN(result);
}
/* And key trees where key1->part < key2 -> part */
static SEL_ARG *
and_all_keys(RANGE_OPT_PARAM *param, SEL_ARG *key1, SEL_ARG *key2,
uint clone_flag)
{
SEL_ARG *next;
ulong use_count=key1->use_count;
if (key1->elements != 1)
{
key2->use_count+=key1->elements-1; //psergey: why we don't count that key1 has n-k-p?
key2->increment_use_count((int) key1->elements-1);
}
if (key1->type == SEL_ARG::MAYBE_KEY)
{
key1->right= key1->left= &null_element;
key1->next= key1->prev= 0;
}
for (next=key1->first(); next ; next=next->next)
{
if (next->next_key_part)
{
SEL_ARG *tmp= key_and(param, next->next_key_part, key2, clone_flag);
if (tmp && tmp->type == SEL_ARG::IMPOSSIBLE)
{
key1=key1->tree_delete(next);
continue;
}
next->next_key_part=tmp;
if (use_count)
next->increment_use_count(use_count);
if (param->alloced_sel_args > SEL_ARG::MAX_SEL_ARGS)
break;
}
else
next->next_key_part=key2;
}
if (!key1)
return &null_element; // Impossible ranges
key1->use_count++;
key1->max_part_no= MY_MAX(key2->max_part_no, key2->part+1);
return key1;
}
/*
Produce a SEL_ARG graph that represents "key1 AND key2"
SYNOPSIS
key_and()
param Range analysis context (needed to track if we have allocated
too many SEL_ARGs)
key1 First argument, root of its RB-tree
key2 Second argument, root of its RB-tree
RETURN
RB-tree root of the resulting SEL_ARG graph.
NULL if the result of AND operation is an empty interval {0}.
*/
static SEL_ARG *
key_and(RANGE_OPT_PARAM *param, SEL_ARG *key1, SEL_ARG *key2, uint clone_flag)
{
if (!key1)
return key2;
if (!key2)
return key1;
if (key1->part != key2->part)
{
if (key1->part > key2->part)
{
swap_variables(SEL_ARG *, key1, key2);
clone_flag=swap_clone_flag(clone_flag);
}
// key1->part < key2->part
key1->use_count--;
if (key1->use_count > 0)
if (!(key1= key1->clone_tree(param)))
return 0; // OOM
return and_all_keys(param, key1, key2, clone_flag);
}
if (((clone_flag & CLONE_KEY2_MAYBE) &&
!(clone_flag & CLONE_KEY1_MAYBE) &&
key2->type != SEL_ARG::MAYBE_KEY) ||
key1->type == SEL_ARG::MAYBE_KEY)
{ // Put simple key in key2
swap_variables(SEL_ARG *, key1, key2);
clone_flag=swap_clone_flag(clone_flag);
}
/* If one of the key is MAYBE_KEY then the found region may be smaller */
if (key2->type == SEL_ARG::MAYBE_KEY)
{
if (key1->use_count > 1)
{
key1->use_count--;
if (!(key1=key1->clone_tree(param)))
return 0; // OOM
key1->use_count++;
}
if (key1->type == SEL_ARG::MAYBE_KEY)
{ // Both are maybe key
key1->next_key_part=key_and(param, key1->next_key_part,
key2->next_key_part, clone_flag);
if (key1->next_key_part &&
key1->next_key_part->type == SEL_ARG::IMPOSSIBLE)
return key1;
}
else
{
key1->maybe_smaller();
if (key2->next_key_part)
{
key1->use_count--; // Incremented in and_all_keys
return and_all_keys(param, key1, key2, clone_flag);
}
key2->use_count--; // Key2 doesn't have a tree
}
return key1;
}
if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
{
/* TODO: why not leave one of the trees? */
key1->free_tree();
key2->free_tree();
return 0; // Can't optimize this
}
key1->use_count--;
key2->use_count--;
SEL_ARG *e1=key1->first(), *e2=key2->first(), *new_tree=0;
uint max_part_no= MY_MAX(key1->max_part_no, key2->max_part_no);
while (e1 && e2)
{
int cmp=e1->cmp_min_to_min(e2);
if (cmp < 0)
{
if (get_range(&e1,&e2,key1))
continue;
}
else if (get_range(&e2,&e1,key2))
continue;
SEL_ARG *next=key_and(param, e1->next_key_part, e2->next_key_part,
clone_flag);
e1->incr_refs();
e2->incr_refs();
if (!next || next->type != SEL_ARG::IMPOSSIBLE)
{
SEL_ARG *new_arg= e1->clone_and(e2);
if (!new_arg)
return &null_element; // End of memory
new_arg->next_key_part=next;
if (!new_tree)
{
new_tree=new_arg;
}
else
new_tree=new_tree->insert(new_arg);
}
if (e1->cmp_max_to_max(e2) < 0)
e1=e1->next; // e1 can't overlapp next e2
else
e2=e2->next;
}
key1->free_tree();
key2->free_tree();
if (!new_tree)
return &null_element; // Impossible range
new_tree->max_part_no= max_part_no;
return new_tree;
}
static bool
get_range(SEL_ARG **e1,SEL_ARG **e2,SEL_ARG *root1)
{
(*e1)=root1->find_range(*e2); // first e1->min < e2->min
if ((*e1)->cmp_max_to_min(*e2) < 0)
{
if (!((*e1)=(*e1)->next))
return 1;
if ((*e1)->cmp_min_to_max(*e2) > 0)
{
(*e2)=(*e2)->next;
return 1;
}
}
return 0;
}
/**
Combine two range expression under a common OR. On a logical level, the
transformation is key_or( expr1, expr2 ) => expr1 OR expr2.
Both expressions are assumed to be in the SEL_ARG format. In a logic sense,
theformat is reminiscent of DNF, since an expression such as the following
( 1 < kp1 < 10 AND p1 ) OR ( 10 <= kp2 < 20 AND p2 )
where there is a key consisting of keyparts ( kp1, kp2, ..., kpn ) and p1
and p2 are valid SEL_ARG expressions over keyparts kp2 ... kpn, is a valid
SEL_ARG condition. The disjuncts appear ordered by the minimum endpoint of
the first range and ranges must not overlap. It follows that they are also
ordered by maximum endpoints. Thus
( 1 < kp1 <= 2 AND ( kp2 = 2 OR kp2 = 3 ) ) OR kp1 = 3
Is a a valid SER_ARG expression for a key of at least 2 keyparts.
For simplicity, we will assume that expr2 is a single range predicate,
i.e. on the form ( a < x < b AND ... ). It is easy to generalize to a
disjunction of several predicates by subsequently call key_or for each
disjunct.
The algorithm iterates over each disjunct of expr1, and for each disjunct
where the first keypart's range overlaps with the first keypart's range in
expr2:
If the predicates are equal for the rest of the keyparts, or if there are
no more, the range in expr2 has its endpoints copied in, and the SEL_ARG
node in expr2 is deallocated. If more ranges became connected in expr1, the
surplus is also dealocated. If they differ, two ranges are created.
- The range leading up to the overlap. Empty if endpoints are equal.
- The overlapping sub-range. May be the entire range if they are equal.
Finally, there may be one more range if expr2's first keypart's range has a
greater maximum endpoint than the last range in expr1.
For the overlapping sub-range, we recursively call key_or. Thus in order to
compute key_or of
(1) ( 1 < kp1 < 10 AND 1 < kp2 < 10 )
(2) ( 2 < kp1 < 20 AND 4 < kp2 < 20 )
We create the ranges 1 < kp <= 2, 2 < kp1 < 10, 10 <= kp1 < 20. For the
first one, we simply hook on the condition for the second keypart from (1)
: 1 < kp2 < 10. For the second range 2 < kp1 < 10, key_or( 1 < kp2 < 10, 4
< kp2 < 20 ) is called, yielding 1 < kp2 < 20. For the last range, we reuse
the range 4 < kp2 < 20 from (2) for the second keypart. The result is thus
( 1 < kp1 <= 2 AND 1 < kp2 < 10 ) OR
( 2 < kp1 < 10 AND 1 < kp2 < 20 ) OR
( 10 <= kp1 < 20 AND 4 < kp2 < 20 )
*/
static SEL_ARG *
key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)
{
if (!key1)
{
if (key2)
{
key2->use_count--;
key2->free_tree();
}
return 0;
}
if (!key2)
{
key1->use_count--;
key1->free_tree();
return 0;
}
key1->use_count--;
key2->use_count--;
if (key1->part != key2->part ||
(key1->min_flag | key2->min_flag) & GEOM_FLAG)
{
key1->free_tree();
key2->free_tree();
return 0; // Can't optimize this
}
// If one of the key is MAYBE_KEY then the found region may be bigger
if (key1->type == SEL_ARG::MAYBE_KEY)
{
key2->free_tree();
key1->use_count++;
return key1;
}
if (key2->type == SEL_ARG::MAYBE_KEY)
{
key1->free_tree();
key2->use_count++;
return key2;
}
if (key1->use_count > 0)
{
if (key2->use_count == 0 || key1->elements > key2->elements)
{
swap_variables(SEL_ARG *,key1,key2);
}
if (key1->use_count > 0 && !(key1=key1->clone_tree(param)))
return 0; // OOM
}
// Add tree at key2 to tree at key1
bool key2_shared=key2->use_count != 0;
key1->maybe_flag|=key2->maybe_flag;
/*
Notation for illustrations used in the rest of this function:
Range: [--------]
^ ^
start stop
Two overlapping ranges:
[-----] [----] [--]
[---] or [---] or [-------]
Ambiguity: ***
The range starts or stops somewhere in the "***" range.
Example: a starts before b and may end before/the same plase/after b
a: [----***]
b: [---]
Adjacent ranges:
Ranges that meet but do not overlap. Example: a = "x < 3", b = "x >= 3"
a: ----]
b: [----
*/
uint max_part_no= MY_MAX(key1->max_part_no, key2->max_part_no);
for (key2=key2->first(); key2; )
{
/*
key1 consists of one or more ranges. tmp is the range currently
being handled.
initialize tmp to the latest range in key1 that starts the same
place or before the range in key2 starts
key2: [------]
key1: [---] [-----] [----]
^
tmp
*/
SEL_ARG *tmp=key1->find_range(key2);
/*
Used to describe how two key values are positioned compared to
each other. Consider key_value_a.<cmp_func>(key_value_b):
-2: key_value_a is smaller than key_value_b, and they are adjacent
-1: key_value_a is smaller than key_value_b (not adjacent)
0: the key values are equal
1: key_value_a is bigger than key_value_b (not adjacent)
-2: key_value_a is bigger than key_value_b, and they are adjacent
Example: "cmp= tmp->cmp_max_to_min(key2)"
key2: [-------- (10 <= x ...)
tmp: -----] (... x < 10) => cmp==-2
tmp: ----] (... x <= 9) => cmp==-1
tmp: ------] (... x = 10) => cmp== 0
tmp: --------] (... x <= 12) => cmp== 1
(cmp == 2 does not make sense for cmp_max_to_min())
*/
int cmp= 0;
if (!tmp)
{
/*
The range in key2 starts before the first range in key1. Use
the first range in key1 as tmp.
key2: [--------]
key1: [****--] [----] [-------]
^
tmp
*/
tmp=key1->first();
cmp= -1;
}
else if ((cmp= tmp->cmp_max_to_min(key2)) < 0)
{
/*
This is the case:
key2: [-------]
tmp: [----**]
*/
SEL_ARG *next=tmp->next;
if (cmp == -2 && eq_tree(tmp->next_key_part,key2->next_key_part))
{
/*
Adjacent (cmp==-2) and equal next_key_parts => ranges can be merged
This is the case:
key2: [-------]
tmp: [----]
Result:
key2: [-------------] => inserted into key1 below
tmp: => deleted
*/
SEL_ARG *key2_next=key2->next;
if (key2_shared)
{
if (!(key2=new SEL_ARG(*key2)))
return 0; // out of memory
key2->increment_use_count(key1->use_count+1);
key2->next=key2_next; // New copy of key2
}
key2->copy_min(tmp);
if (!(key1=key1->tree_delete(tmp)))
{ // Only one key in tree
key1=key2;
key1->make_root();
key2=key2_next;
break;
}
}
if (!(tmp=next)) // Move to next range in key1. Now tmp.min > key2.min
break; // No more ranges in key1. Copy rest of key2
}
if (cmp < 0)
{
/*
This is the case:
key2: [--***]
tmp: [----]
*/
int tmp_cmp;
if ((tmp_cmp=tmp->cmp_min_to_max(key2)) > 0)
{
/*
This is the case:
key2: [------**]
tmp: [----]
*/
if (tmp_cmp == 2 && eq_tree(tmp->next_key_part,key2->next_key_part))
{
/*
Adjacent ranges with equal next_key_part. Merge like this:
This is the case:
key2: [------]
tmp: [-----]
Result:
key2: [------]
tmp: [-------------]
Then move on to next key2 range.
*/
tmp->copy_min_to_min(key2);
key1->merge_flags(key2);
if (tmp->min_flag & NO_MIN_RANGE &&
tmp->max_flag & NO_MAX_RANGE)
{
if (key1->maybe_flag)
return new SEL_ARG(SEL_ARG::MAYBE_KEY);
return 0;
}
key2->increment_use_count(-1); // Free not used tree
key2=key2->next;
continue;
}
else
{
/*
key2 not adjacent to tmp or has different next_key_part.
Insert into key1 and move to next range in key2
This is the case:
key2: [------**]
tmp: [----]
Result:
key1_ [------**][----]
^ ^
insert tmp
*/
SEL_ARG *next=key2->next;
if (key2_shared)
{
SEL_ARG *cpy= new SEL_ARG(*key2); // Must make copy
if (!cpy)
return 0; // OOM
key1=key1->insert(cpy);
key2->increment_use_count(key1->use_count+1);
}
else
key1=key1->insert(key2); // Will destroy key2_root
key2=next;
continue;
}
}
}
/*
The ranges in tmp and key2 are overlapping:
key2: [----------]
tmp: [*****-----*****]
Corollary: tmp.min <= key2.max
*/
if (eq_tree(tmp->next_key_part,key2->next_key_part))
{
// Merge overlapping ranges with equal next_key_part
if (tmp->is_same(key2))
{
/*
Found exact match of key2 inside key1.
Use the relevant range in key1.
*/
tmp->merge_flags(key2); // Copy maybe flags
key2->increment_use_count(-1); // Free not used tree
}
else
{
SEL_ARG *last= tmp;
SEL_ARG *first= tmp;
/*
Find the last range in key1 that overlaps key2 and
where all ranges first...last have the same next_key_part as
key2.
key2: [****----------------------*******]
key1: [--] [----] [---] [-----] [xxxx]
^ ^ ^
first last different next_key_part
Since key2 covers them, the ranges between first and last
are merged into one range by deleting first...last-1 from
the key1 tree. In the figure, this applies to first and the
two consecutive ranges. The range of last is then extended:
* last.min: Set to MY_MIN(key2.min, first.min)
* last.max: If there is a last->next that overlaps key2 (i.e.,
last->next has a different next_key_part):
Set adjacent to last->next.min
Otherwise: Set to MY_MAX(key2.max, last.max)
Result:
key2: [****----------------------*******]
[--] [----] [---] => deleted from key1
key1: [**------------------------***][xxxx]
^ ^
tmp=last different next_key_part
*/
while (last->next && last->next->cmp_min_to_max(key2) <= 0 &&
eq_tree(last->next->next_key_part,key2->next_key_part))
{
/*
last->next is covered by key2 and has same next_key_part.
last can be deleted
*/
SEL_ARG *save=last;
last=last->next;
key1=key1->tree_delete(save);
}
// Redirect tmp to last which will cover the entire range
tmp= last;
/*
We need the minimum endpoint of first so we can compare it
with the minimum endpoint of the enclosing key2 range.
*/
last->copy_min(first);
bool full_range= last->copy_min(key2);
if (!full_range)
{
if (last->next && key2->cmp_max_to_min(last->next) >= 0)
{
/*
This is the case:
key2: [-------------]
key1: [***------] [xxxx]
^ ^
last different next_key_part
Extend range of last up to last->next:
key2: [-------------]
key1: [***--------][xxxx]
*/
last->copy_min_to_max(last->next);
}
else
/*
This is the case:
key2: [--------*****]
key1: [***---------] [xxxx]
^ ^
last different next_key_part
Extend range of last up to MY_MAX(last.max, key2.max):
key2: [--------*****]
key1: [***----------**] [xxxx]
*/
full_range= last->copy_max(key2);
}
if (full_range)
{ // Full range
key1->free_tree();
for (; key2 ; key2=key2->next)
key2->increment_use_count(-1); // Free not used tree
if (key1->maybe_flag)
return new SEL_ARG(SEL_ARG::MAYBE_KEY);
return 0;
}
}
}
if (cmp >= 0 && tmp->cmp_min_to_min(key2) < 0)
{
/*
This is the case ("cmp>=0" means that tmp.max >= key2.min):
key2: [----]
tmp: [------------*****]
*/
if (!tmp->next_key_part)
{
/*
tmp->next_key_part is empty: cut the range that is covered
by tmp from key2.
Reason: (key2->next_key_part OR tmp->next_key_part) will be
empty and therefore equal to tmp->next_key_part. Thus, this
part of the key2 range is completely covered by tmp.
*/
if (tmp->cmp_max_to_max(key2) >= 0)
{
/*
tmp covers the entire range in key2.
key2: [----]
tmp: [-----------------]
Move on to next range in key2
*/
key2->increment_use_count(-1); // Free not used tree
key2=key2->next;
continue;
}
else
{
/*
This is the case:
key2: [-------]
tmp: [---------]
Result:
key2: [---]
tmp: [---------]
*/
if (key2->use_count)
{
SEL_ARG *key2_cpy= new SEL_ARG(*key2);
if (key2_cpy)
return 0;
key2= key2_cpy;
}
key2->copy_max_to_min(tmp);
continue;
}
}
/*
The ranges are overlapping but have not been merged because
next_key_part of tmp and key2 differ.
key2: [----]
tmp: [------------*****]
Split tmp in two where key2 starts:
key2: [----]
key1: [--------][--*****]
^ ^
insert tmp
*/
SEL_ARG *new_arg=tmp->clone_first(key2);
if (!new_arg)
return 0; // OOM
if ((new_arg->next_key_part= tmp->next_key_part))
new_arg->increment_use_count(key1->use_count+1);
tmp->copy_min_to_min(key2);
key1=key1->insert(new_arg);
} // tmp.min >= key2.min due to this if()
/*
Now key2.min <= tmp.min <= key2.max:
key2: [---------]
tmp: [****---*****]
*/
SEL_ARG key2_cpy(*key2); // Get copy we can modify
for (;;)
{
if (tmp->cmp_min_to_min(&key2_cpy) > 0)
{
/*
This is the case:
key2_cpy: [------------]
key1: [-*****]
^
tmp
Result:
key2_cpy: [---]
key1: [-------][-*****]
^ ^
insert tmp
*/
SEL_ARG *new_arg=key2_cpy.clone_first(tmp);
if (!new_arg)
return 0; // OOM
if ((new_arg->next_key_part=key2_cpy.next_key_part))
new_arg->increment_use_count(key1->use_count+1);
key1=key1->insert(new_arg);
key2_cpy.copy_min_to_min(tmp);
}
// Now key2_cpy.min == tmp.min
if ((cmp= tmp->cmp_max_to_max(&key2_cpy)) <= 0)
{
/*
tmp.max <= key2_cpy.max:
key2_cpy: a) [-------] or b) [----]
tmp: [----] [----]
Steps:
1) Update next_key_part of tmp: OR it with key2_cpy->next_key_part.
2) If case a: Insert range [tmp.max, key2_cpy.max] into key1 using
next_key_part of key2_cpy
Result:
key1: a) [----][-] or b) [----]
*/
tmp->maybe_flag|= key2_cpy.maybe_flag;
key2_cpy.increment_use_count(key1->use_count+1);
tmp->next_key_part= key_or(param, tmp->next_key_part,
key2_cpy.next_key_part);
if (!cmp)
break; // case b: done with this key2 range
// Make key2_cpy the range [tmp.max, key2_cpy.max]
key2_cpy.copy_max_to_min(tmp);
if (!(tmp=tmp->next))
{
/*
No more ranges in key1. Insert key2_cpy and go to "end"
label to insert remaining ranges in key2 if any.
*/
SEL_ARG *tmp2= new SEL_ARG(key2_cpy);
if (!tmp2)
return 0; // OOM
key1=key1->insert(tmp2);
key2=key2->next;
goto end;
}
if (tmp->cmp_min_to_max(&key2_cpy) > 0)
{
/*
The next range in key1 does not overlap with key2_cpy.
Insert this range into key1 and move on to the next range
in key2.
*/
SEL_ARG *tmp2= new SEL_ARG(key2_cpy);
if (!tmp2)
return 0; // OOM
key1=key1->insert(tmp2);
break;
}
/*
key2_cpy overlaps with the next range in key1 and the case
is now "key2.min <= tmp.min <= key2.max". Go back to for(;;)
to handle this situation.
*/
continue;
}
else
{
/*
This is the case:
key2_cpy: [-------]
tmp: [------------]
Result:
key1: [-------][---]
^ ^
new_arg tmp
Steps:
0) If tmp->next_key_part is empty: do nothing. Reason:
(key2_cpy->next_key_part OR tmp->next_key_part) will be
empty and therefore equal to tmp->next_key_part. Thus,
the range in key2_cpy is completely covered by tmp
1) Make new_arg with range [tmp.min, key2_cpy.max].
new_arg->next_key_part is OR between next_key_part
of tmp and key2_cpy
2) Make tmp the range [key2.max, tmp.max]
3) Insert new_arg into key1
*/
if (!tmp->next_key_part) // Step 0
{
key2_cpy.increment_use_count(-1); // Free not used tree
break;
}
SEL_ARG *new_arg=tmp->clone_last(&key2_cpy);
if (!new_arg)
return 0; // OOM
tmp->copy_max_to_min(&key2_cpy);
tmp->increment_use_count(key1->use_count+1);
/* Increment key count as it may be used for next loop */
key2_cpy.increment_use_count(1);
new_arg->next_key_part= key_or(param, tmp->next_key_part,
key2_cpy.next_key_part);
key1=key1->insert(new_arg);
break;
}
}
// Move on to next range in key2
key2=key2->next;
}
end:
/*
Add key2 ranges that are non-overlapping with and higher than the
highest range in key1.
*/
while (key2)
{
SEL_ARG *next=key2->next;
if (key2_shared)
{
SEL_ARG *tmp=new SEL_ARG(*key2); // Must make copy
if (!tmp)
return 0;
key2->increment_use_count(key1->use_count+1);
key1=key1->insert(tmp);
}
else
key1=key1->insert(key2); // Will destroy key2_root
key2=next;
}
key1->use_count++;
key1->max_part_no= max_part_no;
return key1;
}
/* Compare if two trees are equal */
static bool eq_tree(SEL_ARG* a,SEL_ARG *b)
{
if (a == b)
return 1;
if (!a || !b || !a->is_same(b))
return 0;
if (a->left != &null_element && b->left != &null_element)
{
if (!eq_tree(a->left,b->left))
return 0;
}
else if (a->left != &null_element || b->left != &null_element)
return 0;
if (a->right != &null_element && b->right != &null_element)
{
if (!eq_tree(a->right,b->right))
return 0;
}
else if (a->right != &null_element || b->right != &null_element)
return 0;
if (a->next_key_part != b->next_key_part)
{ // Sub range
if (!a->next_key_part != !b->next_key_part ||
!eq_tree(a->next_key_part, b->next_key_part))
return 0;
}
return 1;
}
SEL_ARG *
SEL_ARG::insert(SEL_ARG *key)
{
SEL_ARG *element,**UNINIT_VAR(par),*UNINIT_VAR(last_element);
for (element= this; element != &null_element ; )
{
last_element=element;
if (key->cmp_min_to_min(element) > 0)
{
par= &element->right; element= element->right;
}
else
{
par = &element->left; element= element->left;
}
}
*par=key;
key->parent=last_element;
/* Link in list */
if (par == &last_element->left)
{
key->next=last_element;
if ((key->prev=last_element->prev))
key->prev->next=key;
last_element->prev=key;
}
else
{
if ((key->next=last_element->next))
key->next->prev=key;
key->prev=last_element;
last_element->next=key;
}
key->left=key->right= &null_element;
SEL_ARG *root=rb_insert(key); // rebalance tree
root->use_count=this->use_count; // copy root info
root->elements= this->elements+1;
root->maybe_flag=this->maybe_flag;
return root;
}
/*
** Find best key with min <= given key
** Because the call context this should never return 0 to get_range
*/
SEL_ARG *
SEL_ARG::find_range(SEL_ARG *key)
{
SEL_ARG *element=this,*found=0;
for (;;)
{
if (element == &null_element)
return found;
int cmp=element->cmp_min_to_min(key);
if (cmp == 0)
return element;
if (cmp < 0)
{
found=element;
element=element->right;
}
else
element=element->left;
}
}
/*
Remove a element from the tree
SYNOPSIS
tree_delete()
key Key that is to be deleted from tree (this)
NOTE
This also frees all sub trees that is used by the element
RETURN
root of new tree (with key deleted)
*/
SEL_ARG *
SEL_ARG::tree_delete(SEL_ARG *key)
{
enum leaf_color remove_color;
SEL_ARG *root,*nod,**par,*fix_par;
DBUG_ENTER("tree_delete");
root=this;
this->parent= 0;
/* Unlink from list */
if (key->prev)
key->prev->next=key->next;
if (key->next)
key->next->prev=key->prev;
key->increment_use_count(-1);
if (!key->parent)
par= &root;
else
par=key->parent_ptr();
if (key->left == &null_element)
{
*par=nod=key->right;
fix_par=key->parent;
if (nod != &null_element)
nod->parent=fix_par;
remove_color= key->color;
}
else if (key->right == &null_element)
{
*par= nod=key->left;
nod->parent=fix_par=key->parent;
remove_color= key->color;
}
else
{
SEL_ARG *tmp=key->next; // next bigger key (exist!)
nod= *tmp->parent_ptr()= tmp->right; // unlink tmp from tree
fix_par=tmp->parent;
if (nod != &null_element)
nod->parent=fix_par;
remove_color= tmp->color;
tmp->parent=key->parent; // Move node in place of key
(tmp->left=key->left)->parent=tmp;
if ((tmp->right=key->right) != &null_element)
tmp->right->parent=tmp;
tmp->color=key->color;
*par=tmp;
if (fix_par == key) // key->right == key->next
fix_par=tmp; // new parent of nod
}
if (root == &null_element)
DBUG_RETURN(0); // Maybe root later
if (remove_color == BLACK)
root=rb_delete_fixup(root,nod,fix_par);
test_rb_tree(root,root->parent);
root->use_count=this->use_count; // Fix root counters
root->elements=this->elements-1;
root->maybe_flag=this->maybe_flag;
DBUG_RETURN(root);
}
/* Functions to fix up the tree after insert and delete */
static void left_rotate(SEL_ARG **root,SEL_ARG *leaf)
{
SEL_ARG *y=leaf->right;
leaf->right=y->left;
if (y->left != &null_element)
y->left->parent=leaf;
if (!(y->parent=leaf->parent))
*root=y;
else
*leaf->parent_ptr()=y;
y->left=leaf;
leaf->parent=y;
}
static void right_rotate(SEL_ARG **root,SEL_ARG *leaf)
{
SEL_ARG *y=leaf->left;
leaf->left=y->right;
if (y->right != &null_element)
y->right->parent=leaf;
if (!(y->parent=leaf->parent))
*root=y;
else
*leaf->parent_ptr()=y;
y->right=leaf;
leaf->parent=y;
}
SEL_ARG *
SEL_ARG::rb_insert(SEL_ARG *leaf)
{
SEL_ARG *y,*par,*par2,*root;
root= this; root->parent= 0;
leaf->color=RED;
while (leaf != root && (par= leaf->parent)->color == RED)
{ // This can't be root or 1 level under
if (par == (par2= leaf->parent->parent)->left)
{
y= par2->right;
if (y->color == RED)
{
par->color=BLACK;
y->color=BLACK;
leaf=par2;
leaf->color=RED; /* And the loop continues */
}
else
{
if (leaf == par->right)
{
left_rotate(&root,leaf->parent);
par=leaf; /* leaf is now parent to old leaf */
}
par->color=BLACK;
par2->color=RED;
right_rotate(&root,par2);
break;
}
}
else
{
y= par2->left;
if (y->color == RED)
{
par->color=BLACK;
y->color=BLACK;
leaf=par2;
leaf->color=RED; /* And the loop continues */
}
else
{
if (leaf == par->left)
{
right_rotate(&root,par);
par=leaf;
}
par->color=BLACK;
par2->color=RED;
left_rotate(&root,par2);
break;
}
}
}
root->color=BLACK;
test_rb_tree(root,root->parent);
return root;
}
SEL_ARG *rb_delete_fixup(SEL_ARG *root,SEL_ARG *key,SEL_ARG *par)
{
SEL_ARG *x,*w;
root->parent=0;
x= key;
while (x != root && x->color == SEL_ARG::BLACK)
{
if (x == par->left)
{
w=par->right;
if (w->color == SEL_ARG::RED)
{
w->color=SEL_ARG::BLACK;
par->color=SEL_ARG::RED;
left_rotate(&root,par);
w=par->right;
}
if (w->left->color == SEL_ARG::BLACK && w->right->color == SEL_ARG::BLACK)
{
w->color=SEL_ARG::RED;
x=par;
}
else
{
if (w->right->color == SEL_ARG::BLACK)
{
w->left->color=SEL_ARG::BLACK;
w->color=SEL_ARG::RED;
right_rotate(&root,w);
w=par->right;
}
w->color=par->color;
par->color=SEL_ARG::BLACK;
w->right->color=SEL_ARG::BLACK;
left_rotate(&root,par);
x=root;
break;
}
}
else
{
w=par->left;
if (w->color == SEL_ARG::RED)
{
w->color=SEL_ARG::BLACK;
par->color=SEL_ARG::RED;
right_rotate(&root,par);
w=par->left;
}
if (w->right->color == SEL_ARG::BLACK && w->left->color == SEL_ARG::BLACK)
{
w->color=SEL_ARG::RED;
x=par;
}
else
{
if (w->left->color == SEL_ARG::BLACK)
{
w->right->color=SEL_ARG::BLACK;
w->color=SEL_ARG::RED;
left_rotate(&root,w);
w=par->left;
}
w->color=par->color;
par->color=SEL_ARG::BLACK;
w->left->color=SEL_ARG::BLACK;
right_rotate(&root,par);
x=root;
break;
}
}
par=x->parent;
}
x->color=SEL_ARG::BLACK;
return root;
}
/* Test that the properties for a red-black tree hold */
#ifdef EXTRA_DEBUG
int test_rb_tree(SEL_ARG *element,SEL_ARG *parent)
{
int count_l,count_r;
if (element == &null_element)
return 0; // Found end of tree
if (element->parent != parent)
{
sql_print_error("Wrong tree: Parent doesn't point at parent");
return -1;
}
if (element->color == SEL_ARG::RED &&
(element->left->color == SEL_ARG::RED ||
element->right->color == SEL_ARG::RED))
{
sql_print_error("Wrong tree: Found two red in a row");
return -1;
}
if (element->left == element->right && element->left != &null_element)
{ // Dummy test
sql_print_error("Wrong tree: Found right == left");
return -1;
}
count_l=test_rb_tree(element->left,element);
count_r=test_rb_tree(element->right,element);
if (count_l >= 0 && count_r >= 0)
{
if (count_l == count_r)
return count_l+(element->color == SEL_ARG::BLACK);
sql_print_error("Wrong tree: Incorrect black-count: %d - %d",
count_l,count_r);
}
return -1; // Error, no more warnings
}
/**
Count how many times SEL_ARG graph "root" refers to its part "key" via
transitive closure.
@param root An RB-Root node in a SEL_ARG graph.
@param key Another RB-Root node in that SEL_ARG graph.
The passed "root" node may refer to "key" node via root->next_key_part,
root->next->n
This function counts how many times the node "key" is referred (via
SEL_ARG::next_key_part) by
- intervals of RB-tree pointed by "root",
- intervals of RB-trees that are pointed by SEL_ARG::next_key_part from
intervals of RB-tree pointed by "root",
- and so on.
Here is an example (horizontal links represent next_key_part pointers,
vertical links - next/prev prev pointers):
+----+ $
|root|-----------------+
+----+ $ |
| $ |
| $ |
+----+ +---+ $ | +---+ Here the return value
| |- ... -| |---$-+--+->|key| will be 4.
+----+ +---+ $ | | +---+
| $ | |
... $ | |
| $ | |
+----+ +---+ $ | |
| |---| |---------+ |
+----+ +---+ $ |
| | $ |
... +---+ $ |
| |------------+
+---+ $
@return
Number of links to "key" from nodes reachable from "root".
*/
static ulong count_key_part_usage(SEL_ARG *root, SEL_ARG *key)
{
ulong count= 0;
for (root=root->first(); root ; root=root->next)
{
if (root->next_key_part)
{
if (root->next_key_part == key)
count++;
if (root->next_key_part->part < key->part)
count+=count_key_part_usage(root->next_key_part,key);
}
}
return count;
}
/*
Check if SEL_ARG::use_count value is correct
SYNOPSIS
SEL_ARG::test_use_count()
root The root node of the SEL_ARG graph (an RB-tree root node that
has the least value of sel_arg->part in the entire graph, and
thus is the "origin" of the graph)
DESCRIPTION
Check if SEL_ARG::use_count value is correct. See the definition of
use_count for what is "correct".
*/
void SEL_ARG::test_use_count(SEL_ARG *root)
{
uint e_count=0;
if (this->type != SEL_ARG::KEY_RANGE)
return;
for (SEL_ARG *pos=first(); pos ; pos=pos->next)
{
e_count++;
if (pos->next_key_part)
{
ulong count=count_key_part_usage(root,pos->next_key_part);
if (count > pos->next_key_part->use_count)
{
sql_print_information("Use_count: Wrong count for key at 0x%lx, %lu "
"should be %lu", (long unsigned int)pos,
pos->next_key_part->use_count, count);
return;
}
pos->next_key_part->test_use_count(root);
}
}
if (e_count != elements)
sql_print_warning("Wrong use count: %u (should be %u) for tree at 0x%lx",
e_count, elements, (long unsigned int) this);
}
#endif
/*
Calculate cost and E(#rows) for a given index and intervals tree
SYNOPSIS
check_quick_select()
param Parameter from test_quick_select
idx Number of index to use in PARAM::key SEL_TREE::key
index_only TRUE - assume only index tuples will be accessed
FALSE - assume full table rows will be read
tree Transformed selection condition, tree->key[idx] holds
the intervals for the given index.
update_tbl_stats TRUE <=> update table->quick_* with information
about range scan we've evaluated.
mrr_flags INOUT MRR access flags
cost OUT Scan cost
NOTES
param->is_ror_scan is set to reflect if the key scan is a ROR (see
is_key_scan_ror function for more info)
param->table->quick_*, param->range_count (and maybe others) are
updated with data of given key scan, see quick_range_seq_next for details.
RETURN
Estimate # of records to be retrieved.
HA_POS_ERROR if estimate calculation failed due to table handler problems.
*/
static
ha_rows check_quick_select(PARAM *param, uint idx, bool index_only,
SEL_ARG *tree, bool update_tbl_stats,
uint *mrr_flags, uint *bufsize, Cost_estimate *cost)
{
SEL_ARG_RANGE_SEQ seq;
RANGE_SEQ_IF seq_if = {NULL, sel_arg_range_seq_init, sel_arg_range_seq_next, 0, 0};
handler *file= param->table->file;
ha_rows rows= HA_POS_ERROR;
uint keynr= param->real_keynr[idx];
DBUG_ENTER("check_quick_select");
/* Handle cases when we don't have a valid non-empty list of range */
if (!tree)
DBUG_RETURN(HA_POS_ERROR);
if (tree->type == SEL_ARG::IMPOSSIBLE)
DBUG_RETURN(0L);
if (tree->type != SEL_ARG::KEY_RANGE || tree->part != 0)
DBUG_RETURN(HA_POS_ERROR);
seq.keyno= idx;
seq.real_keyno= keynr;
seq.param= param;
seq.start= tree;
param->range_count=0;
param->max_key_part=0;
param->is_ror_scan= TRUE;
if (file->index_flags(keynr, 0, TRUE) & HA_KEY_SCAN_NOT_ROR)
param->is_ror_scan= FALSE;
*mrr_flags= param->force_default_mrr? HA_MRR_USE_DEFAULT_IMPL: 0;
/*
Pass HA_MRR_SORTED to see if MRR implementation can handle sorting.
*/
*mrr_flags|= HA_MRR_NO_ASSOCIATION | HA_MRR_SORTED;
bool pk_is_clustered= file->primary_key_is_clustered();
if (index_only &&
(file->index_flags(keynr, param->max_key_part, 1) & HA_KEYREAD_ONLY) &&
!(file->index_flags(keynr, param->max_key_part, 1) & HA_CLUSTERED_INDEX))
*mrr_flags |= HA_MRR_INDEX_ONLY;
if (param->thd->lex->sql_command != SQLCOM_SELECT)
*mrr_flags |= HA_MRR_USE_DEFAULT_IMPL;
*bufsize= param->thd->variables.mrr_buff_size;
/*
Skip materialized derived table/view result table from MRR check as
they aren't contain any data yet.
*/
if (param->table->pos_in_table_list->is_non_derived())
rows= file->multi_range_read_info_const(keynr, &seq_if, (void*)&seq, 0,
bufsize, mrr_flags, cost);
if (rows != HA_POS_ERROR)
{
param->quick_rows[keynr]= rows;
param->possible_keys.set_bit(keynr);
if (update_tbl_stats)
{
param->table->quick_keys.set_bit(keynr);
param->table->quick_key_parts[keynr]= param->max_key_part+1;
param->table->quick_n_ranges[keynr]= param->range_count;
param->table->quick_condition_rows=
MY_MIN(param->table->quick_condition_rows, rows);
param->table->quick_rows[keynr]= rows;
}
}
/* Figure out if the key scan is ROR (returns rows in ROWID order) or not */
enum ha_key_alg key_alg= param->table->key_info[seq.real_keyno].algorithm;
if ((key_alg != HA_KEY_ALG_BTREE) && (key_alg!= HA_KEY_ALG_UNDEF))
{
/*
All scans are non-ROR scans for those index types.
TODO: Don't have this logic here, make table engines return
appropriate flags instead.
*/
param->is_ror_scan= FALSE;
}
else if (param->table->s->primary_key == keynr && pk_is_clustered)
{
/* Clustered PK scan is always a ROR scan (TODO: same as above) */
param->is_ror_scan= TRUE;
}
else if (param->range_count > 1)
{
/*
Scaning multiple key values in the index: the records are ROR
for each value, but not between values. E.g, "SELECT ... x IN
(1,3)" returns ROR order for all records with x=1, then ROR
order for records with x=3
*/
param->is_ror_scan= FALSE;
}
DBUG_PRINT("exit", ("Records: %lu", (ulong) rows));
DBUG_RETURN(rows); //psergey-merge:todo: maintain first_null_comp.
}
/*
Check if key scan on given index with equality conditions on first n key
parts is a ROR scan.
SYNOPSIS
is_key_scan_ror()
param Parameter from test_quick_select
keynr Number of key in the table. The key must not be a clustered
primary key.
nparts Number of first key parts for which equality conditions
are present.
NOTES
ROR (Rowid Ordered Retrieval) key scan is a key scan that produces
ordered sequence of rowids (ha_xxx::cmp_ref is the comparison function)
This function is needed to handle a practically-important special case:
an index scan is a ROR scan if it is done using a condition in form
"key1_1=c_1 AND ... AND key1_n=c_n"
where the index is defined on (key1_1, ..., key1_N [,a_1, ..., a_n])
and the table has a clustered Primary Key defined as
PRIMARY KEY(a_1, ..., a_n, b1, ..., b_k)
i.e. the first key parts of it are identical to uncovered parts ot the
key being scanned. This function assumes that the index flags do not
include HA_KEY_SCAN_NOT_ROR flag (that is checked elsewhere).
Check (1) is made in quick_range_seq_next()
RETURN
TRUE The scan is ROR-scan
FALSE Otherwise
*/
static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts)
{
KEY *table_key= param->table->key_info + keynr;
KEY_PART_INFO *key_part= table_key->key_part + nparts;
KEY_PART_INFO *key_part_end= (table_key->key_part +
table_key->user_defined_key_parts);
uint pk_number;
for (KEY_PART_INFO *kp= table_key->key_part; kp < key_part; kp++)
{
uint16 fieldnr= param->table->key_info[keynr].
key_part[kp - table_key->key_part].fieldnr - 1;
if (param->table->field[fieldnr]->key_length() != kp->length)
return FALSE;
}
/*
If there are equalities for all key parts, it is a ROR scan. If there are
equalities all keyparts and even some of key parts from "Extended Key"
index suffix, it is a ROR-scan, too.
*/
if (key_part >= key_part_end)
return TRUE;
key_part= table_key->key_part + nparts;
pk_number= param->table->s->primary_key;
if (!param->table->file->primary_key_is_clustered() || pk_number == MAX_KEY)
return FALSE;
KEY_PART_INFO *pk_part= param->table->key_info[pk_number].key_part;
KEY_PART_INFO *pk_part_end= pk_part +
param->table->key_info[pk_number].user_defined_key_parts;
for (;(key_part!=key_part_end) && (pk_part != pk_part_end);
++key_part, ++pk_part)
{
if ((key_part->field != pk_part->field) ||
(key_part->length != pk_part->length))
return FALSE;
}
return (key_part == key_part_end);
}
/*
Create a QUICK_RANGE_SELECT from given key and SEL_ARG tree for that key.
SYNOPSIS
get_quick_select()
param
idx Index of used key in param->key.
key_tree SEL_ARG tree for the used key
mrr_flags MRR parameter for quick select
mrr_buf_size MRR parameter for quick select
parent_alloc If not NULL, use it to allocate memory for
quick select data. Otherwise use quick->alloc.
NOTES
The caller must call QUICK_SELECT::init for returned quick select.
CAUTION! This function may change thd->mem_root to a MEM_ROOT which will be
deallocated when the returned quick select is deleted.
RETURN
NULL on error
otherwise created quick select
*/
QUICK_RANGE_SELECT *
get_quick_select(PARAM *param,uint idx,SEL_ARG *key_tree, uint mrr_flags,
uint mrr_buf_size, MEM_ROOT *parent_alloc)
{
QUICK_RANGE_SELECT *quick;
bool create_err= FALSE;
DBUG_ENTER("get_quick_select");
if (param->table->key_info[param->real_keynr[idx]].flags & HA_SPATIAL)
quick=new QUICK_RANGE_SELECT_GEOM(param->thd, param->table,
param->real_keynr[idx],
MY_TEST(parent_alloc),
parent_alloc, &create_err);
else
quick=new QUICK_RANGE_SELECT(param->thd, param->table,
param->real_keynr[idx],
MY_TEST(parent_alloc), NULL, &create_err);
if (quick)
{
if (create_err ||
get_quick_keys(param,quick,param->key[idx],key_tree,param->min_key,0,
param->max_key,0))
{
delete quick;
quick=0;
}
else
{
KEY *keyinfo= param->table->key_info+param->real_keynr[idx];
quick->mrr_flags= mrr_flags;
quick->mrr_buf_size= mrr_buf_size;
quick->key_parts=(KEY_PART*)
memdup_root(parent_alloc? parent_alloc : &quick->alloc,
(char*) param->key[idx],
sizeof(KEY_PART)*
param->table->actual_n_key_parts(keyinfo));
}
}
DBUG_RETURN(quick);
}
/*
** Fix this to get all possible sub_ranges
*/
bool
get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
SEL_ARG *key_tree, uchar *min_key,uint min_key_flag,
uchar *max_key, uint max_key_flag)
{
QUICK_RANGE *range;
uint flag;
int min_part= key_tree->part-1, // # of keypart values in min_key buffer
max_part= key_tree->part-1; // # of keypart values in max_key buffer
if (key_tree->left != &null_element)
{
if (get_quick_keys(param,quick,key,key_tree->left,
min_key,min_key_flag, max_key, max_key_flag))
return 1;
}
uchar *tmp_min_key=min_key,*tmp_max_key=max_key;
min_part+= key_tree->store_min(key[key_tree->part].store_length,
&tmp_min_key,min_key_flag);
max_part+= key_tree->store_max(key[key_tree->part].store_length,
&tmp_max_key,max_key_flag);
if (key_tree->next_key_part &&
key_tree->next_key_part->type == SEL_ARG::KEY_RANGE &&
key_tree->next_key_part->part == key_tree->part+1)
{ // const key as prefix
if ((tmp_min_key - min_key) == (tmp_max_key - max_key) &&
memcmp(min_key, max_key, (uint)(tmp_max_key - max_key))==0 &&
key_tree->min_flag==0 && key_tree->max_flag==0)
{
if (get_quick_keys(param,quick,key,key_tree->next_key_part,
tmp_min_key, min_key_flag | key_tree->min_flag,
tmp_max_key, max_key_flag | key_tree->max_flag))
return 1;
goto end; // Ugly, but efficient
}
{
uint tmp_min_flag=key_tree->min_flag,tmp_max_flag=key_tree->max_flag;
if (!tmp_min_flag)
min_part+= key_tree->next_key_part->store_min_key(key,
&tmp_min_key,
&tmp_min_flag,
MAX_KEY);
if (!tmp_max_flag)
max_part+= key_tree->next_key_part->store_max_key(key,
&tmp_max_key,
&tmp_max_flag,
MAX_KEY);
flag=tmp_min_flag | tmp_max_flag;
}
}
else
{
flag = (key_tree->min_flag & GEOM_FLAG) ?
key_tree->min_flag : key_tree->min_flag | key_tree->max_flag;
}
/*
Ensure that some part of min_key and max_key are used. If not,
regard this as no lower/upper range
*/
if ((flag & GEOM_FLAG) == 0)
{
if (tmp_min_key != param->min_key)
flag&= ~NO_MIN_RANGE;
else
flag|= NO_MIN_RANGE;
if (tmp_max_key != param->max_key)
flag&= ~NO_MAX_RANGE;
else
flag|= NO_MAX_RANGE;
}
if (flag == 0)
{
uint length= (uint) (tmp_min_key - param->min_key);
if (length == (uint) (tmp_max_key - param->max_key) &&
!memcmp(param->min_key,param->max_key,length))
{
KEY *table_key=quick->head->key_info+quick->index;
flag=EQ_RANGE;
if ((table_key->flags & HA_NOSAME) &&
key_tree->part == table_key->user_defined_key_parts-1)
{
if ((table_key->flags & HA_NULL_PART_KEY) &&
null_part_in_key(key,
param->min_key,
(uint) (tmp_min_key - param->min_key)))
flag|= NULL_RANGE;
else
flag|= UNIQUE_RANGE;
}
}
}
/* Get range for retrieving rows in QUICK_SELECT::get_next */
if (!(range= new QUICK_RANGE(param->min_key,
(uint) (tmp_min_key - param->min_key),
min_part >=0 ? make_keypart_map(min_part) : 0,
param->max_key,
(uint) (tmp_max_key - param->max_key),
max_part >=0 ? make_keypart_map(max_part) : 0,
flag)))
return 1; // out of memory
set_if_bigger(quick->max_used_key_length, range->min_length);
set_if_bigger(quick->max_used_key_length, range->max_length);
set_if_bigger(quick->used_key_parts, (uint) key_tree->part+1);
if (insert_dynamic(&quick->ranges, (uchar*) &range))
return 1;
end:
if (key_tree->right != &null_element)
return get_quick_keys(param,quick,key,key_tree->right,
min_key,min_key_flag,
max_key,max_key_flag);
return 0;
}
/*
Return 1 if there is only one range and this uses the whole unique key
*/
bool QUICK_RANGE_SELECT::unique_key_range()
{
if (ranges.elements == 1)
{
QUICK_RANGE *tmp= *((QUICK_RANGE**)ranges.buffer);
if ((tmp->flag & (EQ_RANGE | NULL_RANGE)) == EQ_RANGE)
{
KEY *key=head->key_info+index;
return (key->flags & HA_NOSAME) && key->key_length == tmp->min_length;
}
}
return 0;
}
/*
Return TRUE if any part of the key is NULL
SYNOPSIS
null_part_in_key()
key_part Array of key parts (index description)
key Key values tuple
length Length of key values tuple in bytes.
RETURN
TRUE The tuple has at least one "keypartX is NULL"
FALSE Otherwise
*/
static bool null_part_in_key(KEY_PART *key_part, const uchar *key, uint length)
{
for (const uchar *end=key+length ;
key < end;
key+= key_part++->store_length)
{
if (key_part->null_bit && *key)
return 1;
}
return 0;
}
bool QUICK_SELECT_I::is_keys_used(const MY_BITMAP *fields)
{
return is_key_used(head, index, fields);
}
bool QUICK_INDEX_SORT_SELECT::is_keys_used(const MY_BITMAP *fields)
{
QUICK_RANGE_SELECT *quick;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
while ((quick= it++))
{
if (is_key_used(head, quick->index, fields))
return 1;
}
return 0;
}
bool QUICK_ROR_INTERSECT_SELECT::is_keys_used(const MY_BITMAP *fields)
{
QUICK_SELECT_WITH_RECORD *qr;
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
while ((qr= it++))
{
if (is_key_used(head, qr->quick->index, fields))
return 1;
}
return 0;
}
bool QUICK_ROR_UNION_SELECT::is_keys_used(const MY_BITMAP *fields)
{
QUICK_SELECT_I *quick;
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
if (quick->is_keys_used(fields))
return 1;
}
return 0;
}
FT_SELECT *get_ft_select(THD *thd, TABLE *table, uint key)
{
bool create_err= FALSE;
FT_SELECT *fts= new FT_SELECT(thd, table, key, &create_err);
if (create_err)
{
delete fts;
return NULL;
}
else
return fts;
}
/*
Create quick select from ref/ref_or_null scan.
SYNOPSIS
get_quick_select_for_ref()
thd Thread handle
table Table to access
ref ref[_or_null] scan parameters
records Estimate of number of records (needed only to construct
quick select)
NOTES
This allocates things in a new memory root, as this may be called many
times during a query.
RETURN
Quick select that retrieves the same rows as passed ref scan
NULL on error.
*/
QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
TABLE_REF *ref, ha_rows records)
{
MEM_ROOT *old_root, *alloc;
QUICK_RANGE_SELECT *quick;
KEY *key_info = &table->key_info[ref->key];
KEY_PART *key_part;
QUICK_RANGE *range;
uint part;
bool create_err= FALSE;
Cost_estimate cost;
uint max_used_key_len;
old_root= thd->mem_root;
/* The following call may change thd->mem_root */
quick= new QUICK_RANGE_SELECT(thd, table, ref->key, 0, 0, &create_err);
/* save mem_root set by QUICK_RANGE_SELECT constructor */
alloc= thd->mem_root;
/*
return back default mem_root (thd->mem_root) changed by
QUICK_RANGE_SELECT constructor
*/
thd->mem_root= old_root;
if (!quick || create_err)
return 0; /* no ranges found */
if (quick->init())
goto err;
quick->records= records;
if ((cp_buffer_from_ref(thd, table, ref) && thd->is_fatal_error) ||
!(range= new(alloc) QUICK_RANGE()))
goto err; // out of memory
range->min_key= range->max_key= ref->key_buff;
range->min_length= range->max_length= ref->key_length;
range->min_keypart_map= range->max_keypart_map=
make_prev_keypart_map(ref->key_parts);
range->flag= EQ_RANGE;
if (!(quick->key_parts=key_part=(KEY_PART *)
alloc_root(&quick->alloc,sizeof(KEY_PART)*ref->key_parts)))
goto err;
max_used_key_len=0;
for (part=0 ; part < ref->key_parts ;part++,key_part++)
{
key_part->part=part;
key_part->field= key_info->key_part[part].field;
key_part->length= key_info->key_part[part].length;
key_part->store_length= key_info->key_part[part].store_length;
key_part->null_bit= key_info->key_part[part].null_bit;
key_part->flag= (uint8) key_info->key_part[part].key_part_flag;
max_used_key_len +=key_info->key_part[part].store_length;
}
quick->max_used_key_length= max_used_key_len;
if (insert_dynamic(&quick->ranges,(uchar*)&range))
goto err;
/*
Add a NULL range if REF_OR_NULL optimization is used.
For example:
if we have "WHERE A=2 OR A IS NULL" we created the (A=2) range above
and have ref->null_ref_key set. Will create a new NULL range here.
*/
if (ref->null_ref_key)
{
QUICK_RANGE *null_range;
*ref->null_ref_key= 1; // Set null byte then create a range
if (!(null_range= new (alloc)
QUICK_RANGE(ref->key_buff, ref->key_length,
make_prev_keypart_map(ref->key_parts),
ref->key_buff, ref->key_length,
make_prev_keypart_map(ref->key_parts), EQ_RANGE)))
goto err;
*ref->null_ref_key= 0; // Clear null byte
if (insert_dynamic(&quick->ranges,(uchar*)&null_range))
goto err;
}
/* Call multi_range_read_info() to get the MRR flags and buffer size */
quick->mrr_flags= HA_MRR_NO_ASSOCIATION |
(table->key_read ? HA_MRR_INDEX_ONLY : 0);
if (thd->lex->sql_command != SQLCOM_SELECT)
quick->mrr_flags |= HA_MRR_USE_DEFAULT_IMPL;
quick->mrr_buf_size= thd->variables.mrr_buff_size;
if (table->file->multi_range_read_info(quick->index, 1, (uint)records,
~0,
&quick->mrr_buf_size,
&quick->mrr_flags, &cost))
goto err;
return quick;
err:
delete quick;
return 0;
}
/*
Perform key scans for all used indexes (except CPK), get rowids and merge
them into an ordered non-recurrent sequence of rowids.
The merge/duplicate removal is performed using Unique class. We put all
rowids into Unique, get the sorted sequence and destroy the Unique.
If table has a clustered primary key that covers all rows (TRUE for bdb
and innodb currently) and one of the index_merge scans is a scan on PK,
then rows that will be retrieved by PK scan are not put into Unique and
primary key scan is not performed here, it is performed later separately.
RETURN
0 OK
other error
*/
int read_keys_and_merge_scans(THD *thd,
TABLE *head,
List<QUICK_RANGE_SELECT> quick_selects,
QUICK_RANGE_SELECT *pk_quick_select,
READ_RECORD *read_record,
bool intersection,
key_map *filtered_scans,
Unique **unique_ptr)
{
List_iterator_fast<QUICK_RANGE_SELECT> cur_quick_it(quick_selects);
QUICK_RANGE_SELECT* cur_quick;
int result;
Unique *unique= *unique_ptr;
handler *file= head->file;
bool with_cpk_filter= pk_quick_select != NULL;
bool enabled_keyread= 0;
DBUG_ENTER("read_keys_and_merge");
/* We're going to just read rowids. */
if (!head->key_read)
{
enabled_keyread= 1;
head->enable_keyread();
}
head->prepare_for_position();
cur_quick_it.rewind();
cur_quick= cur_quick_it++;
bool first_quick= TRUE;
DBUG_ASSERT(cur_quick != 0);
/*
We reuse the same instance of handler so we need to call both init and
reset here.
*/
if (cur_quick->init() || cur_quick->reset())
goto err;
if (unique == NULL)
{
DBUG_EXECUTE_IF("index_merge_may_not_create_a_Unique", DBUG_ABORT(); );
DBUG_EXECUTE_IF("only_one_Unique_may_be_created",
DBUG_SET("+d,index_merge_may_not_create_a_Unique"); );
unique= new Unique(refpos_order_cmp, (void *)file,
file->ref_length,
thd->variables.sortbuff_size,
intersection ? quick_selects.elements : 0);
if (!unique)
goto err;
*unique_ptr= unique;
}
else
{
unique->reset();
filesort_free_buffers(head, false);
}
DBUG_ASSERT(file->ref_length == unique->get_size());
DBUG_ASSERT(thd->variables.sortbuff_size == unique->get_max_in_memory_size());
for (;;)
{
while ((result= cur_quick->get_next()) == HA_ERR_END_OF_FILE)
{
if (intersection)
with_cpk_filter= filtered_scans->is_set(cur_quick->index);
if (first_quick)
{
first_quick= FALSE;
if (intersection && unique->is_in_memory())
unique->close_for_expansion();
}
cur_quick->range_end();
cur_quick= cur_quick_it++;
if (!cur_quick)
break;
if (cur_quick->file->inited != handler::NONE)
cur_quick->file->ha_index_end();
if (cur_quick->init() || cur_quick->reset())
goto err;
}
if (result)
{
if (result != HA_ERR_END_OF_FILE)
{
cur_quick->range_end();
goto err;
}
break;
}
if (thd->killed)
goto err;
if (with_cpk_filter &&
pk_quick_select->row_in_ranges() != intersection )
continue;
cur_quick->file->position(cur_quick->record);
if (unique->unique_add((char*)cur_quick->file->ref))
goto err;
}
/*
Ok all rowids are in the Unique now. The next call will initialize
head->sort structure so it can be used to iterate through the rowids
sequence.
*/
result= unique->get(head);
/*
index merge currently doesn't support "using index" at all
*/
if (enabled_keyread)
head->disable_keyread();
if (init_read_record(read_record, thd, head, (SQL_SELECT*) 0, 1 , 1, TRUE))
result= 1;
DBUG_RETURN(result);
err:
if (enabled_keyread)
head->disable_keyread();
DBUG_RETURN(1);
}
int QUICK_INDEX_MERGE_SELECT::read_keys_and_merge()
{
int result;
DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::read_keys_and_merge");
result= read_keys_and_merge_scans(thd, head, quick_selects, pk_quick_select,
&read_record, FALSE, NULL, &unique);
doing_pk_scan= FALSE;
DBUG_RETURN(result);
}
/*
Get next row for index_merge.
NOTES
The rows are read from
1. rowids stored in Unique.
2. QUICK_RANGE_SELECT with clustered primary key (if any).
The sets of rows retrieved in 1) and 2) are guaranteed to be disjoint.
*/
int QUICK_INDEX_MERGE_SELECT::get_next()
{
int result;
DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::get_next");
if (doing_pk_scan)
DBUG_RETURN(pk_quick_select->get_next());
if ((result= read_record.read_record(&read_record)) == -1)
{
result= HA_ERR_END_OF_FILE;
end_read_record(&read_record);
free_io_cache(head);
/* All rows from Unique have been retrieved, do a clustered PK scan */
if (pk_quick_select)
{
doing_pk_scan= TRUE;
if ((result= pk_quick_select->init()) ||
(result= pk_quick_select->reset()))
DBUG_RETURN(result);
DBUG_RETURN(pk_quick_select->get_next());
}
}
DBUG_RETURN(result);
}
int QUICK_INDEX_INTERSECT_SELECT::read_keys_and_merge()
{
int result;
DBUG_ENTER("QUICK_INDEX_INTERSECT_SELECT::read_keys_and_merge");
result= read_keys_and_merge_scans(thd, head, quick_selects, pk_quick_select,
&read_record, TRUE, &filtered_scans,
&unique);
DBUG_RETURN(result);
}
int QUICK_INDEX_INTERSECT_SELECT::get_next()
{
int result;
DBUG_ENTER("QUICK_INDEX_INTERSECT_SELECT::get_next");
if ((result= read_record.read_record(&read_record)) == -1)
{
result= HA_ERR_END_OF_FILE;
end_read_record(&read_record);
free_io_cache(head);
}
DBUG_RETURN(result);
}
/*
Retrieve next record.
SYNOPSIS
QUICK_ROR_INTERSECT_SELECT::get_next()
NOTES
Invariant on enter/exit: all intersected selects have retrieved all index
records with rowid <= some_rowid_val and no intersected select has
retrieved any index records with rowid > some_rowid_val.
We start fresh and loop until we have retrieved the same rowid in each of
the key scans or we got an error.
If a Clustered PK scan is present, it is used only to check if row
satisfies its condition (and never used for row retrieval).
Locking: to ensure that exclusive locks are only set on records that
are included in the final result we must release the lock
on all rows we read but do not include in the final result. This
must be done on each index that reads the record and the lock
must be released using the same handler (the same quick object) as
used when reading the record.
RETURN
0 - Ok
other - Error code if any error occurred.
*/
int QUICK_ROR_INTERSECT_SELECT::get_next()
{
List_iterator_fast<QUICK_SELECT_WITH_RECORD> quick_it(quick_selects);
QUICK_SELECT_WITH_RECORD *qr;
QUICK_RANGE_SELECT* quick;
/* quick that reads the given rowid first. This is needed in order
to be able to unlock the row using the same handler object that locked
it */
QUICK_RANGE_SELECT* quick_with_last_rowid;
int error, cmp;
uint last_rowid_count=0;
DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::get_next");
do
{
/* Get a rowid for first quick and save it as a 'candidate' */
qr= quick_it++;
quick= qr->quick;
error= quick->get_next();
if (cpk_quick)
{
while (!error && !cpk_quick->row_in_ranges())
{
quick->file->unlock_row(); /* row not in range; unlock */
error= quick->get_next();
}
}
if (error)
DBUG_RETURN(error);
/* Save the read key tuple */
key_copy(qr->key_tuple, record, head->key_info + quick->index,
quick->max_used_key_length);
quick->file->position(quick->record);
memcpy(last_rowid, quick->file->ref, head->file->ref_length);
last_rowid_count= 1;
quick_with_last_rowid= quick;
while (last_rowid_count < quick_selects.elements)
{
if (!(qr= quick_it++))
{
quick_it.rewind();
qr= quick_it++;
}
quick= qr->quick;
do
{
DBUG_EXECUTE_IF("innodb_quick_report_deadlock",
DBUG_SET("+d,innodb_report_deadlock"););
if ((error= quick->get_next()))
{
/* On certain errors like deadlock, trx might be rolled back.*/
if (!current_thd->transaction_rollback_request)
quick_with_last_rowid->file->unlock_row();
DBUG_RETURN(error);
}
quick->file->position(quick->record);
cmp= head->file->cmp_ref(quick->file->ref, last_rowid);
if (cmp < 0)
{
/* This row is being skipped. Release lock on it. */
quick->file->unlock_row();
}
} while (cmp < 0);
key_copy(qr->key_tuple, record, head->key_info + quick->index,
quick->max_used_key_length);
/* Ok, current select 'caught up' and returned ref >= cur_ref */
if (cmp > 0)
{
/* Found a row with ref > cur_ref. Make it a new 'candidate' */
if (cpk_quick)
{
while (!cpk_quick->row_in_ranges())
{
quick->file->unlock_row(); /* row not in range; unlock */
if ((error= quick->get_next()))
{
/* On certain errors like deadlock, trx might be rolled back.*/
if (!current_thd->transaction_rollback_request)
quick_with_last_rowid->file->unlock_row();
DBUG_RETURN(error);
}
}
quick->file->position(quick->record);
}
memcpy(last_rowid, quick->file->ref, head->file->ref_length);
quick_with_last_rowid->file->unlock_row();
last_rowid_count= 1;
quick_with_last_rowid= quick;
//save the fields here
key_copy(qr->key_tuple, record, head->key_info + quick->index,
quick->max_used_key_length);
}
else
{
/* current 'candidate' row confirmed by this select */
last_rowid_count++;
}
}
/* We get here if we got the same row ref in all scans. */
if (need_to_fetch_row)
error= head->file->ha_rnd_pos(head->record[0], last_rowid);
} while (error == HA_ERR_RECORD_DELETED);
if (!need_to_fetch_row)
{
/* Restore the columns we've read/saved with other quick selects */
quick_it.rewind();
while ((qr= quick_it++))
{
if (qr->quick != quick)
{
key_restore(record, qr->key_tuple, head->key_info + qr->quick->index,
qr->quick->max_used_key_length);
}
}
}
DBUG_RETURN(error);
}
/*
Retrieve next record.
SYNOPSIS
QUICK_ROR_UNION_SELECT::get_next()
NOTES
Enter/exit invariant:
For each quick select in the queue a {key,rowid} tuple has been
retrieved but the corresponding row hasn't been passed to output.
RETURN
0 - Ok
other - Error code if any error occurred.
*/
int QUICK_ROR_UNION_SELECT::get_next()
{
int error, dup_row;
QUICK_SELECT_I *quick;
uchar *tmp;
DBUG_ENTER("QUICK_ROR_UNION_SELECT::get_next");
do
{
do
{
if (!queue.elements)
DBUG_RETURN(HA_ERR_END_OF_FILE);
/* Ok, we have a queue with >= 1 scans */
quick= (QUICK_SELECT_I*)queue_top(&queue);
memcpy(cur_rowid, quick->last_rowid, rowid_length);
/* put into queue rowid from the same stream as top element */
if ((error= quick->get_next()))
{
if (error != HA_ERR_END_OF_FILE)
DBUG_RETURN(error);
queue_remove_top(&queue);
}
else
{
quick->save_last_pos();
queue_replace_top(&queue);
}
if (!have_prev_rowid)
{
/* No rows have been returned yet */
dup_row= FALSE;
have_prev_rowid= TRUE;
}
else
dup_row= !head->file->cmp_ref(cur_rowid, prev_rowid);
} while (dup_row);
tmp= cur_rowid;
cur_rowid= prev_rowid;
prev_rowid= tmp;
error= head->file->ha_rnd_pos(quick->record, prev_rowid);
} while (error == HA_ERR_RECORD_DELETED);
DBUG_RETURN(error);
}
int QUICK_RANGE_SELECT::reset()
{
uint buf_size;
uchar *mrange_buff;
int error;
HANDLER_BUFFER empty_buf;
MY_BITMAP * const save_read_set= head->read_set;
MY_BITMAP * const save_write_set= head->write_set;
DBUG_ENTER("QUICK_RANGE_SELECT::reset");
last_range= NULL;
cur_range= (QUICK_RANGE**) ranges.buffer;
RANGE_SEQ_IF seq_funcs= {NULL, quick_range_seq_init, quick_range_seq_next, 0, 0};
if (file->inited == handler::RND)
{
/* Handler could be left in this state by MRR */
if ((error= file->ha_rnd_end()))
DBUG_RETURN(error);
}
if (in_ror_merged_scan)
head->column_bitmaps_set_no_signal(&column_bitmap, &column_bitmap);
if (file->inited == handler::NONE)
{
DBUG_EXECUTE_IF("bug14365043_2",
DBUG_SET("+d,ha_index_init_fail"););
if ((error= file->ha_index_init(index,1)))
{
file->print_error(error, MYF(0));
goto err;
}
}
/* Allocate buffer if we need one but haven't allocated it yet */
if (mrr_buf_size && !mrr_buf_desc)
{
buf_size= mrr_buf_size;
while (buf_size && !my_multi_malloc(MYF(MY_WME),
&mrr_buf_desc, sizeof(*mrr_buf_desc),
&mrange_buff, buf_size,
NullS))
{
/* Try to shrink the buffers until both are 0. */
buf_size/= 2;
}
if (!mrr_buf_desc)
DBUG_RETURN(HA_ERR_OUT_OF_MEM);
/* Initialize the handler buffer. */
mrr_buf_desc->buffer= mrange_buff;
mrr_buf_desc->buffer_end= mrange_buff + buf_size;
mrr_buf_desc->end_of_used_area= mrange_buff;
#ifdef HAVE_valgrind
/*
We need this until ndb will use the buffer efficiently
(Now ndb stores complete row in here, instead of only the used fields
which gives us valgrind warnings in compare_record[])
*/
bzero((char*) mrange_buff, buf_size);
#endif
}
if (!mrr_buf_desc)
empty_buf.buffer= empty_buf.buffer_end= empty_buf.end_of_used_area= NULL;
error= file->multi_range_read_init(&seq_funcs, (void*)this, ranges.elements,
mrr_flags, mrr_buf_desc? mrr_buf_desc:
&empty_buf);
err:
/* Restore bitmaps set on entry */
if (in_ror_merged_scan)
head->column_bitmaps_set_no_signal(save_read_set, save_write_set);
DBUG_RETURN(error);
}
/*
Get next possible record using quick-struct.
SYNOPSIS
QUICK_RANGE_SELECT::get_next()
NOTES
Record is read into table->record[0]
RETURN
0 Found row
HA_ERR_END_OF_FILE No (more) rows in range
# Error code
*/
int QUICK_RANGE_SELECT::get_next()
{
range_id_t dummy;
MY_BITMAP * const save_read_set= head->read_set;
MY_BITMAP * const save_write_set= head->write_set;
DBUG_ENTER("QUICK_RANGE_SELECT::get_next");
if (in_ror_merged_scan)
{
/*
We don't need to signal the bitmap change as the bitmap is always the
same for this head->file
*/
head->column_bitmaps_set_no_signal(&column_bitmap, &column_bitmap);
}
int result= file->multi_range_read_next(&dummy);
if (in_ror_merged_scan)
{
/* Restore bitmaps set on entry */
head->column_bitmaps_set_no_signal(save_read_set, save_write_set);
}
DBUG_RETURN(result);
}
/*
Get the next record with a different prefix.
@param prefix_length length of cur_prefix
@param group_key_parts The number of key parts in the group prefix
@param cur_prefix prefix of a key to be searched for
Each subsequent call to the method retrieves the first record that has a
prefix with length prefix_length and which is different from cur_prefix,
such that the record with the new prefix is within the ranges described by
this->ranges. The record found is stored into the buffer pointed by
this->record. The method is useful for GROUP-BY queries with range
conditions to discover the prefix of the next group that satisfies the range
conditions.
@todo
This method is a modified copy of QUICK_RANGE_SELECT::get_next(), so both
methods should be unified into a more general one to reduce code
duplication.
@retval 0 on success
@retval HA_ERR_END_OF_FILE if returned all keys
@retval other if some error occurred
*/
int QUICK_RANGE_SELECT::get_next_prefix(uint prefix_length,
uint group_key_parts,
uchar *cur_prefix)
{
DBUG_ENTER("QUICK_RANGE_SELECT::get_next_prefix");
const key_part_map keypart_map= make_prev_keypart_map(group_key_parts);
for (;;)
{
int result;
if (last_range)
{
/* Read the next record in the same range with prefix after cur_prefix. */
DBUG_ASSERT(cur_prefix != NULL);
result= file->ha_index_read_map(record, cur_prefix, keypart_map,
HA_READ_AFTER_KEY);
if (result || last_range->max_keypart_map == 0)
DBUG_RETURN(result);
key_range previous_endpoint;
last_range->make_max_endpoint(&previous_endpoint, prefix_length, keypart_map);
if (file->compare_key(&previous_endpoint) <= 0)
DBUG_RETURN(0);
}
uint count= ranges.elements - (cur_range - (QUICK_RANGE**) ranges.buffer);
if (count == 0)
{
/* Ranges have already been used up before. None is left for read. */
last_range= 0;
DBUG_RETURN(HA_ERR_END_OF_FILE);
}
last_range= *(cur_range++);
key_range start_key, end_key;
last_range->make_min_endpoint(&start_key, prefix_length, keypart_map);
last_range->make_max_endpoint(&end_key, prefix_length, keypart_map);
result= file->read_range_first(last_range->min_keypart_map ? &start_key : 0,
last_range->max_keypart_map ? &end_key : 0,
MY_TEST(last_range->flag & EQ_RANGE),
TRUE);
if (last_range->flag == (UNIQUE_RANGE | EQ_RANGE))
last_range= 0; // Stop searching
if (result != HA_ERR_END_OF_FILE)
DBUG_RETURN(result);
last_range= 0; // No matching rows; go to next range
}
}
/* Get next for geometrical indexes */
int QUICK_RANGE_SELECT_GEOM::get_next()
{
DBUG_ENTER("QUICK_RANGE_SELECT_GEOM::get_next");
for (;;)
{
int result;
if (last_range)
{
// Already read through key
result= file->ha_index_next_same(record, last_range->min_key,
last_range->min_length);
if (result != HA_ERR_END_OF_FILE)
DBUG_RETURN(result);
}
uint count= ranges.elements - (cur_range - (QUICK_RANGE**) ranges.buffer);
if (count == 0)
{
/* Ranges have already been used up before. None is left for read. */
last_range= 0;
DBUG_RETURN(HA_ERR_END_OF_FILE);
}
last_range= *(cur_range++);
result= file->ha_index_read_map(record, last_range->min_key,
last_range->min_keypart_map,
(ha_rkey_function)(last_range->flag ^
GEOM_FLAG));
if (result != HA_ERR_KEY_NOT_FOUND && result != HA_ERR_END_OF_FILE)
DBUG_RETURN(result);
last_range= 0; // Not found, to next range
}
}
/*
Check if current row will be retrieved by this QUICK_RANGE_SELECT
NOTES
It is assumed that currently a scan is being done on another index
which reads all necessary parts of the index that is scanned by this
quick select.
The implementation does a binary search on sorted array of disjoint
ranges, without taking size of range into account.
This function is used to filter out clustered PK scan rows in
index_merge quick select.
RETURN
TRUE if current row will be retrieved by this quick select
FALSE if not
*/
bool QUICK_RANGE_SELECT::row_in_ranges()
{
QUICK_RANGE *res;
uint min= 0;
uint max= ranges.elements - 1;
uint mid= (max + min)/2;
while (min != max)
{
if (cmp_next(*(QUICK_RANGE**)dynamic_array_ptr(&ranges, mid)))
{
/* current row value > mid->max */
min= mid + 1;
}
else
max= mid;
mid= (min + max) / 2;
}
res= *(QUICK_RANGE**)dynamic_array_ptr(&ranges, mid);
return (!cmp_next(res) && !cmp_prev(res));
}
/*
This is a hack: we inherit from QUICK_RANGE_SELECT so that we can use the
get_next() interface, but we have to hold a pointer to the original
QUICK_RANGE_SELECT because its data are used all over the place. What
should be done is to factor out the data that is needed into a base
class (QUICK_SELECT), and then have two subclasses (_ASC and _DESC)
which handle the ranges and implement the get_next() function. But
for now, this seems to work right at least.
*/
QUICK_SELECT_DESC::QUICK_SELECT_DESC(QUICK_RANGE_SELECT *q,
uint used_key_parts_arg)
:QUICK_RANGE_SELECT(*q), rev_it(rev_ranges),
used_key_parts (used_key_parts_arg)
{
QUICK_RANGE *r;
/*
Use default MRR implementation for reverse scans. No table engine
currently can do an MRR scan with output in reverse index order.
*/
mrr_buf_desc= NULL;
mrr_flags |= HA_MRR_USE_DEFAULT_IMPL;
mrr_buf_size= 0;
QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
QUICK_RANGE **end_range= pr + ranges.elements;
for (; pr!=end_range; pr++)
rev_ranges.push_front(*pr);
/* Remove EQ_RANGE flag for keys that are not using the full key */
for (r = rev_it++; r; r = rev_it++)
{
if ((r->flag & EQ_RANGE) &&
head->key_info[index].key_length != r->max_length)
r->flag&= ~EQ_RANGE;
}
rev_it.rewind();
q->dont_free=1; // Don't free shared mem
}
int QUICK_SELECT_DESC::get_next()
{
DBUG_ENTER("QUICK_SELECT_DESC::get_next");
/* The max key is handled as follows:
* - if there is NO_MAX_RANGE, start at the end and move backwards
* - if it is an EQ_RANGE, which means that max key covers the entire
* key, go directly to the key and read through it (sorting backwards is
* same as sorting forwards)
* - if it is NEAR_MAX, go to the key or next, step back once, and
* move backwards
* - otherwise (not NEAR_MAX == include the key), go after the key,
* step back once, and move backwards
*/
for (;;)
{
int result;
if (last_range)
{ // Already read through key
result = ((last_range->flag & EQ_RANGE &&
used_key_parts <= head->key_info[index].user_defined_key_parts) ?
file->ha_index_next_same(record, last_range->min_key,
last_range->min_length) :
file->ha_index_prev(record));
if (!result)
{
if (cmp_prev(*rev_it.ref()) == 0)
DBUG_RETURN(0);
}
else if (result != HA_ERR_END_OF_FILE)
DBUG_RETURN(result);
}
if (!(last_range= rev_it++))
DBUG_RETURN(HA_ERR_END_OF_FILE); // All ranges used
key_range start_key;
start_key.key= (const uchar*) last_range->min_key;
start_key.length= last_range->min_length;
start_key.flag= ((last_range->flag & NEAR_MIN) ? HA_READ_AFTER_KEY :
(last_range->flag & EQ_RANGE) ?
HA_READ_KEY_EXACT : HA_READ_KEY_OR_NEXT);
start_key.keypart_map= last_range->min_keypart_map;
key_range end_key;
end_key.key= (const uchar*) last_range->max_key;
end_key.length= last_range->max_length;
end_key.flag= (last_range->flag & NEAR_MAX ? HA_READ_BEFORE_KEY :
HA_READ_AFTER_KEY);
end_key.keypart_map= last_range->max_keypart_map;
result= file->prepare_range_scan((last_range->flag & NO_MIN_RANGE) ? NULL : &start_key,
(last_range->flag & NO_MAX_RANGE) ? NULL : &end_key);
if (result)
{
DBUG_RETURN(result);
}
if (last_range->flag & NO_MAX_RANGE) // Read last record
{
int local_error;
if ((local_error= file->ha_index_last(record)))
DBUG_RETURN(local_error); // Empty table
if (cmp_prev(last_range) == 0)
DBUG_RETURN(0);
last_range= 0; // No match; go to next range
continue;
}
if (last_range->flag & EQ_RANGE &&
used_key_parts <= head->key_info[index].user_defined_key_parts)
{
result= file->ha_index_read_map(record, last_range->max_key,
last_range->max_keypart_map,
HA_READ_KEY_EXACT);
}
else
{
DBUG_ASSERT(last_range->flag & NEAR_MAX ||
(last_range->flag & EQ_RANGE &&
used_key_parts > head->key_info[index].user_defined_key_parts) ||
range_reads_after_key(last_range));
result= file->ha_index_read_map(record, last_range->max_key,
last_range->max_keypart_map,
((last_range->flag & NEAR_MAX) ?
HA_READ_BEFORE_KEY :
HA_READ_PREFIX_LAST_OR_PREV));
}
if (result)
{
if (result != HA_ERR_KEY_NOT_FOUND && result != HA_ERR_END_OF_FILE)
DBUG_RETURN(result);
last_range= 0; // Not found, to next range
continue;
}
if (cmp_prev(last_range) == 0)
{
if (last_range->flag == (UNIQUE_RANGE | EQ_RANGE))
last_range= 0; // Stop searching
DBUG_RETURN(0); // Found key is in range
}
last_range= 0; // To next range
}
}
/**
Create a compatible quick select with the result ordered in an opposite way
@param used_key_parts_arg Number of used key parts
@retval NULL in case of errors (OOM etc)
@retval pointer to a newly created QUICK_SELECT_DESC if success
*/
QUICK_SELECT_I *QUICK_RANGE_SELECT::make_reverse(uint used_key_parts_arg)
{
QUICK_SELECT_DESC *new_quick= new QUICK_SELECT_DESC(this, used_key_parts_arg);
if (new_quick == NULL)
{
delete new_quick;
return NULL;
}
return new_quick;
}
/*
Compare if found key is over max-value
Returns 0 if key <= range->max_key
TODO: Figure out why can't this function be as simple as cmp_prev().
*/
int QUICK_RANGE_SELECT::cmp_next(QUICK_RANGE *range_arg)
{
if (range_arg->flag & NO_MAX_RANGE)
return 0; /* key can't be to large */
KEY_PART *key_part=key_parts;
uint store_length;
for (uchar *key=range_arg->max_key, *end=key+range_arg->max_length;
key < end;
key+= store_length, key_part++)
{
int cmp;
store_length= key_part->store_length;
if (key_part->null_bit)
{
if (*key)
{
if (!key_part->field->is_null())
return 1;
continue;
}
else if (key_part->field->is_null())
return 0;
key++; // Skip null byte
store_length--;
}
if ((cmp=key_part->field->key_cmp(key, key_part->length)) < 0)
return 0;
if (cmp > 0)
return 1;
}
return (range_arg->flag & NEAR_MAX) ? 1 : 0; // Exact match
}
/*
Returns 0 if found key is inside range (found key >= range->min_key).
*/
int QUICK_RANGE_SELECT::cmp_prev(QUICK_RANGE *range_arg)
{
int cmp;
if (range_arg->flag & NO_MIN_RANGE)
return 0; /* key can't be to small */
cmp= key_cmp(key_part_info, range_arg->min_key,
range_arg->min_length);
if (cmp > 0 || (cmp == 0 && !(range_arg->flag & NEAR_MIN)))
return 0;
return 1; // outside of range
}
/*
* TRUE if this range will require using HA_READ_AFTER_KEY
See comment in get_next() about this
*/
bool QUICK_SELECT_DESC::range_reads_after_key(QUICK_RANGE *range_arg)
{
return ((range_arg->flag & (NO_MAX_RANGE | NEAR_MAX)) ||
!(range_arg->flag & EQ_RANGE) ||
head->key_info[index].key_length != range_arg->max_length) ? 1 : 0;
}
void QUICK_SELECT_I::add_key_name(String *str, bool *first)
{
KEY *key_info= head->key_info + index;
if (*first)
*first= FALSE;
else
str->append(',');
str->append(key_info->name);
}
Explain_quick_select* QUICK_RANGE_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
if ((res= new (alloc) Explain_quick_select(QS_TYPE_RANGE)))
res->range.set(alloc, head->key_info[index].name, max_used_key_length);
return res;
}
Explain_quick_select* QUICK_GROUP_MIN_MAX_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
if ((res= new (alloc) Explain_quick_select(QS_TYPE_GROUP_MIN_MAX)))
res->range.set(alloc, head->key_info[index].name, max_used_key_length);
return res;
}
Explain_quick_select* QUICK_INDEX_SORT_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
if (!(res= new (alloc) Explain_quick_select(get_type())))
return NULL;
QUICK_RANGE_SELECT *quick;
Explain_quick_select *child_explain;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
while ((quick= it++))
{
if ((child_explain= quick->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
if (pk_quick_select)
{
if ((child_explain= pk_quick_select->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
return res;
}
/*
Same as QUICK_INDEX_SORT_SELECT::get_explain(), but primary key is printed
first
*/
Explain_quick_select* QUICK_INDEX_INTERSECT_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
Explain_quick_select *child_explain;
if (!(res= new (alloc) Explain_quick_select(get_type())))
return NULL;
if (pk_quick_select)
{
if ((child_explain= pk_quick_select->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
QUICK_RANGE_SELECT *quick;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
while ((quick= it++))
{
if ((child_explain= quick->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
return res;
}
Explain_quick_select* QUICK_ROR_INTERSECT_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
Explain_quick_select *child_explain;
if (!(res= new (alloc) Explain_quick_select(get_type())))
return NULL;
QUICK_SELECT_WITH_RECORD *qr;
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
while ((qr= it++))
{
if ((child_explain= qr->quick->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
if (cpk_quick)
{
if ((child_explain= cpk_quick->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
return res;
}
Explain_quick_select* QUICK_ROR_UNION_SELECT::get_explain(MEM_ROOT *alloc)
{
Explain_quick_select *res;
Explain_quick_select *child_explain;
if (!(res= new (alloc) Explain_quick_select(get_type())))
return NULL;
QUICK_SELECT_I *quick;
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
if ((child_explain= quick->get_explain(alloc)))
res->children.push_back(child_explain);
else
return NULL;
}
return res;
}
void QUICK_SELECT_I::add_key_and_length(String *key_names,
String *used_lengths,
bool *first)
{
char buf[64];
uint length;
KEY *key_info= head->key_info + index;
if (*first)
*first= FALSE;
else
{
key_names->append(',');
used_lengths->append(',');
}
key_names->append(key_info->name);
length= longlong10_to_str(max_used_key_length, buf, 10) - buf;
used_lengths->append(buf, length);
}
void QUICK_RANGE_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
bool first= TRUE;
add_key_and_length(key_names, used_lengths, &first);
}
void QUICK_INDEX_MERGE_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
QUICK_RANGE_SELECT *quick;
bool first= TRUE;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
while ((quick= it++))
{
quick->add_key_and_length(key_names, used_lengths, &first);
}
if (pk_quick_select)
pk_quick_select->add_key_and_length(key_names, used_lengths, &first);
}
void QUICK_INDEX_INTERSECT_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
QUICK_RANGE_SELECT *quick;
bool first= TRUE;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
if (pk_quick_select)
pk_quick_select->add_key_and_length(key_names, used_lengths, &first);
while ((quick= it++))
{
quick->add_key_and_length(key_names, used_lengths, &first);
}
}
void QUICK_ROR_INTERSECT_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
QUICK_SELECT_WITH_RECORD *qr;
bool first= TRUE;
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
while ((qr= it++))
{
qr->quick->add_key_and_length(key_names, used_lengths, &first);
}
if (cpk_quick)
cpk_quick->add_key_and_length(key_names, used_lengths, &first);
}
void QUICK_ROR_UNION_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
QUICK_SELECT_I *quick;
bool first= TRUE;
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
if (first)
first= FALSE;
else
{
used_lengths->append(',');
key_names->append(',');
}
quick->add_keys_and_lengths(key_names, used_lengths);
}
}
void QUICK_RANGE_SELECT::add_used_key_part_to_set(MY_BITMAP *col_set)
{
uint key_len;
KEY_PART *part= key_parts;
for (key_len=0; key_len < max_used_key_length;
key_len += (part++)->store_length)
{
bitmap_set_bit(col_set, part->field->field_index);
}
}
void QUICK_GROUP_MIN_MAX_SELECT::add_used_key_part_to_set(MY_BITMAP *col_set)
{
uint key_len;
KEY_PART_INFO *part= index_info->key_part;
for (key_len=0; key_len < max_used_key_length;
key_len += (part++)->store_length)
{
bitmap_set_bit(col_set, part->field->field_index);
}
}
void QUICK_ROR_INTERSECT_SELECT::add_used_key_part_to_set(MY_BITMAP *col_set)
{
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
QUICK_SELECT_WITH_RECORD *quick;
while ((quick= it++))
{
quick->quick->add_used_key_part_to_set(col_set);
}
}
void QUICK_INDEX_SORT_SELECT::add_used_key_part_to_set(MY_BITMAP *col_set)
{
QUICK_RANGE_SELECT *quick;
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
while ((quick= it++))
{
quick->add_used_key_part_to_set(col_set);
}
if (pk_quick_select)
pk_quick_select->add_used_key_part_to_set(col_set);
}
void QUICK_ROR_UNION_SELECT::add_used_key_part_to_set(MY_BITMAP *col_set)
{
QUICK_SELECT_I *quick;
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
while ((quick= it++))
{
quick->add_used_key_part_to_set(col_set);
}
}
/*******************************************************************************
* Implementation of QUICK_GROUP_MIN_MAX_SELECT
*******************************************************************************/
static inline uint get_field_keypart(KEY *index, Field *field);
static inline SEL_ARG * get_index_range_tree(uint index, SEL_TREE* range_tree,
PARAM *param, uint *param_idx);
static bool get_sel_arg_for_keypart(Field *field, SEL_ARG *index_range_tree,
SEL_ARG **cur_range);
static bool get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
KEY_PART_INFO *first_non_group_part,
KEY_PART_INFO *min_max_arg_part,
KEY_PART_INFO *last_part, THD *thd,
uchar *key_infix, uint *key_infix_len,
KEY_PART_INFO **first_non_infix_part);
static bool
check_group_min_max_predicates(Item *cond, Item_field *min_max_arg_item,
Field::imagetype image_type,
bool *has_min_max_fld, bool *has_other_fld);
static void
cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
uint group_key_parts, SEL_TREE *range_tree,
SEL_ARG *index_tree, ha_rows quick_prefix_records,
bool have_min, bool have_max,
double *read_cost, ha_rows *records);
/**
Test if this access method is applicable to a GROUP query with MIN/MAX
functions, and if so, construct a new TRP object.
DESCRIPTION
Test whether a query can be computed via a QUICK_GROUP_MIN_MAX_SELECT.
Queries computable via a QUICK_GROUP_MIN_MAX_SELECT must satisfy the
following conditions:
A) Table T has at least one compound index I of the form:
I = <A_1, ...,A_k, [B_1,..., B_m], C, [D_1,...,D_n]>
B) Query conditions:
B0. Q is over a single table T.
B1. The attributes referenced by Q are a subset of the attributes of I.
B2. All attributes QA in Q can be divided into 3 overlapping groups:
- SA = {S_1, ..., S_l, [C]} - from the SELECT clause, where C is
referenced by any number of MIN and/or MAX functions if present.
- WA = {W_1, ..., W_p} - from the WHERE clause
- GA = <G_1, ..., G_k> - from the GROUP BY clause (if any)
= SA - if Q is a DISTINCT query (based on the
equivalence of DISTINCT and GROUP queries.
- NGA = QA - (GA union C) = {NG_1, ..., NG_m} - the ones not in
GROUP BY and not referenced by MIN/MAX functions.
with the following properties specified below.
B3. If Q has a GROUP BY WITH ROLLUP clause the access method is not
applicable.
SA1. There is at most one attribute in SA referenced by any number of
MIN and/or MAX functions which, which if present, is denoted as C.
SA2. The position of the C attribute in the index is after the last A_k.
SA3. The attribute C can be referenced in the WHERE clause only in
predicates of the forms:
- (C {< | <= | > | >= | =} const)
- (const {< | <= | > | >= | =} C)
- (C between const_i and const_j)
- C IS NULL
- C IS NOT NULL
- C != const
SA4. If Q has a GROUP BY clause, there are no other aggregate functions
except MIN and MAX. For queries with DISTINCT, aggregate functions
are allowed.
SA5. The select list in DISTINCT queries should not contain expressions.
SA6. Clustered index can not be used by GROUP_MIN_MAX quick select
for AGG_FUNC(DISTINCT ...) optimization because cursor position is
never stored after a unique key lookup in the clustered index and
furhter index_next/prev calls can not be used. So loose index scan
optimization can not be used in this case.
SA7. If Q has both AGG_FUNC(DISTINCT ...) and MIN/MAX() functions then this
access method is not used.
For above queries MIN/MAX() aggregation has to be done at
nested_loops_join (end_send_group). But with current design MIN/MAX()
is always set as part of loose index scan. Because of this mismatch
MIN() and MAX() values will be set incorrectly. For such queries to
work we need a new interface for loose index scan. This new interface
should only fetch records with min and max values and let
end_send_group to do aggregation. Until then do not use
loose_index_scan.
GA1. If Q has a GROUP BY clause, then GA is a prefix of I. That is, if
G_i = A_j => i = j.
GA2. If Q has a DISTINCT clause, then there is a permutation of SA that
forms a prefix of I. This permutation is used as the GROUP clause
when the DISTINCT query is converted to a GROUP query.
GA3. The attributes in GA may participate in arbitrary predicates, divided
into two groups:
- RNG(G_1,...,G_q ; where q <= k) is a range condition over the
attributes of a prefix of GA
- PA(G_i1,...G_iq) is an arbitrary predicate over an arbitrary subset
of GA. Since P is applied to only GROUP attributes it filters some
groups, and thus can be applied after the grouping.
GA4. There are no expressions among G_i, just direct column references.
NGA1.If in the index I there is a gap between the last GROUP attribute G_k,
and the MIN/MAX attribute C, then NGA must consist of exactly the
index attributes that constitute the gap. As a result there is a
permutation of NGA, BA=<B_1,...,B_m>, that coincides with the gap
in the index.
NGA2.If BA <> {}, then the WHERE clause must contain a conjunction EQ of
equality conditions for all NG_i of the form (NG_i = const) or
(const = NG_i), such that each NG_i is referenced in exactly one
conjunct. Informally, the predicates provide constants to fill the
gap in the index.
NGA3.If BA <> {}, there can only be one range. TODO: This is a code
limitation and is not strictly needed. See BUG#15947433
WA1. There are no other attributes in the WHERE clause except the ones
referenced in predicates RNG, PA, PC, EQ defined above. Therefore
WA is subset of (GA union NGA union C) for GA,NGA,C that pass the
above tests. By transitivity then it also follows that each WA_i
participates in the index I (if this was already tested for GA, NGA
and C).
WA2. If there is a predicate on C, then it must be in conjunction
to all predicates on all earlier keyparts in I.
C) Overall query form:
SELECT EXPR([A_1,...,A_k], [B_1,...,B_m], [MIN(C)], [MAX(C)])
FROM T
WHERE [RNG(A_1,...,A_p ; where p <= k)]
[AND EQ(B_1,...,B_m)]
[AND PC(C)]
[AND PA(A_i1,...,A_iq)]
GROUP BY A_1,...,A_k
[HAVING PH(A_1, ..., B_1,..., C)]
where EXPR(...) is an arbitrary expression over some or all SELECT fields,
or:
SELECT DISTINCT A_i1,...,A_ik
FROM T
WHERE [RNG(A_1,...,A_p ; where p <= k)]
[AND PA(A_i1,...,A_iq)];
NOTES
If the current query satisfies the conditions above, and if
(mem_root! = NULL), then the function constructs and returns a new TRP
object, that is later used to construct a new QUICK_GROUP_MIN_MAX_SELECT.
If (mem_root == NULL), then the function only tests whether the current
query satisfies the conditions above, and, if so, sets
is_applicable = TRUE.
Queries with DISTINCT for which index access can be used are transformed
into equivalent group-by queries of the form:
SELECT A_1,...,A_k FROM T
WHERE [RNG(A_1,...,A_p ; where p <= k)]
[AND PA(A_i1,...,A_iq)]
GROUP BY A_1,...,A_k;
The group-by list is a permutation of the select attributes, according
to their order in the index.
TODO
- What happens if the query groups by the MIN/MAX field, and there is no
other field as in: "select MY_MIN(a) from t1 group by a" ?
- We assume that the general correctness of the GROUP-BY query was checked
before this point. Is this correct, or do we have to check it completely?
- Lift the limitation in condition (B3), that is, make this access method
applicable to ROLLUP queries.
@param param Parameter from test_quick_select
@param sel_tree Range tree generated by get_mm_tree
@param read_time Best read time so far (=table/index scan time)
@return table read plan
@retval NULL Loose index scan not applicable or mem_root == NULL
@retval !NULL Loose index scan table read plan
*/
static TRP_GROUP_MIN_MAX *
get_best_group_min_max(PARAM *param, SEL_TREE *tree, double read_time)
{
THD *thd= param->thd;
JOIN *join= thd->lex->current_select->join;
TABLE *table= param->table;
bool have_min= FALSE; /* TRUE if there is a MIN function. */
bool have_max= FALSE; /* TRUE if there is a MAX function. */
Item_field *min_max_arg_item= NULL; // The argument of all MIN/MAX functions
KEY_PART_INFO *min_max_arg_part= NULL; /* The corresponding keypart. */
uint group_prefix_len= 0; /* Length (in bytes) of the key prefix. */
KEY *index_info= NULL; /* The index chosen for data access. */
uint index= 0; /* The id of the chosen index. */
uint group_key_parts= 0; // Number of index key parts in the group prefix.
uint used_key_parts= 0; /* Number of index key parts used for access. */
uchar key_infix[MAX_KEY_LENGTH]; /* Constants from equality predicates.*/
uint key_infix_len= 0; /* Length of key_infix. */
TRP_GROUP_MIN_MAX *read_plan= NULL; /* The eventually constructed TRP. */
uint key_part_nr;
ORDER *tmp_group;
Item *item;
Item_field *item_field;
bool is_agg_distinct;
List<Item_field> agg_distinct_flds;
DBUG_ENTER("get_best_group_min_max");
/* Perform few 'cheap' tests whether this access method is applicable. */
if (!join)
DBUG_RETURN(NULL); /* This is not a select statement. */
if ((join->table_count != 1) || /* The query must reference one table. */
(join->select_lex->olap == ROLLUP_TYPE)) /* Check (B3) for ROLLUP */
DBUG_RETURN(NULL);
if (table->s->keys == 0) /* There are no indexes to use. */
DBUG_RETURN(NULL);
if (join->conds && join->conds->used_tables() & OUTER_REF_TABLE_BIT)
DBUG_RETURN(NULL); /* Cannot execute with correlated conditions. */
/* Check (SA1,SA4) and store the only MIN/MAX argument - the C attribute.*/
if (join->make_sum_func_list(join->all_fields, join->fields_list, 1))
DBUG_RETURN(NULL);
List_iterator<Item> select_items_it(join->fields_list);
is_agg_distinct = is_indexed_agg_distinct(join, &agg_distinct_flds);
if ((!join->group_list) && /* Neither GROUP BY nor a DISTINCT query. */
(!join->select_distinct) &&
!is_agg_distinct)
DBUG_RETURN(NULL);
/* Analyze the query in more detail. */
if (join->sum_funcs[0])
{
Item_sum *min_max_item;
Item_sum **func_ptr= join->sum_funcs;
while ((min_max_item= *(func_ptr++)))
{
if (min_max_item->sum_func() == Item_sum::MIN_FUNC)
have_min= TRUE;
else if (min_max_item->sum_func() == Item_sum::MAX_FUNC)
have_max= TRUE;
else if (is_agg_distinct &&
(min_max_item->sum_func() == Item_sum::COUNT_DISTINCT_FUNC ||
min_max_item->sum_func() == Item_sum::SUM_DISTINCT_FUNC ||
min_max_item->sum_func() == Item_sum::AVG_DISTINCT_FUNC))
continue;
else
DBUG_RETURN(NULL);
/* The argument of MIN/MAX. */
Item *expr= min_max_item->get_arg(0)->real_item();
if (expr->type() == Item::FIELD_ITEM) /* Is it an attribute? */
{
if (! min_max_arg_item)
min_max_arg_item= (Item_field*) expr;
else if (! min_max_arg_item->eq(expr, 1))
DBUG_RETURN(NULL);
}
else
DBUG_RETURN(NULL);
}
}
/* Check (SA7). */
if (is_agg_distinct && (have_max || have_min))
{
DBUG_RETURN(NULL);
}
/* Check (SA5). */
if (join->select_distinct)
{
while ((item= select_items_it++))
{
if (item->real_item()->type() != Item::FIELD_ITEM)
DBUG_RETURN(NULL);
}
}
/* Check (GA4) - that there are no expressions among the group attributes. */
for (tmp_group= join->group_list; tmp_group; tmp_group= tmp_group->next)
{
if ((*tmp_group->item)->real_item()->type() != Item::FIELD_ITEM)
DBUG_RETURN(NULL);
}
/*
Check that table has at least one compound index such that the conditions
(GA1,GA2) are all TRUE. If there is more than one such index, select the
first one. Here we set the variables: group_prefix_len and index_info.
*/
KEY *cur_index_info= table->key_info;
KEY *cur_index_info_end= cur_index_info + table->s->keys;
/* Cost-related variables for the best index so far. */
double best_read_cost= DBL_MAX;
ha_rows best_records= 0;
SEL_ARG *best_index_tree= NULL;
ha_rows best_quick_prefix_records= 0;
uint best_param_idx= 0;
const uint pk= param->table->s->primary_key;
uint max_key_part;
SEL_ARG *cur_index_tree= NULL;
ha_rows cur_quick_prefix_records= 0;
uint cur_param_idx=MAX_KEY;
for (uint cur_index= 0 ; cur_index_info != cur_index_info_end ;
cur_index_info++, cur_index++)
{
KEY_PART_INFO *cur_part;
KEY_PART_INFO *end_part; /* Last part for loops. */
/* Last index part. */
KEY_PART_INFO *last_part;
KEY_PART_INFO *first_non_group_part;
KEY_PART_INFO *first_non_infix_part;
uint key_parts;
uint key_infix_parts;
uint cur_group_key_parts= 0;
uint cur_group_prefix_len= 0;
double cur_read_cost;
ha_rows cur_records;
key_map used_key_parts_map;
uint cur_key_infix_len= 0;
uchar cur_key_infix[MAX_KEY_LENGTH];
uint cur_used_key_parts;
/*
Check (B1) - if current index is covering.
(was also: "Exclude UNIQUE indexes ..." but this was removed because
there are cases Loose Scan over a multi-part index is useful).
*/
if (!table->covering_keys.is_set(cur_index))
goto next_index;
/*
Unless extended keys can be used for cur_index:
If the current storage manager is such that it appends the primary key to
each index, then the above condition is insufficient to check if the
index is covering. In such cases it may happen that some fields are
covered by the PK index, but not by the current index. Since we can't
use the concatenation of both indexes for index lookup, such an index
does not qualify as covering in our case. If this is the case, below
we check that all query fields are indeed covered by 'cur_index'.
*/
if (cur_index_info->user_defined_key_parts == table->actual_n_key_parts(cur_index_info)
&& pk < MAX_KEY && cur_index != pk &&
(table->file->ha_table_flags() & HA_PRIMARY_KEY_IN_READ_INDEX))
{
/* For each table field */
for (uint i= 0; i < table->s->fields; i++)
{
Field *cur_field= table->field[i];
/*
If the field is used in the current query ensure that it's
part of 'cur_index'
*/
if (bitmap_is_set(table->read_set, cur_field->field_index) &&
!cur_field->part_of_key_not_clustered.is_set(cur_index))
goto next_index; // Field was not part of key
}
}
max_key_part= 0;
used_key_parts_map.clear_all();
/*
Check (GA1) for GROUP BY queries.
*/
if (join->group_list)
{
cur_part= cur_index_info->key_part;
end_part= cur_part + table->actual_n_key_parts(cur_index_info);
/* Iterate in parallel over the GROUP list and the index parts. */
for (tmp_group= join->group_list; tmp_group && (cur_part != end_part);
tmp_group= tmp_group->next, cur_part++)
{
/*
TODO:
tmp_group::item is an array of Item, is it OK to consider only the
first Item? If so, then why? What is the array for?
*/
/* Above we already checked that all group items are fields. */
DBUG_ASSERT((*tmp_group->item)->real_item()->type() == Item::FIELD_ITEM);
Item_field *group_field= (Item_field *) (*tmp_group->item)->real_item();
if (group_field->field->eq(cur_part->field))
{
cur_group_prefix_len+= cur_part->store_length;
++cur_group_key_parts;
max_key_part= cur_part - cur_index_info->key_part + 1;
used_key_parts_map.set_bit(max_key_part);
}
else
goto next_index;
}
/*
This function is called on the precondition that the index is covering.
Therefore if the GROUP BY list contains more elements than the index,
these are duplicates. The GROUP BY list cannot be a prefix of the index.
*/
if (cur_part == end_part && tmp_group)
goto next_index;
}
/*
Check (GA2) if this is a DISTINCT query.
If GA2, then Store a new ORDER object in group_fields_array at the
position of the key part of item_field->field. Thus we get the ORDER
objects for each field ordered as the corresponding key parts.
Later group_fields_array of ORDER objects is used to convert the query
to a GROUP query.
*/
if ((!join->group_list && join->select_distinct) ||
is_agg_distinct)
{
if (!is_agg_distinct)
{
select_items_it.rewind();
}
List_iterator<Item_field> agg_distinct_flds_it (agg_distinct_flds);
while (NULL != (item = (is_agg_distinct ?
(Item *) agg_distinct_flds_it++ : select_items_it++)))
{
/* (SA5) already checked above. */
item_field= (Item_field*) item->real_item();
DBUG_ASSERT(item->real_item()->type() == Item::FIELD_ITEM);
/* not doing loose index scan for derived tables */
if (!item_field->field)
goto next_index;
/* Find the order of the key part in the index. */
key_part_nr= get_field_keypart(cur_index_info, item_field->field);
/*
Check if this attribute was already present in the select list.
If it was present, then its corresponding key part was alredy used.
*/
if (used_key_parts_map.is_set(key_part_nr))
continue;
if (key_part_nr < 1 ||
(!is_agg_distinct && key_part_nr > join->fields_list.elements))
goto next_index;
cur_part= cur_index_info->key_part + key_part_nr - 1;
cur_group_prefix_len+= cur_part->store_length;
used_key_parts_map.set_bit(key_part_nr);
++cur_group_key_parts;
max_key_part= MY_MAX(max_key_part,key_part_nr);
}
/*
Check that used key parts forms a prefix of the index.
To check this we compare bits in all_parts and cur_parts.
all_parts have all bits set from 0 to (max_key_part-1).
cur_parts have bits set for only used keyparts.
*/
ulonglong all_parts, cur_parts;
all_parts= (1ULL << max_key_part) - 1;
cur_parts= used_key_parts_map.to_ulonglong() >> 1;
if (all_parts != cur_parts)
goto next_index;
}
/* Check (SA2). */
if (min_max_arg_item)
{
key_part_nr= get_field_keypart(cur_index_info, min_max_arg_item->field);
if (key_part_nr <= cur_group_key_parts)
goto next_index;
min_max_arg_part= cur_index_info->key_part + key_part_nr - 1;
}
/*
Aplly a heuristic: there is no point to use loose index scan when we're
using the whole unique index.
*/
if (cur_index_info->flags & HA_NOSAME &&
cur_group_key_parts == cur_index_info->user_defined_key_parts)
{
goto next_index;
}
/*
Check (NGA1, NGA2) and extract a sequence of constants to be used as part
of all search keys.
*/
/*
If there is MIN/MAX, each keypart between the last group part and the
MIN/MAX part must participate in one equality with constants, and all
keyparts after the MIN/MAX part must not be referenced in the query.
If there is no MIN/MAX, the keyparts after the last group part can be
referenced only in equalities with constants, and the referenced keyparts
must form a sequence without any gaps that starts immediately after the
last group keypart.
*/
key_parts= table->actual_n_key_parts(cur_index_info);
last_part= cur_index_info->key_part + key_parts;
first_non_group_part= (cur_group_key_parts < key_parts) ?
cur_index_info->key_part + cur_group_key_parts :
NULL;
first_non_infix_part= min_max_arg_part ?
(min_max_arg_part < last_part) ?
min_max_arg_part :
NULL :
NULL;
if (first_non_group_part &&
(!min_max_arg_part || (min_max_arg_part - first_non_group_part > 0)))
{
if (tree)
{
uint dummy;
SEL_ARG *index_range_tree= get_index_range_tree(cur_index, tree, param,
&dummy);
if (!get_constant_key_infix(cur_index_info, index_range_tree,
first_non_group_part, min_max_arg_part,
last_part, thd, cur_key_infix,
&cur_key_infix_len,
&first_non_infix_part))
goto next_index;
}
else if (min_max_arg_part &&
(min_max_arg_part - first_non_group_part > 0))
{
/*
There is a gap but no range tree, thus no predicates at all for the
non-group keyparts.
*/
goto next_index;
}
else if (first_non_group_part && join->conds)
{
/*
If there is no MIN/MAX function in the query, but some index
key part is referenced in the WHERE clause, then this index
cannot be used because the WHERE condition over the keypart's
field cannot be 'pushed' to the index (because there is no
range 'tree'), and the WHERE clause must be evaluated before
GROUP BY/DISTINCT.
*/
/*
Store the first and last keyparts that need to be analyzed
into one array that can be passed as parameter.
*/
KEY_PART_INFO *key_part_range[2];
key_part_range[0]= first_non_group_part;
key_part_range[1]= last_part;
/* Check if cur_part is referenced in the WHERE clause. */
if (join->conds->walk(&Item::find_item_in_field_list_processor, 0,
(uchar*) key_part_range))
goto next_index;
}
}
/*
Test (WA1) partially - that no other keypart after the last infix part is
referenced in the query.
*/
if (first_non_infix_part)
{
cur_part= first_non_infix_part +
(min_max_arg_part && (min_max_arg_part < last_part));
for (; cur_part != last_part; cur_part++)
{
if (bitmap_is_set(table->read_set, cur_part->field->field_index))
goto next_index;
}
}
/**
Test WA2:If there are conditions on a column C participating in
MIN/MAX, those conditions must be conjunctions to all earlier
keyparts. Otherwise, Loose Index Scan cannot be used.
*/
if (tree && min_max_arg_item)
{
uint dummy;
SEL_ARG *index_range_tree= get_index_range_tree(cur_index, tree, param,
&dummy);
SEL_ARG *cur_range= NULL;
if (get_sel_arg_for_keypart(min_max_arg_part->field,
index_range_tree, &cur_range) ||
(cur_range && cur_range->type != SEL_ARG::KEY_RANGE))
{
goto next_index;
}
}
/* If we got to this point, cur_index_info passes the test. */
key_infix_parts= cur_key_infix_len ? (uint)
(first_non_infix_part - first_non_group_part) : 0;
cur_used_key_parts= cur_group_key_parts + key_infix_parts;
/* Compute the cost of using this index. */
if (tree)
{
/* Find the SEL_ARG sub-tree that corresponds to the chosen index. */
cur_index_tree= get_index_range_tree(cur_index, tree, param,
&cur_param_idx);
/* Check if this range tree can be used for prefix retrieval. */
Cost_estimate dummy_cost;
uint mrr_flags= HA_MRR_USE_DEFAULT_IMPL;
uint mrr_bufsize=0;
cur_quick_prefix_records= check_quick_select(param, cur_param_idx,
FALSE /*don't care*/,
cur_index_tree, TRUE,
&mrr_flags, &mrr_bufsize,
&dummy_cost);
}
cost_group_min_max(table, cur_index_info, cur_used_key_parts,
cur_group_key_parts, tree, cur_index_tree,
cur_quick_prefix_records, have_min, have_max,
&cur_read_cost, &cur_records);
/*
If cur_read_cost is lower than best_read_cost use cur_index.
Do not compare doubles directly because they may have different
representations (64 vs. 80 bits).
*/
if (cur_read_cost < best_read_cost - (DBL_EPSILON * cur_read_cost))
{
index_info= cur_index_info;
index= cur_index;
best_read_cost= cur_read_cost;
best_records= cur_records;
best_index_tree= cur_index_tree;
best_quick_prefix_records= cur_quick_prefix_records;
best_param_idx= cur_param_idx;
group_key_parts= cur_group_key_parts;
group_prefix_len= cur_group_prefix_len;
key_infix_len= cur_key_infix_len;
if (key_infix_len)
memcpy (key_infix, cur_key_infix, sizeof (key_infix));
used_key_parts= cur_used_key_parts;
}
next_index:;
}
if (!index_info) /* No usable index found. */
DBUG_RETURN(NULL);
/* Check (SA3) for the where clause. */
bool has_min_max_fld= false, has_other_fld= false;
if (join->conds && min_max_arg_item &&
!check_group_min_max_predicates(join->conds, min_max_arg_item,
(index_info->flags & HA_SPATIAL) ?
Field::itMBR : Field::itRAW,
&has_min_max_fld, &has_other_fld))
DBUG_RETURN(NULL);
/*
Check (SA6) if clustered key is used
*/
if (is_agg_distinct && index == table->s->primary_key &&
table->file->primary_key_is_clustered())
DBUG_RETURN(NULL);
/* The query passes all tests, so construct a new TRP object. */
read_plan= new (param->mem_root)
TRP_GROUP_MIN_MAX(have_min, have_max, is_agg_distinct,
min_max_arg_part,
group_prefix_len, used_key_parts,
group_key_parts, index_info, index,
key_infix_len,
(key_infix_len > 0) ? key_infix : NULL,
tree, best_index_tree, best_param_idx,
best_quick_prefix_records);
if (read_plan)
{
if (tree && read_plan->quick_prefix_records == 0)
DBUG_RETURN(NULL);
read_plan->read_cost= best_read_cost;
read_plan->records= best_records;
if (read_time < best_read_cost && is_agg_distinct)
{
read_plan->read_cost= 0;
read_plan->use_index_scan();
}
DBUG_PRINT("info",
("Returning group min/max plan: cost: %g, records: %lu",
read_plan->read_cost, (ulong) read_plan->records));
}
DBUG_RETURN(read_plan);
}
/*
Check that the MIN/MAX attribute participates only in range predicates
with constants.
SYNOPSIS
check_group_min_max_predicates()
cond [in] the expression tree being analyzed
min_max_arg [in] the field referenced by the MIN/MAX function(s)
image_type [in]
has_min_max_arg [out] true if the subtree being analyzed references
min_max_arg
has_other_arg [out] true if the subtree being analyzed references a
column other min_max_arg
DESCRIPTION
The function walks recursively over the cond tree representing a WHERE
clause, and checks condition (SA3) - if a field is referenced by a MIN/MAX
aggregate function, it is referenced only by one of the following
predicates $FUNC$:
{=, !=, <, <=, >, >=, between, is [not] null, multiple equal}.
In addition the function checks that the WHERE condition is equivalent to
"cond1 AND cond2" where :
cond1 - does not use min_max_column at all.
cond2 - is an AND/OR tree with leaves in form
"$FUNC$(min_max_column[, const])".
RETURN
TRUE if cond passes the test
FALSE o/w
*/
static bool
check_group_min_max_predicates(Item *cond, Item_field *min_max_arg_item,
Field::imagetype image_type,
bool *has_min_max_arg, bool *has_other_arg)
{
DBUG_ENTER("check_group_min_max_predicates");
DBUG_ASSERT(cond && min_max_arg_item);
cond= cond->real_item();
Item::Type cond_type= cond->real_type();
if (cond_type == Item::COND_ITEM) /* 'AND' or 'OR' */
{
DBUG_PRINT("info", ("Analyzing: %s", ((Item_func*) cond)->func_name()));
List_iterator_fast<Item> li(*((Item_cond*) cond)->argument_list());
Item *and_or_arg;
Item_func::Functype func_type= ((Item_cond*) cond)->functype();
bool has_min_max= false, has_other= false;
while ((and_or_arg= li++))
{
/*
The WHERE clause doesn't pass the condition if:
(1) any subtree doesn't pass the condition or
(2) the subtree passes the test, but it is an OR and it references both
the min/max argument and other columns.
*/
if (!check_group_min_max_predicates(and_or_arg, min_max_arg_item, //1
image_type,
&has_min_max, &has_other) ||
(func_type == Item_func::COND_OR_FUNC && has_min_max && has_other))//2
DBUG_RETURN(FALSE);
}
*has_min_max_arg= has_min_max || *has_min_max_arg;
*has_other_arg= has_other || *has_other_arg;
DBUG_RETURN(TRUE);
}
/*
Disallow loose index scan if the MIN/MAX argument field is referenced by
a subquery in the WHERE clause.
*/
if (unlikely(cond_type == Item::SUBSELECT_ITEM))
{
Item_subselect *subs_cond= (Item_subselect*) cond;
if (subs_cond->is_correlated)
{
DBUG_ASSERT(subs_cond->upper_refs.elements > 0);
List_iterator_fast<Item_subselect::Ref_to_outside>
li(subs_cond->upper_refs);
Item_subselect::Ref_to_outside *dep;
while ((dep= li++))
{
if (dep->item->eq(min_max_arg_item, FALSE))
DBUG_RETURN(FALSE);
}
}
DBUG_RETURN(TRUE);
}
/*
Subquery with IS [NOT] NULL
TODO: Look into the cache_item and optimize it like we do for
subselect's above
*/
if (unlikely(cond_type == Item::CACHE_ITEM))
DBUG_RETURN(cond->const_item());
/*
Condition of the form 'field' is equivalent to 'field <> 0' and thus
satisfies the SA3 condition.
*/
if (cond_type == Item::FIELD_ITEM)
{
DBUG_PRINT("info", ("Analyzing: %s", cond->full_name()));
if (min_max_arg_item->eq((Item_field*)cond, 1))
*has_min_max_arg= true;
else
*has_other_arg= true;
DBUG_RETURN(TRUE);
}
/* We presume that at this point there are no other Items than functions. */
DBUG_ASSERT(cond_type == Item::FUNC_ITEM);
if (unlikely(cond_type != Item::FUNC_ITEM)) /* Safety */
DBUG_RETURN(FALSE);
/* Test if cond references only group-by or non-group fields. */
Item_func *pred= (Item_func*) cond;
Item_func::Functype pred_type= pred->functype();
DBUG_PRINT("info", ("Analyzing: %s", pred->func_name()));
if (pred_type == Item_func::MULT_EQUAL_FUNC)
{
/*
Check that each field in a multiple equality is either a constant or
it is a reference to the min/max argument, or it doesn't contain the
min/max argument at all.
*/
Item_equal_fields_iterator eq_it(*((Item_equal*)pred));
Item *eq_item;
bool has_min_max= false, has_other= false;
while ((eq_item= eq_it++))
{
if (min_max_arg_item->eq(eq_item->real_item(), 1))
has_min_max= true;
else
has_other= true;
}
*has_min_max_arg= has_min_max || *has_min_max_arg;
*has_other_arg= has_other || *has_other_arg;
DBUG_RETURN(!(has_min_max && has_other));
}
Item **arguments= pred->arguments();
Item *cur_arg;
bool has_min_max= false, has_other= false;
for (uint arg_idx= 0; arg_idx < pred->argument_count (); arg_idx++)
{
cur_arg= arguments[arg_idx]->real_item();
DBUG_PRINT("info", ("cur_arg: %s", cur_arg->full_name()));
if (cur_arg->type() == Item::FIELD_ITEM)
{
if (min_max_arg_item->eq(cur_arg, 1))
{
has_min_max= true;
/*
If pred references the MIN/MAX argument, check whether pred is a range
condition that compares the MIN/MAX argument with a constant.
*/
if (pred_type != Item_func::EQUAL_FUNC &&
pred_type != Item_func::LT_FUNC &&
pred_type != Item_func::LE_FUNC &&
pred_type != Item_func::GT_FUNC &&
pred_type != Item_func::GE_FUNC &&
pred_type != Item_func::BETWEEN &&
pred_type != Item_func::ISNULL_FUNC &&
pred_type != Item_func::ISNOTNULL_FUNC &&
pred_type != Item_func::EQ_FUNC &&
pred_type != Item_func::NE_FUNC)
DBUG_RETURN(FALSE);
/* Check that pred compares min_max_arg_item with a constant. */
Item *args[3];
bzero(args, 3 * sizeof(Item*));
bool inv;
/* Test if this is a comparison of a field and a constant. */
if (!simple_pred(pred, args, &inv))
DBUG_RETURN(FALSE);
/* Check for compatible string comparisons - similar to get_mm_leaf. */
if (args[0] && args[1] && !args[2]) // this is a binary function
{
if (args[1]->cmp_type() == TIME_RESULT &&
min_max_arg_item->field->cmp_type() != TIME_RESULT)
DBUG_RETURN(FALSE);
/*
Can't use GROUP_MIN_MAX optimization for ENUM and SET,
because the values are stored as numbers in index,
while MIN() and MAX() work as strings.
It would return the records with min and max enum numeric indexes.
"Bug#45300 MAX() and ENUM type" should be fixed first.
*/
if (min_max_arg_item->field->real_type() == MYSQL_TYPE_ENUM ||
min_max_arg_item->field->real_type() == MYSQL_TYPE_SET)
DBUG_RETURN(FALSE);
if (min_max_arg_item->result_type() == STRING_RESULT &&
/*
Don't use an index when comparing strings of different collations.
*/
((args[1]->result_type() == STRING_RESULT &&
image_type == Field::itRAW &&
min_max_arg_item->field->charset() !=
pred->compare_collation()) ||
/*
We can't always use indexes when comparing a string index to a
number.
*/
(args[1]->result_type() != STRING_RESULT &&
min_max_arg_item->field->cmp_type() != args[1]->result_type())))
DBUG_RETURN(FALSE);
}
}
else
has_other= true;
}
else if (cur_arg->type() == Item::FUNC_ITEM)
{
if (!check_group_min_max_predicates(cur_arg, min_max_arg_item, image_type,
&has_min_max, &has_other))
DBUG_RETURN(FALSE);
}
else if (cur_arg->const_item() && !cur_arg->is_expensive())
{
/*
For predicates of the form "const OP expr" we also have to check 'expr'
to make a decision.
*/
continue;
}
else
DBUG_RETURN(FALSE);
if(has_min_max && has_other)
DBUG_RETURN(FALSE);
}
*has_min_max_arg= has_min_max || *has_min_max_arg;
*has_other_arg= has_other || *has_other_arg;
DBUG_RETURN(TRUE);
}
/*
Get the SEL_ARG tree 'tree' for the keypart covering 'field', if
any. 'tree' must be a unique conjunction to ALL predicates in earlier
keyparts of 'keypart_tree'.
E.g., if 'keypart_tree' is for a composite index (kp1,kp2) and kp2
covers 'field', all these conditions satisfies the requirement:
1. "(kp1=2 OR kp1=3) AND kp2=10" => returns "kp2=10"
2. "(kp1=2 AND kp2=10) OR (kp1=3 AND kp2=10)" => returns "kp2=10"
3. "(kp1=2 AND (kp2=10 OR kp2=11)) OR (kp1=3 AND (kp2=10 OR kp2=11))"
=> returns "kp2=10 OR kp2=11"
whereas these do not
1. "(kp1=2 AND kp2=10) OR kp1=3"
2. "(kp1=2 AND kp2=10) OR (kp1=3 AND kp2=11)"
3. "(kp1=2 AND kp2=10) OR (kp1=3 AND (kp2=10 OR kp2=11))"
This function effectively tests requirement WA2. In combination with
a test that the returned tree has no more than one range it is also
a test of NGA3.
@param[in] field The field we want the SEL_ARG tree for
@param[in] keypart_tree Root node of the SEL_ARG* tree for the index
@param[out] cur_range The SEL_ARG tree, if any, for the keypart
covering field 'keypart_field'
@retval true 'keypart_tree' contained a predicate for 'field' that
is not conjunction to all predicates on earlier keyparts
@retval false otherwise
*/
static bool
get_sel_arg_for_keypart(Field *field,
SEL_ARG *keypart_tree,
SEL_ARG **cur_range)
{
if (keypart_tree == NULL)
return false;
if (keypart_tree->field->eq(field))
{
*cur_range= keypart_tree;
return false;
}
SEL_ARG *tree_first_range= NULL;
SEL_ARG *first_kp= keypart_tree->first();
for (SEL_ARG *cur_kp= first_kp; cur_kp; cur_kp= cur_kp->next)
{
SEL_ARG *curr_tree= NULL;
if (cur_kp->next_key_part)
{
if (get_sel_arg_for_keypart(field,
cur_kp->next_key_part,
&curr_tree))
return true;
}
/*
Check if the SEL_ARG tree for 'field' is identical for all ranges in
'keypart_tree
*/
if (cur_kp == first_kp)
tree_first_range= curr_tree;
else if (!all_same(tree_first_range, curr_tree))
return true;
}
*cur_range= tree_first_range;
return false;
}
/*
Extract a sequence of constants from a conjunction of equality predicates.
SYNOPSIS
get_constant_key_infix()
index_info [in] Descriptor of the chosen index.
index_range_tree [in] Range tree for the chosen index
first_non_group_part [in] First index part after group attribute parts
min_max_arg_part [in] The keypart of the MIN/MAX argument if any
last_part [in] Last keypart of the index
thd [in] Current thread
key_infix [out] Infix of constants to be used for index lookup
key_infix_len [out] Lenghth of the infix
first_non_infix_part [out] The first keypart after the infix (if any)
DESCRIPTION
Test conditions (NGA1, NGA2, NGA3) from get_best_group_min_max(). Namely,
for each keypart field NG_i not in GROUP-BY, check that there is exactly one
constant equality predicate among conds with the form (NG_i = const_ci) or
(const_ci = NG_i).. In addition, there can only be one range when there is
such a gap.
Thus all the NGF_i attributes must fill the 'gap' between the last group-by
attribute and the MIN/MAX attribute in the index (if present). Also ensure
that there is only a single range on NGF_i (NGA3). If these
conditions hold, copy each constant from its corresponding predicate into
key_infix, in the order its NG_i attribute appears in the index, and update
key_infix_len with the total length of the key parts in key_infix.
RETURN
TRUE if the index passes the test
FALSE o/w
*/
static bool
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
KEY_PART_INFO *first_non_group_part,
KEY_PART_INFO *min_max_arg_part,
KEY_PART_INFO *last_part, THD *thd,
uchar *key_infix, uint *key_infix_len,
KEY_PART_INFO **first_non_infix_part)
{
SEL_ARG *cur_range;
KEY_PART_INFO *cur_part;
/* End part for the first loop below. */
KEY_PART_INFO *end_part= min_max_arg_part ? min_max_arg_part : last_part;
*key_infix_len= 0;
uchar *key_ptr= key_infix;
for (cur_part= first_non_group_part; cur_part != end_part; cur_part++)
{
cur_range= NULL;
/*
Check NGA3:
1. get_sel_arg_for_keypart gets the range tree for the 'field' and also
checks for a unique conjunction of this tree with all the predicates
on the earlier keyparts in the index.
2. Check for multiple ranges on the found keypart tree.
We assume that index_range_tree points to the leftmost keypart in
the index.
*/
if (get_sel_arg_for_keypart(cur_part->field, index_range_tree,
&cur_range))
return false;
if (cur_range && cur_range->elements > 1)
return false;
if (!cur_range || cur_range->type != SEL_ARG::KEY_RANGE)
{
if (min_max_arg_part)
return false; /* The current keypart has no range predicates at all. */
else
{
*first_non_infix_part= cur_part;
return true;
}
}
if ((cur_range->min_flag & NO_MIN_RANGE) ||
(cur_range->max_flag & NO_MAX_RANGE) ||
(cur_range->min_flag & NEAR_MIN) || (cur_range->max_flag & NEAR_MAX))
return false;
uint field_length= cur_part->store_length;
if (cur_range->maybe_null &&
cur_range->min_value[0] && cur_range->max_value[0])
{
/*
cur_range specifies 'IS NULL'. In this case the argument points
to a "null value" (is_null_string) that may not always be long
enough for a direct memcpy to a field.
*/
DBUG_ASSERT (field_length > 0);
*key_ptr= 1;
bzero(key_ptr+1,field_length-1);
key_ptr+= field_length;
*key_infix_len+= field_length;
}
else if (memcmp(cur_range->min_value, cur_range->max_value, field_length) == 0)
{ /* cur_range specifies an equality condition. */
memcpy(key_ptr, cur_range->min_value, field_length);
key_ptr+= field_length;
*key_infix_len+= field_length;
}
else
return false;
}
if (!min_max_arg_part && (cur_part == last_part))
*first_non_infix_part= last_part;
return TRUE;
}
/*
Find the key part referenced by a field.
SYNOPSIS
get_field_keypart()
index descriptor of an index
field field that possibly references some key part in index
NOTES
The return value can be used to get a KEY_PART_INFO pointer by
part= index->key_part + get_field_keypart(...) - 1;
RETURN
Positive number which is the consecutive number of the key part, or
0 if field does not reference any index field.
*/
static inline uint
get_field_keypart(KEY *index, Field *field)
{
KEY_PART_INFO *part, *end;
for (part= index->key_part,
end= part + field->table->actual_n_key_parts(index);
part < end; part++)
{
if (field->eq(part->field))
return part - index->key_part + 1;
}
return 0;
}
/*
Find the SEL_ARG sub-tree that corresponds to the chosen index.
SYNOPSIS
get_index_range_tree()
index [in] The ID of the index being looked for
range_tree[in] Tree of ranges being searched
param [in] PARAM from SQL_SELECT::test_quick_select
param_idx [out] Index in the array PARAM::key that corresponds to 'index'
DESCRIPTION
A SEL_TREE contains range trees for all usable indexes. This procedure
finds the SEL_ARG sub-tree for 'index'. The members of a SEL_TREE are
ordered in the same way as the members of PARAM::key, thus we first find
the corresponding index in the array PARAM::key. This index is returned
through the variable param_idx, to be used later as argument of
check_quick_select().
RETURN
Pointer to the SEL_ARG subtree that corresponds to index.
*/
SEL_ARG * get_index_range_tree(uint index, SEL_TREE* range_tree, PARAM *param,
uint *param_idx)
{
uint idx= 0; /* Index nr in param->key_parts */
while (idx < param->keys)
{
if (index == param->real_keynr[idx])
break;
idx++;
}
*param_idx= idx;
return(range_tree->keys[idx]);
}
/*
Compute the cost of a quick_group_min_max_select for a particular index.
SYNOPSIS
cost_group_min_max()
table [in] The table being accessed
index_info [in] The index used to access the table
used_key_parts [in] Number of key parts used to access the index
group_key_parts [in] Number of index key parts in the group prefix
range_tree [in] Tree of ranges for all indexes
index_tree [in] The range tree for the current index
quick_prefix_records [in] Number of records retrieved by the internally
used quick range select if any
have_min [in] True if there is a MIN function
have_max [in] True if there is a MAX function
read_cost [out] The cost to retrieve rows via this quick select
records [out] The number of rows retrieved
DESCRIPTION
This method computes the access cost of a TRP_GROUP_MIN_MAX instance and
the number of rows returned.
NOTES
The cost computation distinguishes several cases:
1) No equality predicates over non-group attributes (thus no key_infix).
If groups are bigger than blocks on the average, then we assume that it
is very unlikely that block ends are aligned with group ends, thus even
if we look for both MIN and MAX keys, all pairs of neighbor MIN/MAX
keys, except for the first MIN and the last MAX keys, will be in the
same block. If groups are smaller than blocks, then we are going to
read all blocks.
2) There are equality predicates over non-group attributes.
In this case the group prefix is extended by additional constants, and
as a result the min/max values are inside sub-groups of the original
groups. The number of blocks that will be read depends on whether the
ends of these sub-groups will be contained in the same or in different
blocks. We compute the probability for the two ends of a subgroup to be
in two different blocks as the ratio of:
- the number of positions of the left-end of a subgroup inside a group,
such that the right end of the subgroup is past the end of the buffer
containing the left-end, and
- the total number of possible positions for the left-end of the
subgroup, which is the number of keys in the containing group.
We assume it is very unlikely that two ends of subsequent subgroups are
in the same block.
3) The are range predicates over the group attributes.
Then some groups may be filtered by the range predicates. We use the
selectivity of the range predicates to decide how many groups will be
filtered.
TODO
- Take into account the optional range predicates over the MIN/MAX
argument.
- Check if we have a PK index and we use all cols - then each key is a
group, and it will be better to use an index scan.
RETURN
None
*/
void cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
uint group_key_parts, SEL_TREE *range_tree,
SEL_ARG *index_tree, ha_rows quick_prefix_records,
bool have_min, bool have_max,
double *read_cost, ha_rows *records)
{
ha_rows table_records;
ha_rows num_groups;
ha_rows num_blocks;
uint keys_per_block;
ha_rows keys_per_group;
ha_rows keys_per_subgroup; /* Average number of keys in sub-groups */
/* formed by a key infix. */
double p_overlap; /* Probability that a sub-group overlaps two blocks. */
double quick_prefix_selectivity;
double io_cost;
DBUG_ENTER("cost_group_min_max");
table_records= table->stat_records();
keys_per_block= (uint) (table->file->stats.block_size / 2 /
(index_info->key_length + table->file->ref_length)
+ 1);
num_blocks= (ha_rows)(table_records / keys_per_block) + 1;
/* Compute the number of keys in a group. */
keys_per_group= (ha_rows) index_info->actual_rec_per_key(group_key_parts - 1);
if (keys_per_group == 0) /* If there is no statistics try to guess */
/* each group contains 10% of all records */
keys_per_group= (table_records / 10) + 1;
num_groups= (table_records / keys_per_group) + 1;
/* Apply the selectivity of the quick select for group prefixes. */
if (range_tree && (quick_prefix_records != HA_POS_ERROR))
{
quick_prefix_selectivity= (double) quick_prefix_records /
(double) table_records;
num_groups= (ha_rows) rint(num_groups * quick_prefix_selectivity);
set_if_bigger(num_groups, 1);
}
if (used_key_parts > group_key_parts)
{ /*
Compute the probability that two ends of a subgroup are inside
different blocks.
*/
keys_per_subgroup= (ha_rows) index_info->actual_rec_per_key(used_key_parts - 1);
if (keys_per_subgroup >= keys_per_block) /* If a subgroup is bigger than */
p_overlap= 1.0; /* a block, it will overlap at least two blocks. */
else
{
double blocks_per_group= (double) num_blocks / (double) num_groups;
p_overlap= (blocks_per_group * (keys_per_subgroup - 1)) / keys_per_group;
p_overlap= MY_MIN(p_overlap, 1.0);
}
io_cost= (double) MY_MIN(num_groups * (1 + p_overlap), num_blocks);
}
else
io_cost= (keys_per_group > keys_per_block) ?
(have_min && have_max) ? (double) (num_groups + 1) :
(double) num_groups :
(double) num_blocks;
/*
CPU cost must be comparable to that of an index scan as computed
in SQL_SELECT::test_quick_select(). When the groups are small,
e.g. for a unique index, using index scan will be cheaper since it
reads the next record without having to re-position to it on every
group. To make the CPU cost reflect this, we estimate the CPU cost
as the sum of:
1. Cost for evaluating the condition (similarly as for index scan).
2. Cost for navigating the index structure (assuming a b-tree).
Note: We only add the cost for one comparision per block. For a
b-tree the number of comparisons will be larger.
TODO: This cost should be provided by the storage engine.
*/
const double tree_traversal_cost=
ceil(log(static_cast<double>(table_records))/
log(static_cast<double>(keys_per_block))) *
1/double(2*TIME_FOR_COMPARE);
const double cpu_cost= num_groups *
(tree_traversal_cost + 1/double(TIME_FOR_COMPARE));
*read_cost= io_cost + cpu_cost;
*records= num_groups;
DBUG_PRINT("info",
("table rows: %lu keys/block: %u keys/group: %lu result rows: %lu blocks: %lu",
(ulong)table_records, keys_per_block, (ulong) keys_per_group,
(ulong) *records, (ulong) num_blocks));
DBUG_VOID_RETURN;
}
/*
Construct a new quick select object for queries with group by with min/max.
SYNOPSIS
TRP_GROUP_MIN_MAX::make_quick()
param Parameter from test_quick_select
retrieve_full_rows ignored
parent_alloc Memory pool to use, if any.
NOTES
Make_quick ignores the retrieve_full_rows parameter because
QUICK_GROUP_MIN_MAX_SELECT always performs 'index only' scans.
The other parameter are ignored as well because all necessary
data to create the QUICK object is computed at this TRP creation
time.
RETURN
New QUICK_GROUP_MIN_MAX_SELECT object if successfully created,
NULL otherwise.
*/
QUICK_SELECT_I *
TRP_GROUP_MIN_MAX::make_quick(PARAM *param, bool retrieve_full_rows,
MEM_ROOT *parent_alloc)
{
QUICK_GROUP_MIN_MAX_SELECT *quick;
DBUG_ENTER("TRP_GROUP_MIN_MAX::make_quick");
quick= new QUICK_GROUP_MIN_MAX_SELECT(param->table,
param->thd->lex->current_select->join,
have_min, have_max,
have_agg_distinct, min_max_arg_part,
group_prefix_len, group_key_parts,
used_key_parts, index_info, index,
read_cost, records, key_infix_len,
key_infix, parent_alloc, is_index_scan);
if (!quick)
DBUG_RETURN(NULL);
if (quick->init())
{
delete quick;
DBUG_RETURN(NULL);
}
if (range_tree)
{
DBUG_ASSERT(quick_prefix_records > 0);
if (quick_prefix_records == HA_POS_ERROR)
quick->quick_prefix_select= NULL; /* Can't construct a quick select. */
else
/* Make a QUICK_RANGE_SELECT to be used for group prefix retrieval. */
quick->quick_prefix_select= get_quick_select(param, param_idx,
index_tree,
HA_MRR_USE_DEFAULT_IMPL, 0,
&quick->alloc);
/*
Extract the SEL_ARG subtree that contains only ranges for the MIN/MAX
attribute, and create an array of QUICK_RANGES to be used by the
new quick select.
*/
if (min_max_arg_part)
{
SEL_ARG *min_max_range= index_tree;
while (min_max_range) /* Find the tree for the MIN/MAX key part. */
{
if (min_max_range->field->eq(min_max_arg_part->field))
break;
min_max_range= min_max_range->next_key_part;
}
/* Scroll to the leftmost interval for the MIN/MAX argument. */
while (min_max_range && min_max_range->prev)
min_max_range= min_max_range->prev;
/* Create an array of QUICK_RANGEs for the MIN/MAX argument. */
while (min_max_range)
{
if (quick->add_range(min_max_range))
{
delete quick;
quick= NULL;
DBUG_RETURN(NULL);
}
min_max_range= min_max_range->next;
}
}
}
else
quick->quick_prefix_select= NULL;
quick->update_key_stat();
quick->adjust_prefix_ranges();
DBUG_RETURN(quick);
}
/*
Construct new quick select for group queries with min/max.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::QUICK_GROUP_MIN_MAX_SELECT()
table The table being accessed
join Descriptor of the current query
have_min TRUE if the query selects a MIN function
have_max TRUE if the query selects a MAX function
min_max_arg_part The only argument field of all MIN/MAX functions
group_prefix_len Length of all key parts in the group prefix
prefix_key_parts All key parts in the group prefix
index_info The index chosen for data access
use_index The id of index_info
read_cost Cost of this access method
records Number of records returned
key_infix_len Length of the key infix appended to the group prefix
key_infix Infix of constants from equality predicates
parent_alloc Memory pool for this and quick_prefix_select data
is_index_scan get the next different key not by jumping on it via
index read, but by scanning until the end of the
rows with equal key value.
RETURN
None
*/
QUICK_GROUP_MIN_MAX_SELECT::
QUICK_GROUP_MIN_MAX_SELECT(TABLE *table, JOIN *join_arg, bool have_min_arg,
bool have_max_arg, bool have_agg_distinct_arg,
KEY_PART_INFO *min_max_arg_part_arg,
uint group_prefix_len_arg, uint group_key_parts_arg,
uint used_key_parts_arg, KEY *index_info_arg,
uint use_index, double read_cost_arg,
ha_rows records_arg, uint key_infix_len_arg,
uchar *key_infix_arg, MEM_ROOT *parent_alloc,
bool is_index_scan_arg)
:file(table->file), join(join_arg), index_info(index_info_arg),
group_prefix_len(group_prefix_len_arg),
group_key_parts(group_key_parts_arg), have_min(have_min_arg),
have_max(have_max_arg), have_agg_distinct(have_agg_distinct_arg),
seen_first_key(FALSE), doing_key_read(FALSE), min_max_arg_part(min_max_arg_part_arg),
key_infix(key_infix_arg), key_infix_len(key_infix_len_arg),
min_functions_it(NULL), max_functions_it(NULL),
is_index_scan(is_index_scan_arg)
{
head= table;
index= use_index;
record= head->record[0];
tmp_record= head->record[1];
read_time= read_cost_arg;
records= records_arg;
used_key_parts= used_key_parts_arg;
real_key_parts= used_key_parts_arg;
real_prefix_len= group_prefix_len + key_infix_len;
group_prefix= NULL;
min_max_arg_len= min_max_arg_part ? min_max_arg_part->store_length : 0;
/*
We can't have parent_alloc set as the init function can't handle this case
yet.
*/
DBUG_ASSERT(!parent_alloc);
if (!parent_alloc)
{
init_sql_alloc(&alloc, join->thd->variables.range_alloc_block_size, 0,
MYF(MY_THREAD_SPECIFIC));
join->thd->mem_root= &alloc;
}
else
bzero(&alloc, sizeof(MEM_ROOT)); // ensure that it's not used
}
/*
Do post-constructor initialization.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::init()
DESCRIPTION
The method performs initialization that cannot be done in the constructor
such as memory allocations that may fail. It allocates memory for the
group prefix and inifix buffers, and for the lists of MIN/MAX item to be
updated during execution.
RETURN
0 OK
other Error code
*/
int QUICK_GROUP_MIN_MAX_SELECT::init()
{
if (group_prefix) /* Already initialized. */
return 0;
/*
We allocate one byte more to serve the case when the last field in
the buffer is compared using uint3korr (e.g. a Field_newdate field)
*/
if (!(last_prefix= (uchar*) alloc_root(&alloc, group_prefix_len+1)))
return 1;
/*
We may use group_prefix to store keys with all select fields, so allocate
enough space for it.
We allocate one byte more to serve the case when the last field in
the buffer is compared using uint3korr (e.g. a Field_newdate field)
*/
if (!(group_prefix= (uchar*) alloc_root(&alloc,
real_prefix_len+min_max_arg_len+1)))
return 1;
if (key_infix_len > 0)
{
/*
The memory location pointed to by key_infix will be deleted soon, so
allocate a new buffer and copy the key_infix into it.
*/
uchar *tmp_key_infix= (uchar*) alloc_root(&alloc, key_infix_len);
if (!tmp_key_infix)
return 1;
memcpy(tmp_key_infix, this->key_infix, key_infix_len);
this->key_infix= tmp_key_infix;
}
if (min_max_arg_part)
{
if (my_init_dynamic_array(&min_max_ranges, sizeof(QUICK_RANGE*), 16, 16,
MYF(MY_THREAD_SPECIFIC)))
return 1;
if (have_min)
{
if (!(min_functions= new List<Item_sum>))
return 1;
}
else
min_functions= NULL;
if (have_max)
{
if (!(max_functions= new List<Item_sum>))
return 1;
}
else
max_functions= NULL;
Item_sum *min_max_item;
Item_sum **func_ptr= join->sum_funcs;
while ((min_max_item= *(func_ptr++)))
{
if (have_min && (min_max_item->sum_func() == Item_sum::MIN_FUNC))
min_functions->push_back(min_max_item);
else if (have_max && (min_max_item->sum_func() == Item_sum::MAX_FUNC))
max_functions->push_back(min_max_item);
}
if (have_min)
{
if (!(min_functions_it= new List_iterator<Item_sum>(*min_functions)))
return 1;
}
if (have_max)
{
if (!(max_functions_it= new List_iterator<Item_sum>(*max_functions)))
return 1;
}
}
else
min_max_ranges.elements= 0;
return 0;
}
QUICK_GROUP_MIN_MAX_SELECT::~QUICK_GROUP_MIN_MAX_SELECT()
{
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::~QUICK_GROUP_MIN_MAX_SELECT");
if (file->inited != handler::NONE)
{
DBUG_ASSERT(file == head->file);
if (doing_key_read)
head->disable_keyread();
/*
There may be a code path when the same table was first accessed by index,
then the index is closed, and the table is scanned (order by + loose scan).
*/
file->ha_index_or_rnd_end();
}
if (min_max_arg_part)
delete_dynamic(&min_max_ranges);
free_root(&alloc,MYF(0));
delete min_functions_it;
delete max_functions_it;
delete quick_prefix_select;
DBUG_VOID_RETURN;
}
/*
Eventually create and add a new quick range object.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::add_range()
sel_range Range object from which a
NOTES
Construct a new QUICK_RANGE object from a SEL_ARG object, and
add it to the array min_max_ranges. If sel_arg is an infinite
range, e.g. (x < 5 or x > 4), then skip it and do not construct
a quick range.
RETURN
FALSE on success
TRUE otherwise
*/
bool QUICK_GROUP_MIN_MAX_SELECT::add_range(SEL_ARG *sel_range)
{
QUICK_RANGE *range;
uint range_flag= sel_range->min_flag | sel_range->max_flag;
/* Skip (-inf,+inf) ranges, e.g. (x < 5 or x > 4). */
if ((range_flag & NO_MIN_RANGE) && (range_flag & NO_MAX_RANGE))
return FALSE;
if (!(sel_range->min_flag & NO_MIN_RANGE) &&
!(sel_range->max_flag & NO_MAX_RANGE))
{
if (sel_range->maybe_null &&
sel_range->min_value[0] && sel_range->max_value[0])
range_flag|= NULL_RANGE; /* IS NULL condition */
else if (memcmp(sel_range->min_value, sel_range->max_value,
min_max_arg_len) == 0)
range_flag|= EQ_RANGE; /* equality condition */
}
range= new QUICK_RANGE(sel_range->min_value, min_max_arg_len,
make_keypart_map(sel_range->part),
sel_range->max_value, min_max_arg_len,
make_keypart_map(sel_range->part),
range_flag);
if (!range)
return TRUE;
if (insert_dynamic(&min_max_ranges, (uchar*)&range))
return TRUE;
return FALSE;
}
/*
Opens the ranges if there are more conditions in quick_prefix_select than
the ones used for jumping through the prefixes.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::adjust_prefix_ranges()
NOTES
quick_prefix_select is made over the conditions on the whole key.
It defines a number of ranges of length x.
However when jumping through the prefixes we use only the the first
few most significant keyparts in the range key. However if there
are more keyparts to follow the ones we are using we must make the
condition on the key inclusive (because x < "ab" means
x[0] < 'a' OR (x[0] == 'a' AND x[1] < 'b').
To achive the above we must turn off the NEAR_MIN/NEAR_MAX
*/
void QUICK_GROUP_MIN_MAX_SELECT::adjust_prefix_ranges ()
{
if (quick_prefix_select &&
group_prefix_len < quick_prefix_select->max_used_key_length)
{
DYNAMIC_ARRAY *arr;
uint inx;
for (inx= 0, arr= &quick_prefix_select->ranges; inx < arr->elements; inx++)
{
QUICK_RANGE *range;
get_dynamic(arr, (uchar*)&range, inx);
range->flag &= ~(NEAR_MIN | NEAR_MAX);
}
}
}
/*
Determine the total number and length of the keys that will be used for
index lookup.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::update_key_stat()
DESCRIPTION
The total length of the keys used for index lookup depends on whether
there are any predicates referencing the min/max argument, and/or if
the min/max argument field can be NULL.
This function does an optimistic analysis whether the search key might
be extended by a constant for the min/max keypart. It is 'optimistic'
because during actual execution it may happen that a particular range
is skipped, and then a shorter key will be used. However this is data
dependent and can't be easily estimated here.
RETURN
None
*/
void QUICK_GROUP_MIN_MAX_SELECT::update_key_stat()
{
max_used_key_length= real_prefix_len;
if (min_max_ranges.elements > 0)
{
QUICK_RANGE *cur_range;
if (have_min)
{ /* Check if the right-most range has a lower boundary. */
get_dynamic(&min_max_ranges, (uchar*)&cur_range,
min_max_ranges.elements - 1);
if (!(cur_range->flag & NO_MIN_RANGE))
{
max_used_key_length+= min_max_arg_len;
used_key_parts++;
return;
}
}
if (have_max)
{ /* Check if the left-most range has an upper boundary. */
get_dynamic(&min_max_ranges, (uchar*)&cur_range, 0);
if (!(cur_range->flag & NO_MAX_RANGE))
{
max_used_key_length+= min_max_arg_len;
used_key_parts++;
return;
}
}
}
else if (have_min && min_max_arg_part &&
min_max_arg_part->field->real_maybe_null())
{
/*
If a MIN/MAX argument value is NULL, we can quickly determine
that we're in the beginning of the next group, because NULLs
are always < any other value. This allows us to quickly
determine the end of the current group and jump to the next
group (see next_min()) and thus effectively increases the
usable key length.
*/
max_used_key_length+= min_max_arg_len;
used_key_parts++;
}
}
/*
Initialize a quick group min/max select for key retrieval.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::reset()
DESCRIPTION
Initialize the index chosen for access and find and store the prefix
of the last group. The method is expensive since it performs disk access.
RETURN
0 OK
other Error code
*/
int QUICK_GROUP_MIN_MAX_SELECT::reset(void)
{
int result;
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::reset");
seen_first_key= FALSE;
if (!head->key_read)
{
doing_key_read= 1;
head->enable_keyread(); /* We need only the key attributes */
}
if ((result= file->ha_index_init(index,1)))
{
head->file->print_error(result, MYF(0));
DBUG_RETURN(result);
}
if (quick_prefix_select && quick_prefix_select->reset())
DBUG_RETURN(1);
result= file->ha_index_last(record);
if (result == HA_ERR_END_OF_FILE)
DBUG_RETURN(0);
/* Save the prefix of the last group. */
key_copy(last_prefix, record, index_info, group_prefix_len);
DBUG_RETURN(0);
}
/*
Get the next key containing the MIN and/or MAX key for the next group.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::get_next()
DESCRIPTION
The method finds the next subsequent group of records that satisfies the
query conditions and finds the keys that contain the MIN/MAX values for
the key part referenced by the MIN/MAX function(s). Once a group and its
MIN/MAX values are found, store these values in the Item_sum objects for
the MIN/MAX functions. The rest of the values in the result row are stored
in the Item_field::result_field of each select field. If the query does
not contain MIN and/or MAX functions, then the function only finds the
group prefix, which is a query answer itself.
NOTES
If both MIN and MAX are computed, then we use the fact that if there is
no MIN key, there can't be a MAX key as well, so we can skip looking
for a MAX key in this case.
RETURN
0 on success
HA_ERR_END_OF_FILE if returned all keys
other if some error occurred
*/
int QUICK_GROUP_MIN_MAX_SELECT::get_next()
{
int min_res= 0;
int max_res= 0;
#ifdef HPUX11
/*
volatile is required by a bug in the HP compiler due to which the
last test of result fails.
*/
volatile int result;
#else
int result;
#endif
int is_last_prefix= 0;
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::get_next");
/*
Loop until a group is found that satisfies all query conditions or the last
group is reached.
*/
do
{
result= next_prefix();
/*
Check if this is the last group prefix. Notice that at this point
this->record contains the current prefix in record format.
*/
if (!result)
{
is_last_prefix= key_cmp(index_info->key_part, last_prefix,
group_prefix_len);
DBUG_ASSERT(is_last_prefix <= 0);
}
else
{
if (result == HA_ERR_KEY_NOT_FOUND)
continue;
break;
}
if (have_min)
{
min_res= next_min();
if (min_res == 0)
update_min_result();
}
/* If there is no MIN in the group, there is no MAX either. */
if ((have_max && !have_min) ||
(have_max && have_min && (min_res == 0)))
{
max_res= next_max();
if (max_res == 0)
update_max_result();
/* If a MIN was found, a MAX must have been found as well. */
DBUG_ASSERT((have_max && !have_min) ||
(have_max && have_min && (max_res == 0)));
}
/*
If this is just a GROUP BY or DISTINCT without MIN or MAX and there
are equality predicates for the key parts after the group, find the
first sub-group with the extended prefix.
*/
if (!have_min && !have_max && key_infix_len > 0)
result= file->ha_index_read_map(record, group_prefix,
make_prev_keypart_map(real_key_parts),
HA_READ_KEY_EXACT);
result= have_min ? min_res : have_max ? max_res : result;
} while ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
is_last_prefix != 0);
if (result == HA_ERR_KEY_NOT_FOUND)
result= HA_ERR_END_OF_FILE;
DBUG_RETURN(result);
}
/*
Retrieve the minimal key in the next group.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::next_min()
DESCRIPTION
Find the minimal key within this group such that the key satisfies the query
conditions and NULL semantics. The found key is loaded into this->record.
IMPLEMENTATION
Depending on the values of min_max_ranges.elements, key_infix_len, and
whether there is a NULL in the MIN field, this function may directly
return without any data access. In this case we use the key loaded into
this->record by the call to this->next_prefix() just before this call.
RETURN
0 on success
HA_ERR_KEY_NOT_FOUND if no MIN key was found that fulfills all conditions.
HA_ERR_END_OF_FILE - "" -
other if some error occurred
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_min()
{
int result= 0;
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_min");
/* Find the MIN key using the eventually extended group prefix. */
if (min_max_ranges.elements > 0)
{
if ((result= next_min_in_range()))
DBUG_RETURN(result);
}
else
{
/* Apply the constant equality conditions to the non-group select fields */
if (key_infix_len > 0)
{
if ((result=
file->ha_index_read_map(record, group_prefix,
make_prev_keypart_map(real_key_parts),
HA_READ_KEY_EXACT)))
DBUG_RETURN(result);
}
/*
If the min/max argument field is NULL, skip subsequent rows in the same
group with NULL in it. Notice that:
- if the first row in a group doesn't have a NULL in the field, no row
in the same group has (because NULL < any other value),
- min_max_arg_part->field->ptr points to some place in 'record'.
*/
if (min_max_arg_part && min_max_arg_part->field->is_null())
{
uchar *tmp_key_buff= (uchar*)my_alloca(max_used_key_length);
/* Find the first subsequent record without NULL in the MIN/MAX field. */
key_copy(tmp_key_buff, record, index_info, max_used_key_length);
result= file->ha_index_read_map(record, tmp_key_buff,
make_keypart_map(real_key_parts),
HA_READ_AFTER_KEY);
/*
Check if the new record belongs to the current group by comparing its
prefix with the group's prefix. If it is from the next group, then the
whole group has NULLs in the MIN/MAX field, so use the first record in
the group as a result.
TODO:
It is possible to reuse this new record as the result candidate for the
next call to next_min(), and to save one lookup in the next call. For
this add a new member 'this->next_group_prefix'.
*/
if (!result)
{
if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
key_restore(record, tmp_key_buff, index_info, 0);
}
else if (result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE)
result= 0; /* There is a result in any case. */
my_afree(tmp_key_buff);
}
}
/*
If the MIN attribute is non-nullable, this->record already contains the
MIN key in the group, so just return.
*/
DBUG_RETURN(result);
}
/*
Retrieve the maximal key in the next group.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::next_max()
DESCRIPTION
Lookup the maximal key of the group, and store it into this->record.
RETURN
0 on success
HA_ERR_KEY_NOT_FOUND if no MAX key was found that fulfills all conditions.
HA_ERR_END_OF_FILE - "" -
other if some error occurred
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_max()
{
int result;
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_max");
/* Get the last key in the (possibly extended) group. */
if (min_max_ranges.elements > 0)
result= next_max_in_range();
else
result= file->ha_index_read_map(record, group_prefix,
make_prev_keypart_map(real_key_parts),
HA_READ_PREFIX_LAST);
DBUG_RETURN(result);
}
/**
Find the next different key value by skiping all the rows with the same key
value.
Implements a specialized loose index access method for queries
containing aggregate functions with distinct of the form:
SELECT [SUM|COUNT|AVG](DISTINCT a,...) FROM t
This method comes to replace the index scan + Unique class
(distinct selection) for loose index scan that visits all the rows of a
covering index instead of jumping in the begining of each group.
TODO: Placeholder function. To be replaced by a handler API call
@param is_index_scan hint to use index scan instead of random index read
to find the next different value.
@param file table handler
@param key_part group key to compare
@param record row data
@param group_prefix current key prefix data
@param group_prefix_len length of the current key prefix data
@param group_key_parts number of the current key prefix columns
@return status
@retval 0 success
@retval !0 failure
*/
static int index_next_different (bool is_index_scan, handler *file,
KEY_PART_INFO *key_part, uchar * record,
const uchar * group_prefix,
uint group_prefix_len,
uint group_key_parts)
{
if (is_index_scan)
{
int result= 0;
while (!key_cmp (key_part, group_prefix, group_prefix_len))
{
result= file->ha_index_next(record);
if (result)
return(result);
}
return result;
}
else
return file->ha_index_read_map(record, group_prefix,
make_prev_keypart_map(group_key_parts),
HA_READ_AFTER_KEY);
}
/*
Determine the prefix of the next group.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::next_prefix()
DESCRIPTION
Determine the prefix of the next group that satisfies the query conditions.
If there is a range condition referencing the group attributes, use a
QUICK_RANGE_SELECT object to retrieve the *first* key that satisfies the
condition. If there is a key infix of constants, append this infix
immediately after the group attributes. The possibly extended prefix is
stored in this->group_prefix. The first key of the found group is stored in
this->record, on which relies this->next_min().
RETURN
0 on success
HA_ERR_KEY_NOT_FOUND if there is no key with the formed prefix
HA_ERR_END_OF_FILE if there are no more keys
other if some error occurred
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_prefix()
{
int result;
DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_prefix");
if (quick_prefix_select)
{
uchar *cur_prefix= seen_first_key ? group_prefix : NULL;
if ((result= quick_prefix_select->get_next_prefix(group_prefix_len,
group_key_parts,
cur_prefix)))
DBUG_RETURN(result);
seen_first_key= TRUE;
}
else
{
if (!seen_first_key)
{
result= file->ha_index_first(record);
if (result)
DBUG_RETURN(result);
seen_first_key= TRUE;
}
else
{
/* Load the first key in this group into record. */
result= index_next_different (is_index_scan, file, index_info->key_part,
record, group_prefix, group_prefix_len,
group_key_parts);
if (result)
DBUG_RETURN(result);
}
}
/* Save the prefix of this group for subsequent calls. */
key_copy(group_prefix, record, index_info, group_prefix_len);
/* Append key_infix to group_prefix. */
if (key_infix_len > 0)
memcpy(group_prefix + group_prefix_len,
key_infix, key_infix_len);
DBUG_RETURN(0);
}
/*
Find the minimal key in a group that satisfies some range conditions for the
min/max argument field.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::next_min_in_range()
DESCRIPTION
Given the sequence of ranges min_max_ranges, find the minimal key that is
in the left-most possible range. If there is no such key, then the current
group does not have a MIN key that satisfies the WHERE clause. If a key is
found, its value is stored in this->record.
RETURN
0 on success
HA_ERR_KEY_NOT_FOUND if there is no key with the given prefix in any of
the ranges
HA_ERR_END_OF_FILE - "" -
other if some error
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_min_in_range()
{
ha_rkey_function find_flag;
key_part_map keypart_map;
QUICK_RANGE *cur_range;
bool found_null= FALSE;
int result= HA_ERR_KEY_NOT_FOUND;
DBUG_ASSERT(min_max_ranges.elements > 0);
for (uint range_idx= 0; range_idx < min_max_ranges.elements; range_idx++)
{ /* Search from the left-most range to the right. */
get_dynamic(&min_max_ranges, (uchar*)&cur_range, range_idx);
/*
If the current value for the min/max argument is bigger than the right
boundary of cur_range, there is no need to check this range.
*/
if (range_idx != 0 && !(cur_range->flag & NO_MAX_RANGE) &&
(key_cmp(min_max_arg_part, (const uchar*) cur_range->max_key,
min_max_arg_len) == 1))
continue;
if (cur_range->flag & NO_MIN_RANGE)
{
keypart_map= make_prev_keypart_map(real_key_parts);
find_flag= HA_READ_KEY_EXACT;
}
else
{
/* Extend the search key with the lower boundary for this range. */
memcpy(group_prefix + real_prefix_len, cur_range->min_key,
cur_range->min_length);
keypart_map= make_keypart_map(real_key_parts);
find_flag= (cur_range->flag & (EQ_RANGE | NULL_RANGE)) ?
HA_READ_KEY_EXACT : (cur_range->flag & NEAR_MIN) ?
HA_READ_AFTER_KEY : HA_READ_KEY_OR_NEXT;
}
result= file->ha_index_read_map(record, group_prefix, keypart_map,
find_flag);
if (result)
{
if ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
(cur_range->flag & (EQ_RANGE | NULL_RANGE)))
continue; /* Check the next range. */
/*
In all other cases (HA_ERR_*, HA_READ_KEY_EXACT with NO_MIN_RANGE,
HA_READ_AFTER_KEY, HA_READ_KEY_OR_NEXT) if the lookup failed for this
range, it can't succeed for any other subsequent range.
*/
break;
}
/* A key was found. */
if (cur_range->flag & EQ_RANGE)
break; /* No need to perform the checks below for equal keys. */
if (cur_range->flag & NULL_RANGE)
{
/*
Remember this key, and continue looking for a non-NULL key that
satisfies some other condition.
*/
memcpy(tmp_record, record, head->s->rec_buff_length);
found_null= TRUE;
continue;
}
/* Check if record belongs to the current group. */
if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
{
result= HA_ERR_KEY_NOT_FOUND;
continue;
}
/* If there is an upper limit, check if the found key is in the range. */
if ( !(cur_range->flag & NO_MAX_RANGE) )
{
/* Compose the MAX key for the range. */
uchar *max_key= (uchar*) my_alloca(real_prefix_len + min_max_arg_len);
memcpy(max_key, group_prefix, real_prefix_len);
memcpy(max_key + real_prefix_len, cur_range->max_key,
cur_range->max_length);
/* Compare the found key with max_key. */
int cmp_res= key_cmp(index_info->key_part, max_key,
real_prefix_len + min_max_arg_len);
my_afree(max_key);
/*
The key is outside of the range if:
the interval is open and the key is equal to the maximum boundry
or
the key is greater than the maximum
*/
if (((cur_range->flag & NEAR_MAX) && cmp_res == 0) ||
cmp_res > 0)
{
result= HA_ERR_KEY_NOT_FOUND;
continue;
}
}
/* If we got to this point, the current key qualifies as MIN. */
DBUG_ASSERT(result == 0);
break;
}
/*
If there was a key with NULL in the MIN/MAX field, and there was no other
key without NULL from the same group that satisfies some other condition,
then use the key with the NULL.
*/
if (found_null && result)
{
memcpy(record, tmp_record, head->s->rec_buff_length);
result= 0;
}
return result;
}
/*
Find the maximal key in a group that satisfies some range conditions for the
min/max argument field.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::next_max_in_range()
DESCRIPTION
Given the sequence of ranges min_max_ranges, find the maximal key that is
in the right-most possible range. If there is no such key, then the current
group does not have a MAX key that satisfies the WHERE clause. If a key is
found, its value is stored in this->record.
RETURN
0 on success
HA_ERR_KEY_NOT_FOUND if there is no key with the given prefix in any of
the ranges
HA_ERR_END_OF_FILE - "" -
other if some error
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_max_in_range()
{
ha_rkey_function find_flag;
key_part_map keypart_map;
QUICK_RANGE *cur_range;
int result;
DBUG_ASSERT(min_max_ranges.elements > 0);
for (uint range_idx= min_max_ranges.elements; range_idx > 0; range_idx--)
{ /* Search from the right-most range to the left. */
get_dynamic(&min_max_ranges, (uchar*)&cur_range, range_idx - 1);
/*
If the current value for the min/max argument is smaller than the left
boundary of cur_range, there is no need to check this range.
*/
if (range_idx != min_max_ranges.elements &&
!(cur_range->flag & NO_MIN_RANGE) &&
(key_cmp(min_max_arg_part, (const uchar*) cur_range->min_key,
min_max_arg_len) == -1))
continue;
if (cur_range->flag & NO_MAX_RANGE)
{
keypart_map= make_prev_keypart_map(real_key_parts);
find_flag= HA_READ_PREFIX_LAST;
}
else
{
/* Extend the search key with the upper boundary for this range. */
memcpy(group_prefix + real_prefix_len, cur_range->max_key,
cur_range->max_length);
keypart_map= make_keypart_map(real_key_parts);
find_flag= (cur_range->flag & EQ_RANGE) ?
HA_READ_KEY_EXACT : (cur_range->flag & NEAR_MAX) ?
HA_READ_BEFORE_KEY : HA_READ_PREFIX_LAST_OR_PREV;
}
result= file->ha_index_read_map(record, group_prefix, keypart_map,
find_flag);
if (result)
{
if ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
(cur_range->flag & EQ_RANGE))
continue; /* Check the next range. */
/*
In no key was found with this upper bound, there certainly are no keys
in the ranges to the left.
*/
return result;
}
/* A key was found. */
if (cur_range->flag & EQ_RANGE)
return 0; /* No need to perform the checks below for equal keys. */
/* Check if record belongs to the current group. */
if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
continue; // Row not found
/* If there is a lower limit, check if the found key is in the range. */
if ( !(cur_range->flag & NO_MIN_RANGE) )
{
/* Compose the MIN key for the range. */
uchar *min_key= (uchar*) my_alloca(real_prefix_len + min_max_arg_len);
memcpy(min_key, group_prefix, real_prefix_len);
memcpy(min_key + real_prefix_len, cur_range->min_key,
cur_range->min_length);
/* Compare the found key with min_key. */
int cmp_res= key_cmp(index_info->key_part, min_key,
real_prefix_len + min_max_arg_len);
my_afree(min_key);
/*
The key is outside of the range if:
the interval is open and the key is equal to the minimum boundry
or
the key is less than the minimum
*/
if (((cur_range->flag & NEAR_MIN) && cmp_res == 0) ||
cmp_res < 0)
continue;
}
/* If we got to this point, the current key qualifies as MAX. */
return result;
}
return HA_ERR_KEY_NOT_FOUND;
}
/*
Update all MIN function results with the newly found value.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::update_min_result()
DESCRIPTION
The method iterates through all MIN functions and updates the result value
of each function by calling Item_sum::reset(), which in turn picks the new
result value from this->head->record[0], previously updated by
next_min(). The updated value is stored in a member variable of each of the
Item_sum objects, depending on the value type.
IMPLEMENTATION
The update must be done separately for MIN and MAX, immediately after
next_min() was called and before next_max() is called, because both MIN and
MAX take their result value from the same buffer this->head->record[0]
(i.e. this->record).
RETURN
None
*/
void QUICK_GROUP_MIN_MAX_SELECT::update_min_result()
{
Item_sum *min_func;
min_functions_it->rewind();
while ((min_func= (*min_functions_it)++))
min_func->reset_and_add();
}
/*
Update all MAX function results with the newly found value.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::update_max_result()
DESCRIPTION
The method iterates through all MAX functions and updates the result value
of each function by calling Item_sum::reset(), which in turn picks the new
result value from this->head->record[0], previously updated by
next_max(). The updated value is stored in a member variable of each of the
Item_sum objects, depending on the value type.
IMPLEMENTATION
The update must be done separately for MIN and MAX, immediately after
next_max() was called, because both MIN and MAX take their result value
from the same buffer this->head->record[0] (i.e. this->record).
RETURN
None
*/
void QUICK_GROUP_MIN_MAX_SELECT::update_max_result()
{
Item_sum *max_func;
max_functions_it->rewind();
while ((max_func= (*max_functions_it)++))
max_func->reset_and_add();
}
/*
Append comma-separated list of keys this quick select uses to key_names;
append comma-separated list of corresponding used lengths to used_lengths.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::add_keys_and_lengths()
key_names [out] Names of used indexes
used_lengths [out] Corresponding lengths of the index names
DESCRIPTION
This method is used by select_describe to extract the names of the
indexes used by a quick select.
*/
void QUICK_GROUP_MIN_MAX_SELECT::add_keys_and_lengths(String *key_names,
String *used_lengths)
{
bool first= TRUE;
add_key_and_length(key_names, used_lengths, &first);
}
#ifndef DBUG_OFF
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
const char *msg)
{
SEL_ARG **key,**end;
int idx;
char buff[1024];
DBUG_ENTER("print_sel_tree");
String tmp(buff,sizeof(buff),&my_charset_bin);
tmp.length(0);
for (idx= 0,key=tree->keys, end=key+param->keys ;
key != end ;
key++,idx++)
{
if (tree_map->is_set(idx))
{
uint keynr= param->real_keynr[idx];
if (tmp.length())
tmp.append(',');
tmp.append(param->table->key_info[keynr].name);
}
}
if (!tmp.length())
tmp.append(STRING_WITH_LEN("(empty)"));
DBUG_PRINT("info", ("SEL_TREE: 0x%lx (%s) scans: %s", (long) tree, msg,
tmp.c_ptr_safe()));
DBUG_VOID_RETURN;
}
static void print_ror_scans_arr(TABLE *table, const char *msg,
struct st_ror_scan_info **start,
struct st_ror_scan_info **end)
{
DBUG_ENTER("print_ror_scans_arr");
char buff[1024];
String tmp(buff,sizeof(buff),&my_charset_bin);
tmp.length(0);
for (;start != end; start++)
{
if (tmp.length())
tmp.append(',');
tmp.append(table->key_info[(*start)->keynr].name);
}
if (!tmp.length())
tmp.append(STRING_WITH_LEN("(empty)"));
DBUG_PRINT("info", ("ROR key scans (%s): %s", msg, tmp.c_ptr()));
DBUG_VOID_RETURN;
}
/*****************************************************************************
** Print a quick range for debugging
** TODO:
** This should be changed to use a String to store each row instead
** of locking the DEBUG stream !
*****************************************************************************/
static void
print_key(KEY_PART *key_part, const uchar *key, uint used_length)
{
char buff[1024];
const uchar *key_end= key+used_length;
uint store_length;
TABLE *table= key_part->field->table;
my_bitmap_map *old_sets[2];
dbug_tmp_use_all_columns(table, old_sets, table->read_set, table->write_set);
for (; key < key_end; key+=store_length, key_part++)
{
String tmp(buff,sizeof(buff),&my_charset_bin);
Field *field= key_part->field;
store_length= key_part->store_length;
if (field->real_maybe_null())
{
if (*key)
{
fwrite("NULL",sizeof(char),4,DBUG_FILE);
continue;
}
key++; // Skip null byte
store_length--;
}
field->set_key_image(key, key_part->length);
if (field->type() == MYSQL_TYPE_BIT)
(void) field->val_int_as_str(&tmp, 1);
else
field->val_str(&tmp);
fwrite(tmp.ptr(),sizeof(char),tmp.length(),DBUG_FILE);
if (key+store_length < key_end)
fputc('/',DBUG_FILE);
}
dbug_tmp_restore_column_maps(table->read_set, table->write_set, old_sets);
}
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg)
{
char buf[MAX_KEY/8+1];
TABLE *table;
my_bitmap_map *old_sets[2];
DBUG_ENTER("print_quick");
if (!quick)
DBUG_VOID_RETURN;
DBUG_LOCK_FILE;
table= quick->head;
dbug_tmp_use_all_columns(table, old_sets, table->read_set, table->write_set);
quick->dbug_dump(0, TRUE);
dbug_tmp_restore_column_maps(table->read_set, table->write_set, old_sets);
fprintf(DBUG_FILE,"other_keys: 0x%s:\n", needed_reg->print(buf));
DBUG_UNLOCK_FILE;
DBUG_VOID_RETURN;
}
void QUICK_RANGE_SELECT::dbug_dump(int indent, bool verbose)
{
/* purecov: begin inspected */
fprintf(DBUG_FILE, "%*squick range select, key %s, length: %d\n",
indent, "", head->key_info[index].name, max_used_key_length);
if (verbose)
{
QUICK_RANGE *range;
QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
QUICK_RANGE **end_range= pr + ranges.elements;
for (; pr != end_range; ++pr)
{
fprintf(DBUG_FILE, "%*s", indent + 2, "");
range= *pr;
if (!(range->flag & NO_MIN_RANGE))
{
print_key(key_parts, range->min_key, range->min_length);
if (range->flag & NEAR_MIN)
fputs(" < ",DBUG_FILE);
else
fputs(" <= ",DBUG_FILE);
}
fputs("X",DBUG_FILE);
if (!(range->flag & NO_MAX_RANGE))
{
if (range->flag & NEAR_MAX)
fputs(" < ",DBUG_FILE);
else
fputs(" <= ",DBUG_FILE);
print_key(key_parts, range->max_key, range->max_length);
}
fputs("\n",DBUG_FILE);
}
}
/* purecov: end */
}
void QUICK_INDEX_SORT_SELECT::dbug_dump(int indent, bool verbose)
{
List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
QUICK_RANGE_SELECT *quick;
fprintf(DBUG_FILE, "%*squick index_merge select\n", indent, "");
fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
while ((quick= it++))
quick->dbug_dump(indent+2, verbose);
if (pk_quick_select)
{
fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
pk_quick_select->dbug_dump(indent+2, verbose);
}
fprintf(DBUG_FILE, "%*s}\n", indent, "");
}
void QUICK_ROR_INTERSECT_SELECT::dbug_dump(int indent, bool verbose)
{
List_iterator_fast<QUICK_SELECT_WITH_RECORD> it(quick_selects);
QUICK_SELECT_WITH_RECORD *qr;
fprintf(DBUG_FILE, "%*squick ROR-intersect select, %scovering\n",
indent, "", need_to_fetch_row? "":"non-");
fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
while ((qr= it++))
qr->quick->dbug_dump(indent+2, verbose);
if (cpk_quick)
{
fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
cpk_quick->dbug_dump(indent+2, verbose);
}
fprintf(DBUG_FILE, "%*s}\n", indent, "");
}
void QUICK_ROR_UNION_SELECT::dbug_dump(int indent, bool verbose)
{
List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
QUICK_SELECT_I *quick;
fprintf(DBUG_FILE, "%*squick ROR-union select\n", indent, "");
fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
while ((quick= it++))
quick->dbug_dump(indent+2, verbose);
fprintf(DBUG_FILE, "%*s}\n", indent, "");
}
/*
Print quick select information to DBUG_FILE.
SYNOPSIS
QUICK_GROUP_MIN_MAX_SELECT::dbug_dump()
indent Indentation offset
verbose If TRUE show more detailed output.
DESCRIPTION
Print the contents of this quick select to DBUG_FILE. The method also
calls dbug_dump() for the used quick select if any.
IMPLEMENTATION
Caller is responsible for locking DBUG_FILE before this call and unlocking
it afterwards.
RETURN
None
*/
void QUICK_GROUP_MIN_MAX_SELECT::dbug_dump(int indent, bool verbose)
{
fprintf(DBUG_FILE,
"%*squick_group_min_max_select: index %s (%d), length: %d\n",
indent, "", index_info->name, index, max_used_key_length);
if (key_infix_len > 0)
{
fprintf(DBUG_FILE, "%*susing key_infix with length %d:\n",
indent, "", key_infix_len);
}
if (quick_prefix_select)
{
fprintf(DBUG_FILE, "%*susing quick_range_select:\n", indent, "");
quick_prefix_select->dbug_dump(indent + 2, verbose);
}
if (min_max_ranges.elements > 0)
{
fprintf(DBUG_FILE, "%*susing %d quick_ranges for MIN/MAX:\n",
indent, "", min_max_ranges.elements);
}
}
#endif /* !DBUG_OFF */
|