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
|
from __future__ import annotations
import asyncio
import gc
import importlib
import itertools
import logging
import os
import sys
import tempfile
import threading
import traceback
import warnings
import weakref
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor
from concurrent.futures.process import BrokenProcessPool
from numbers import Number
from operator import add
from time import sleep
from unittest import mock
import psutil
import pytest
from tlz import first, merge, pluck, sliding_window
from tornado.ioloop import IOLoop
import dask
from dask import delayed
from dask.system import CPU_COUNT
from dask.utils import tmpfile
import distributed
from distributed import (
Client,
Event,
Nanny,
WorkerPlugin,
default_client,
get_client,
get_worker,
profile,
wait,
)
from distributed.comm.registry import backends
from distributed.compatibility import LINUX, WINDOWS, to_thread
from distributed.core import CommClosedError, Status, rpc
from distributed.diagnostics import nvml
from distributed.diagnostics.plugin import (
CondaInstall,
ForwardOutput,
PackageInstall,
PipInstall,
)
from distributed.metrics import time
from distributed.protocol import pickle
from distributed.scheduler import KilledWorker, Scheduler
from distributed.utils_test import (
NO_AMM,
BlockedExecute,
BlockedGatherDep,
BlockedGetData,
TaskStateMetadataPlugin,
_LockedCommPool,
assert_story,
captured_logger,
dec,
div,
freeze_batched_send,
freeze_data_fetching,
gen_cluster,
gen_test,
inc,
mul,
nodebug,
raises_with_cause,
slowinc,
slowsum,
wait_for_state,
)
from distributed.worker import (
Worker,
benchmark_disk,
benchmark_memory,
benchmark_network,
error_message,
logger,
)
from distributed.worker_state_machine import (
AcquireReplicasEvent,
ComputeTaskEvent,
ExecuteFailureEvent,
ExecuteSuccessEvent,
RemoveReplicasEvent,
SerializedTask,
StealRequestEvent,
)
pytestmark = pytest.mark.ci1
@gen_cluster(nthreads=[])
async def test_worker_nthreads(s):
async with Worker(s.address) as w:
assert w.executor._max_workers == CPU_COUNT
@gen_cluster()
async def test_str(s, a, b):
assert a.address in str(a)
assert a.address in repr(a)
assert str(a.state.nthreads) in str(a)
assert str(a.state.nthreads) in repr(a)
assert str(a.state.executing_count) in repr(a)
@gen_cluster(nthreads=[])
async def test_identity(s):
async with Worker(s.address) as w:
ident = w.identity()
assert "Worker" in ident["type"]
assert ident["scheduler"] == s.address
assert isinstance(ident["nthreads"], int)
assert isinstance(ident["memory_limit"], Number)
@gen_cluster(client=True)
async def test_worker_bad_args(c, s, a, b):
class NoReprObj:
"""This object cannot be properly represented as a string."""
def __str__(self):
raise ValueError("I have no str representation.")
def __repr__(self):
raise ValueError("I have no repr representation.")
x = c.submit(NoReprObj, workers=a.address)
await wait(x)
assert not a.state.executing_count
assert a.data
def bad_func(*args, **kwargs):
1 / 0
class MockLoggingHandler(logging.Handler):
"""Mock logging handler to check for expected logs."""
def __init__(self, *args, **kwargs):
self.reset()
super().__init__(*args, **kwargs)
def emit(self, record):
self.messages[record.levelname.lower()].append(record.getMessage())
def reset(self):
self.messages = {
"debug": [],
"info": [],
"warning": [],
"error": [],
"critical": [],
}
hdlr = MockLoggingHandler()
old_level = logger.level
logger.setLevel(logging.DEBUG)
logger.addHandler(hdlr)
y = c.submit(bad_func, x, k=x, workers=b.address)
await wait(y)
assert not b.state.executing_count
assert y.status == "error"
# Make sure job died because of bad func and not because of bad
# argument.
with pytest.raises(ZeroDivisionError):
await y
tb = await y._traceback()
assert any("1 / 0" in line for line in pluck(3, traceback.extract_tb(tb)) if line)
assert "Compute Failed" in hdlr.messages["warning"][0]
assert y.key in hdlr.messages["warning"][0]
logger.setLevel(old_level)
# Now we check that both workers are still alive.
xx = c.submit(add, 1, 2, workers=a.address)
yy = c.submit(add, 3, 4, workers=b.address)
results = await c._gather([xx, yy])
assert tuple(results) == (3, 7)
@gen_cluster(client=True)
async def test_upload_file(c, s, a, b):
assert not os.path.exists(os.path.join(a.local_directory, "foobar.py"))
assert not os.path.exists(os.path.join(b.local_directory, "foobar.py"))
assert a.local_directory != b.local_directory
async with rpc(a.address) as aa, rpc(b.address) as bb:
await asyncio.gather(
aa.upload_file(filename="foobar.py", data=b"x = 123"),
bb.upload_file(filename="foobar.py", data="x = 123"),
)
assert os.path.exists(os.path.join(a.local_directory, "foobar.py"))
assert os.path.exists(os.path.join(b.local_directory, "foobar.py"))
def g():
import foobar
return foobar.x
future = c.submit(g, workers=a.address)
result = await future
assert result == 123
await c.close()
await s.close()
assert not os.path.exists(os.path.join(a.local_directory, "foobar.py"))
@pytest.mark.skip(reason="don't yet support uploading pyc files")
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_upload_file_pyc(c, s, w):
with tmpfile() as dirname:
os.mkdir(dirname)
with open(os.path.join(dirname, "foo.py"), mode="w") as f:
f.write("def f():\n return 123")
sys.path.append(dirname)
try:
import foo
assert foo.f() == 123
pyc = importlib.util.cache_from_source(os.path.join(dirname, "foo.py"))
assert os.path.exists(pyc)
await c.upload_file(pyc)
def g():
import foo
return foo.x
future = c.submit(g)
result = await future
assert result == 123
finally:
sys.path.remove(dirname)
@gen_cluster(client=True)
async def test_upload_egg(c, s, a, b):
eggname = "testegg-1.0.0-py3.4.egg"
if 'AUTOPKGTEST_TMP' in os.environ:
local_file = os.path.abspath(eggname)
else:
local_file = __file__.replace('test_worker.py', eggname)
assert not os.path.exists(os.path.join(a.local_directory, eggname))
assert not os.path.exists(os.path.join(b.local_directory, eggname))
assert a.local_directory != b.local_directory
await c.upload_file(filename=local_file)
assert os.path.exists(os.path.join(a.local_directory, eggname))
assert os.path.exists(os.path.join(b.local_directory, eggname))
def g(x):
import testegg
return testegg.inc(x)
future = c.submit(g, 10, workers=a.address)
result = await future
assert result == 10 + 1
await c.close()
await s.close()
await a.close()
await b.close()
assert not os.path.exists(os.path.join(a.local_directory, eggname))
@gen_cluster(client=True)
async def test_upload_pyz(c, s, a, b):
pyzname = "mytest.pyz"
if 'AUTOPKGTEST_TMP' in os.environ:
local_file = os.path.abspath(pyzname)
else:
local_file = __file__.replace('test_worker.py', pyzname)
assert not os.path.exists(os.path.join(a.local_directory, pyzname))
assert not os.path.exists(os.path.join(b.local_directory, pyzname))
assert a.local_directory != b.local_directory
await c.upload_file(filename=local_file)
assert os.path.exists(os.path.join(a.local_directory, pyzname))
assert os.path.exists(os.path.join(b.local_directory, pyzname))
def g(x):
from mytest import mytest
return mytest.inc(x)
future = c.submit(g, 10, workers=a.address)
result = await future
assert result == 10 + 1
await c.close()
await s.close()
await a.close()
await b.close()
assert not os.path.exists(os.path.join(a.local_directory, pyzname))
@pytest.mark.xfail(reason="Still lose time to network I/O")
@gen_cluster(client=True)
async def test_upload_large_file(c, s, a, b):
pytest.importorskip("crick")
await asyncio.sleep(0.05)
async with rpc(a.address) as aa:
await aa.upload_file(filename="myfile.dat", data=b"0" * 100000000)
await asyncio.sleep(0.05)
assert a.digests["tick-duration"].components[0].max() < 0.050
@gen_cluster()
async def test_broadcast(s, a, b):
async with rpc(s.address) as cc:
results = await cc.broadcast(msg={"op": "ping"})
assert results == {a.address: b"pong", b.address: b"pong"}
@gen_cluster(nthreads=[])
async def test_worker_with_port_zero(s):
async with Worker(s.address) as w:
assert isinstance(w.port, int)
assert w.port > 1024
@gen_cluster(nthreads=[])
async def test_worker_port_range(s):
port = "9867:9868"
async with Worker(s.address, port=port) as w1:
assert w1.port == 9867 # Selects first port in range
async with Worker(s.address, port=port) as w2:
assert w2.port == 9868 # Selects next port in range
with raises_with_cause(
RuntimeError, None, ValueError, match_cause="Could not start Worker"
): # No more ports left
async with Worker(s.address, port=port):
pass
@pytest.mark.slow
@gen_test(timeout=60)
async def test_worker_waits_for_scheduler():
w = Worker("127.0.0.1:8724")
async def f():
await w
task = asyncio.create_task(f())
await asyncio.sleep(3)
assert not task.done()
task.cancel()
assert w.status not in (Status.closed, Status.running, Status.paused)
await w.close(timeout=0.1)
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_worker_task_data(c, s, w):
x = delayed(2)
xx = c.persist(x)
await wait(xx)
assert w.data[x.key] == 2
def test_error_message():
class MyException(Exception):
def __init__(self, a, b):
self.args = (a + b,)
def __str__(self):
return "MyException(%s)" % self.args
msg = error_message(MyException("Hello", "World!"))
assert "Hello" in str(msg["exception"])
max_error_len = 100
with dask.config.set({"distributed.admin.max-error-length": max_error_len}):
msg = error_message(RuntimeError("-" * max_error_len))
assert len(msg["exception_text"]) <= max_error_len + 30
assert len(msg["exception_text"]) < max_error_len * 2
msg = error_message(RuntimeError("-" * max_error_len * 20))
max_error_len = 1000000
with dask.config.set({"distributed.admin.max-error-length": max_error_len}):
msg = error_message(RuntimeError("-" * max_error_len * 2))
assert len(msg["exception_text"]) > 10100 # default + 100
@gen_cluster(client=True)
async def test_chained_error_message(c, s, a, b):
def chained_exception_fn():
class MyException(Exception):
def __init__(self, msg):
self.msg = msg
def __str__(self):
return "MyException(%s)" % self.msg
exception = MyException("Foo")
inner_exception = MyException("Bar")
try:
raise inner_exception
except Exception as e:
raise exception from e
f = c.submit(chained_exception_fn)
try:
await f
except Exception as e:
assert e.__cause__ is not None
assert "Bar" in str(e.__cause__)
@pytest.mark.slow
@pytest.mark.parametrize("sync", [True, False])
@pytest.mark.parametrize(
"exc_type", [BaseException, SystemExit, KeyboardInterrupt, asyncio.CancelledError]
)
@gen_cluster(
nthreads=[("", 1)],
client=True,
Worker=Nanny,
config={"distributed.scheduler.allowed-failures": 0},
)
async def test_base_exception_in_task(c, s, a, sync, exc_type):
if sync:
def raiser():
raise exc_type(f"this is a {exc_type}")
else:
async def raiser():
raise exc_type(f"this is a {exc_type}")
f = c.submit(raiser)
try:
with pytest.raises(
KilledWorker if exc_type in (SystemExit, KeyboardInterrupt) else exc_type
):
await f
except BaseException as e:
# Prevent test failure from killing the whole pytest process
traceback.print_exc()
pytest.fail(f"BaseException propagated back to test: {e!r}. See stdout.")
# Nanny restarts it
await c.wait_for_workers(1)
@gen_test()
async def test_plugin_exception():
class MyPlugin:
def setup(self, worker=None):
raise ValueError("Setup failed")
async with Scheduler(port=0, dashboard_address=":0") as s:
with raises_with_cause(
RuntimeError, "Worker failed to start", ValueError, "Setup failed"
):
async with Worker(
s.address,
plugins={
MyPlugin(),
},
) as w:
pass
@gen_test()
async def test_plugin_multiple_exceptions():
class MyPlugin1:
def setup(self, worker=None):
raise ValueError("MyPlugin1 Error")
class MyPlugin2:
def setup(self, worker=None):
raise RuntimeError("MyPlugin2 Error")
async with Scheduler(port=0, dashboard_address=":0") as s:
# There's no guarantee on the order of which exception is raised first
with raises_with_cause(
RuntimeError,
None,
(ValueError, RuntimeError),
match_cause="MyPlugin.* Error",
):
with captured_logger("distributed.worker") as logger:
async with Worker(
s.address,
plugins={
MyPlugin1(),
MyPlugin2(),
},
) as w:
pass
text = logger.getvalue()
assert "MyPlugin1 Error" in text
assert "MyPlugin2 Error" in text
@gen_test()
async def test_plugin_internal_exception():
async with Scheduler(port=0) as s:
with raises_with_cause(
RuntimeError,
"Worker failed to start",
UnicodeDecodeError,
match_cause="codec can't decode",
):
async with Worker(
s.address,
plugins={
b"corrupting pickle" + pickle.dumps(lambda: None),
},
) as w:
pass
@gen_cluster(client=True)
async def test_gather(c, s, a, b):
x, y = await c.scatter(["x", "y"], workers=[b.address])
async with rpc(a.address) as aa:
resp = await aa.gather(who_has={x.key: [b.address], y.key: [b.address]})
assert resp == {"status": "OK"}
assert a.data[x.key] == b.data[x.key] == "x"
assert a.data[y.key] == b.data[y.key] == "y"
@gen_cluster(client=True)
async def test_gather_missing_keys(c, s, a, b):
"""A key is missing. Other keys are gathered successfully."""
x = await c.scatter("x", workers=[b.address])
async with rpc(a.address) as aa:
resp = await aa.gather(who_has={x.key: [b.address], "y": [b.address]})
assert resp == {"status": "partial-fail", "keys": {"y": (b.address,)}}
assert a.data[x.key] == b.data[x.key] == "x"
@gen_cluster(client=True, worker_kwargs={"timeout": "100ms"})
async def test_gather_missing_workers(c, s, a, b):
"""A worker owning the only copy of a key is missing.
Keys from other workers are gathered successfully.
"""
assert b.address.startswith("tcp://127.0.0.1:")
bad_addr = "tcp://127.0.0.1:12345"
x = await c.scatter("x", workers=[b.address])
async with rpc(a.address) as aa:
resp = await aa.gather(who_has={x.key: [b.address], "y": [bad_addr]})
assert resp == {"status": "partial-fail", "keys": {"y": (bad_addr,)}}
assert a.data[x.key] == b.data[x.key] == "x"
@pytest.mark.parametrize("missing_first", [False, True])
@gen_cluster(client=True, worker_kwargs={"timeout": "100ms"})
async def test_gather_missing_workers_replicated(c, s, a, b, missing_first):
"""A worker owning a redundant copy of a key is missing.
The key is successfully gathered from other workers.
"""
assert b.address.startswith("tcp://127.0.0.1:")
x = await c.scatter("x", workers=[b.address])
bad_addr = "tcp://127.0.0.1:12345"
# Order matters! Test both
addrs = [bad_addr, b.address] if missing_first else [b.address, bad_addr]
async with rpc(a.address) as aa:
resp = await aa.gather(who_has={x.key: addrs})
assert resp == {"status": "OK"}
assert a.data[x.key] == b.data[x.key] == "x"
@gen_cluster(nthreads=[])
async def test_io_loop(s):
async with Worker(s.address) as w:
assert w.io_loop is w.loop is s.loop
@gen_cluster(nthreads=[])
async def test_io_loop_alternate_loop(s, loop):
async def main():
with pytest.warns(
DeprecationWarning,
match=r"The `loop` argument to `Worker` is ignored, and will be "
r"removed in a future release. The Worker always binds to the current loop",
):
async with Worker(s.address, loop=loop) as w:
assert w.io_loop is w.loop is IOLoop.current()
await to_thread(asyncio.run, main())
@gen_cluster(client=True)
async def test_access_key(c, s, a, b):
def f(i):
from distributed.worker import thread_state
return thread_state.key
futures = [c.submit(f, i, key="x-%d" % i) for i in range(20)]
results = await c._gather(futures)
assert list(results) == ["x-%d" % i for i in range(20)]
@gen_cluster(client=True)
async def test_run_dask_worker(c, s, a, b):
def f(dask_worker=None):
return dask_worker.id
response = await c._run(f)
assert response == {a.address: a.id, b.address: b.id}
@gen_cluster(client=True)
async def test_run_dask_worker_kwonlyarg(c, s, a, b):
def f(*, dask_worker=None):
return dask_worker.id
response = await c._run(f)
assert response == {a.address: a.id, b.address: b.id}
@gen_cluster(client=True)
async def test_run_coroutine_dask_worker(c, s, a, b):
async def f(dask_worker=None):
await asyncio.sleep(0.001)
return dask_worker.id
response = await c.run(f)
assert response == {a.address: a.id, b.address: b.id}
@gen_cluster(client=True, nthreads=[])
async def test_Executor(c, s):
with ThreadPoolExecutor(2) as e:
async with Worker(s.address, executor=e) as w:
assert w.executor is e
future = c.submit(inc, 1)
result = await future
assert result == 2
assert e._threads # had to do some work
@gen_cluster(nthreads=[("127.0.0.1", 1)])
async def test_close_on_disconnect(s, w):
await s.close()
start = time()
while w.status != Status.closed:
await asyncio.sleep(0.01)
assert time() < start + 5
@gen_cluster(nthreads=[])
async def test_memory_limit_auto(s):
async with Worker(s.address, nthreads=1) as a, Worker(
s.address, nthreads=2
) as b, Worker(s.address, nthreads=100) as c, Worker(s.address, nthreads=200) as d:
assert isinstance(a.memory_manager.memory_limit, Number)
assert isinstance(b.memory_manager.memory_limit, Number)
if CPU_COUNT > 1:
assert a.memory_manager.memory_limit < b.memory_manager.memory_limit
assert c.memory_manager.memory_limit == d.memory_manager.memory_limit
@gen_cluster(client=True)
async def test_inter_worker_communication(c, s, a, b):
[x, y] = await c._scatter([1, 2], workers=a.address)
future = c.submit(add, x, y, workers=b.address)
result = await future
assert result == 3
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_clean(c, s, a):
x = c.submit(inc, 1)
await x
collections = [a.state.tasks, a.data, a.threads]
assert all(collections)
x.release()
while x.key in a.state.tasks:
await asyncio.sleep(0.01)
assert not any(collections)
@gen_cluster(client=True)
async def test_message_breakup(c, s, a, b):
n = 100_000
a.state.transfer_message_bytes_limit = 10 * n
b.state.transfer_message_bytes_limit = 10 * n
xs = [
c.submit(mul, b"%d" % i, n, key=f"x{i}", workers=[a.address]) for i in range(30)
]
y = c.submit(lambda _: None, xs, key="y", workers=[b.address])
await y
assert 2 <= len(b.transfer_incoming_log) <= 20
assert 2 <= len(a.transfer_outgoing_log) <= 20
assert all(msg["who"] == b.address for msg in a.transfer_outgoing_log)
assert all(msg["who"] == a.address for msg in a.transfer_incoming_log)
@gen_cluster(client=True)
async def test_types(c, s, a, b):
assert all(ts.type is None for ts in a.state.tasks.values())
assert all(ts.type is None for ts in b.state.tasks.values())
x = c.submit(inc, 1, workers=a.address)
await wait(x)
assert a.state.tasks[x.key].type == int
y = c.submit(inc, x, workers=b.address)
await wait(y)
assert b.state.tasks[x.key].type == int
assert b.state.tasks[y.key].type == int
await c._cancel(y)
start = time()
while y.key in b.data:
await asyncio.sleep(0.01)
assert time() < start + 5
assert y.key not in b.state.tasks
@gen_cluster()
async def test_system_monitor(s, a, b):
assert b.monitor
b.monitor.update()
@gen_cluster(
client=True, nthreads=[("127.0.0.1", 2, {"resources": {"A": 1}}), ("127.0.0.1", 1)]
)
async def test_restrictions(c, s, a, b):
# Worker has resource available
assert a.state.available_resources == {"A": 1}
# Resource restrictions
x = c.submit(inc, 1, resources={"A": 1})
await x
ts = a.state.tasks[x.key]
assert ts.resource_restrictions == {"A": 1}
await c.cancel([x])
while ts.state == "executing":
# Resource should be unavailable while task isn't finished
assert a.state.available_resources == {"A": 0}
await asyncio.sleep(0.01)
# Resource restored after task is in memory
assert a.state.available_resources["A"] == 1
@gen_cluster(client=True)
async def test_clean_nbytes(c, s, a, b):
L = [delayed(inc)(i) for i in range(10)]
for _ in range(5):
L = [delayed(add)(x, y) for x, y in sliding_window(2, L)]
total = delayed(sum)(L)
future = c.compute(total)
await wait(future)
await asyncio.sleep(1)
assert (
len(list(filter(None, [ts.nbytes for ts in a.state.tasks.values()])))
+ len(list(filter(None, [ts.nbytes for ts in b.state.tasks.values()])))
== 1
)
@pytest.mark.parametrize("as_deps", [True, False])
@gen_cluster(client=True, nthreads=[("", 1)] * 21)
async def test_gather_many_small(c, s, a, *snd_workers, as_deps):
"""If the dependencies of a given task are very small, do not limit the number of
concurrent outgoing connections. If multiple small fetches from the same worker are
scheduled all at once, they will result in a single call to gather_dep.
"""
a.state.transfer_incoming_count_limit = 2
futures = await c.scatter(
{f"x{i}": i for i in range(100)},
workers=[w.address for w in snd_workers],
)
assert all(w.data for w in snd_workers)
if as_deps:
future = c.submit(lambda _: None, futures, key="y", workers=[a.address])
await wait(future)
else:
s.request_acquire_replicas(a.address, list(futures), stimulus_id="test")
while len(a.data) < 100:
await asyncio.sleep(0.01)
assert a.state.transfer_incoming_bytes == 0
story = a.state.story("request-dep", "receive-dep")
assert len(story) == 40 # 1 request-dep + 1 receive-dep per sender worker
# All GatherDep instructions are fired at the same time; each fetches all keys
# available on the sender worker
for ev in story[:20]:
assert ev[0] == "request-dep"
assert len(ev[2]) > 1
for ev in story[20:]:
assert ev[0] == "receive-dep"
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)] * 3)
async def test_multiple_transfers(c, s, w1, w2, w3):
x = c.submit(inc, 1, workers=w1.address)
y = c.submit(inc, 2, workers=w2.address)
z = c.submit(add, x, y, workers=w3.address)
await wait(z)
r = w3.state.tasks[z.key].startstops
transfers = [t for t in r if t["action"] == "transfer"]
assert len(transfers) == 2
@pytest.mark.xfail(reason="very high flakiness")
@gen_cluster(
client=True,
nthreads=[("127.0.0.1", 1)] * 3,
config=NO_AMM,
)
async def test_share_communication(c, s, w1, w2, w3):
x = c.submit(
mul, b"1", int(w3.transfer_message_bytes_limit + 1), workers=w1.address
)
y = c.submit(
mul, b"2", int(w3.transfer_message_bytes_limit + 1), workers=w2.address
)
await wait([x, y])
await c.replicate([x, y], workers=[w1.address, w2.address])
z = c.submit(add, x, y, workers=w3.address)
await wait(z)
assert len(w3.transfer_incoming_log) == 2
assert w1.transfer_outgoing_log
assert w2.transfer_outgoing_log
@pytest.mark.xfail(reason="very high flakiness")
@gen_cluster(client=True)
async def test_dont_overlap_communications_to_same_worker(c, s, a, b):
x = c.submit(mul, b"1", int(b.transfer_message_bytes_limit + 1), workers=a.address)
y = c.submit(mul, b"2", int(b.transfer_message_bytes_limit + 1), workers=a.address)
await wait([x, y])
z = c.submit(add, x, y, workers=b.address)
await wait(z)
assert len(b.transfer_incoming_log) == 2
l1, l2 = b.transfer_incoming_log
assert l1["stop"] < l2["start"]
@gen_cluster(client=True)
async def test_log_exception_on_failed_task(c, s, a, b):
with captured_logger("distributed.worker") as logger:
future = c.submit(div, 1, 0)
await wait(future)
await asyncio.sleep(0.1)
text = logger.getvalue()
assert "ZeroDivisionError" in text
assert "Exception" in text
@gen_cluster(client=True)
async def test_clean_up_dependencies(c, s, a, b):
x = delayed(inc)(1)
y = delayed(inc)(2)
xx = delayed(inc)(x)
yy = delayed(inc)(y)
z = delayed(add)(xx, yy)
zz = c.persist(z)
await wait(zz)
while len(a.data) + len(b.data) > 1:
await asyncio.sleep(0.01)
assert set(a.data) | set(b.data) == {zz.key}
@gen_cluster(client=True, config=NO_AMM)
async def test_hold_onto_dependents(c, s, a, b):
x = c.submit(inc, 1, workers=a.address)
y = c.submit(inc, x, workers=b.address)
await wait(y)
assert x.key in b.data
await c._cancel(y)
while x.key not in b.data:
await asyncio.sleep(0.1)
@pytest.mark.xfail(reason="asyncio.wait_for bug")
@gen_test()
async def test_worker_death_timeout():
w = Worker("tcp://127.0.0.1:12345", death_timeout=0.1)
with pytest.raises(asyncio.TimeoutError) as info:
await w
assert "Worker" in str(info.value)
assert "timed out" in str(info.value)
assert w.status == Status.failed
@gen_cluster(client=True)
async def test_stop_doing_unnecessary_work(c, s, a, b):
futures = c.map(slowinc, range(1000), delay=0.01)
await asyncio.sleep(0.1)
del futures
start = time()
while a.state.executing_count:
await asyncio.sleep(0.01)
assert time() - start < 0.5
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_priorities(c, s, w):
values = []
for i in range(10):
a = delayed(slowinc)(i, dask_key_name="a-%d" % i, delay=0.01)
a1 = delayed(inc)(a, dask_key_name="a1-%d" % i)
a2 = delayed(inc)(a1, dask_key_name="a2-%d" % i)
b1 = delayed(dec)(a, dask_key_name="b1-%d" % i) # <<-- least favored
values.append(a2)
values.append(b1)
futures = c.compute(values)
await wait(futures)
log = [
t[0]
for t in w.state.log
if t[1] == "executing" and t[2] == "memory" and not t[0].startswith("finalize")
]
assert any(key.startswith("b1") for key in log[: len(log) // 2])
@gen_cluster(client=True)
async def test_heartbeats(c, s, a, b):
x = s.workers[a.address].last_seen
start = time()
await asyncio.sleep(a.periodic_callbacks["heartbeat"].callback_time / 1000 + 0.1)
while s.workers[a.address].last_seen == x:
await asyncio.sleep(0.01)
assert time() < start + 2
assert a.periodic_callbacks["heartbeat"].callback_time < 1000
@pytest.mark.parametrize("worker", [Worker, Nanny])
def test_worker_dir(worker, tmpdir):
@gen_cluster(client=True, worker_kwargs={"local_directory": str(tmpdir)})
async def test_worker_dir(c, s, a, b):
directories = [w.local_directory for w in s.workers.values()]
assert all(d.startswith(str(tmpdir)) for d in directories)
assert len(set(directories)) == 2 # distinct
test_worker_dir()
@gen_cluster(client=True, nthreads=[], config={"temporary-directory": None})
async def test_default_worker_dir(c, s):
expect = os.path.join(tempfile.gettempdir(), "dask-worker-space")
async with Worker(s.address) as w:
assert os.path.dirname(w.local_directory) == expect
async with Nanny(s.address) as n:
assert n.local_directory == expect
results = await c.run(lambda dask_worker: dask_worker.local_directory)
assert os.path.dirname(results[n.worker_address]) == expect
@gen_cluster(client=True)
async def test_dataframe_attribute_error(c, s, a, b):
class BadSize:
def __init__(self, data):
self.data = data
def __sizeof__(self):
raise TypeError("Hello")
future = c.submit(BadSize, 123)
result = await future
assert result.data == 123
@gen_cluster()
async def test_pid(s, a, b):
assert s.workers[a.address].pid == os.getpid()
@gen_cluster(client=True)
async def test_get_client(c, s, a, b):
def f(x):
cc = get_client()
future = cc.submit(inc, x)
return future.result()
assert default_client() is c
future = c.submit(f, 10, workers=a.address)
result = await future
assert result == 11
assert a._client
assert not b._client
assert a._client is c
assert default_client() is c
a_client = a._client
for i in range(10):
await wait(c.submit(f, i))
assert a._client is a_client
def test_get_client_sync(client):
def f(x):
cc = get_client()
future = cc.submit(inc, x)
return future.result()
future = client.submit(f, 10)
assert future.result() == 11
@gen_cluster(client=True)
async def test_get_client_coroutine(c, s, a, b):
async def f():
# TODO: the existence of `await get_client()` implies the possibility
# of `async with get_client()` and that will kill the workers' client
# if you do that. We really don't want users to do that.
# https://github.com/dask/distributed/pull/6921/
client = get_client()
future = client.submit(inc, 10)
result = await future
return result
results = await c.run(f)
assert results == {a.address: 11, b.address: 11}
def test_get_client_coroutine_sync(client, s, a, b):
async def f():
client = await get_client()
future = client.submit(inc, 10)
result = await future
return result
results = client.run(f)
assert results == {a["address"]: 11, b["address"]: 11}
@gen_cluster()
async def test_global_workers(s, a, b):
n = len(Worker._instances)
w = first(Worker._instances)
assert w is a or w is b
@pytest.mark.skipif(WINDOWS, reason="num_fds not supported on windows")
@gen_cluster(nthreads=[])
async def test_worker_fds(s):
proc = psutil.Process()
before = psutil.Process().num_fds()
async with Worker(s.address):
assert proc.num_fds() > before
while proc.num_fds() > before:
await asyncio.sleep(0.01)
@gen_cluster(nthreads=[])
async def test_service_hosts_match_worker(s):
async with Worker(s.address, host="tcp://0.0.0.0") as w:
sock = first(w.http_server._sockets.values())
assert sock.getsockname()[0] in ("::", "0.0.0.0")
async with Worker(
s.address, host="tcp://127.0.0.1", dashboard_address="0.0.0.0:0"
) as w:
sock = first(w.http_server._sockets.values())
assert sock.getsockname()[0] in ("::", "0.0.0.0")
async with Worker(s.address, host="tcp://127.0.0.1") as w:
sock = first(w.http_server._sockets.values())
assert sock.getsockname()[0] in ("::", "0.0.0.0")
# See what happens with e.g. `dask worker --listen-address tcp://:8811`
async with Worker(s.address, host="") as w:
sock = first(w.http_server._sockets.values())
assert sock.getsockname()[0] in ("::", "0.0.0.0")
# Address must be a connectable address. 0.0.0.0 is not!
address_all = w.address.rsplit(":", 1)[0]
assert address_all in ("tcp://[::1]", "tcp://127.0.0.1")
# Check various malformed IPv6 addresses
# Since these hostnames get passed to distributed.comm.address_from_user_args,
# bracketing is mandatory for IPv6.
with pytest.raises(ValueError) as exc:
async with Worker(s.address, host="::") as w:
pass
assert "bracketed" in str(exc)
with pytest.raises(ValueError) as exc:
async with Worker(s.address, host="tcp://::1") as w:
pass
assert "bracketed" in str(exc)
@gen_cluster(nthreads=[])
async def test_start_services(s):
async with Worker(s.address, dashboard_address=1234) as w:
assert w.http_server.port == 1234
@gen_test()
async def test_scheduler_file():
with tmpfile() as fn:
async with Scheduler(scheduler_file=fn, dashboard_address=":0") as s:
async with Worker(scheduler_file=fn) as w:
assert set(s.workers) == {w.address}
@gen_cluster(client=True)
async def test_scheduler_delay(c, s, a, b):
old = a.scheduler_delay
assert abs(a.scheduler_delay) < 0.6
assert abs(b.scheduler_delay) < 0.6
await asyncio.sleep(a.periodic_callbacks["heartbeat"].callback_time / 1000 + 0.6)
assert a.scheduler_delay != old
@pytest.mark.flaky(reruns=10, reruns_delay=5)
@gen_cluster(
client=True,
config={
"distributed.worker.profile.enabled": True,
},
)
async def test_statistical_profiling(c, s, a, b):
futures = c.map(slowinc, range(10), delay=0.1)
await wait(futures)
profile = a.profile_keys["slowinc"]
assert profile["count"]
@pytest.mark.slow
@nodebug
@gen_cluster(
client=True,
timeout=30,
config={
"distributed.worker.profile.enabled": True,
"distributed.worker.profile.interval": "1ms",
"distributed.worker.profile.cycle": "100ms",
},
)
async def test_statistical_profiling_2(c, s, a, b):
da = pytest.importorskip("dask.array")
while True:
x = da.random.random(1000000, chunks=(10000,))
y = (x + x * 2) - x.sum().persist()
await wait(y)
profile = await a.get_profile()
text = str(profile)
if profile["count"] and "sum" in text and "random" in text:
break
@gen_cluster(
client=True,
config={
"distributed.worker.profile.enabled": True,
"distributed.worker.profile.cycle": "100ms",
},
)
async def test_statistical_profiling_cycle(c, s, a, b):
futures = c.map(slowinc, range(20), delay=0.05)
await wait(futures)
await asyncio.sleep(0.01)
end = time()
assert len(a.profile_history) > 3
x = await a.get_profile(start=time() + 10, stop=time() + 20)
assert not x["count"]
x = await a.get_profile(start=0, stop=time() + 10)
recent = a.profile_recent["count"]
actual = sum(p["count"] for _, p in a.profile_history) + a.profile_recent["count"]
x2 = await a.get_profile(start=0, stop=time() + 10)
assert x["count"] <= actual <= x2["count"]
y = await a.get_profile(start=end - 0.300, stop=time())
assert 0 < y["count"] <= x["count"]
@gen_cluster(client=True)
async def test_get_current_task(c, s, a, b):
def some_name():
return get_worker().get_current_task()
result = await c.submit(some_name)
assert result.startswith("some_name")
@gen_cluster(nthreads=[])
async def test_deque_handler(s):
from distributed.worker import logger
async with Worker(s.address) as w:
deque_handler = w._deque_handler
logger.info("foo456")
assert deque_handler.deque
msg = deque_handler.deque[-1]
assert "distributed.worker" in deque_handler.format(msg)
assert any(msg.msg == "foo456" for msg in deque_handler.deque)
def test_get_worker_name(client):
def f():
get_client().submit(inc, 1).result()
client.run(f)
def func(dask_scheduler):
return list(dask_scheduler.clients)
start = time()
while not any("worker" in n for n in client.run_on_scheduler(func)):
sleep(0.1)
assert time() < start + 10
@gen_cluster(nthreads=[], client=True)
async def test_scheduler_address_config(c, s):
with dask.config.set({"scheduler-address": s.address}):
async with Worker() as w:
assert w.scheduler.address == s.address
@pytest.mark.xfail(reason="very high flakiness")
@pytest.mark.slow
@gen_cluster(client=True)
async def test_wait_for_outgoing(c, s, a, b):
np = pytest.importorskip("numpy")
x = np.random.random(10000000)
future = await c.scatter(x, workers=a.address)
y = c.submit(inc, future, workers=b.address)
await wait(y)
assert len(b.transfer_incoming_log) == len(a.transfer_outgoing_log) == 1
bb = b.transfer_incoming_log[0]["duration"]
aa = a.outgoing_transfer_log[0]["duration"]
ratio = aa / bb
assert 1 / 3 < ratio < 3
@pytest.mark.skipif(not LINUX, reason="Need 127.0.0.2 to mean localhost")
@gen_cluster(
nthreads=[("127.0.0.1", 1), ("127.0.0.1", 1), ("127.0.0.2", 1)],
client=True,
config=NO_AMM,
)
async def test_prefer_gather_from_local_address(c, s, w1, w2, w3):
x = await c.scatter(123, workers=[w1.address, w3.address], broadcast=True)
y = c.submit(inc, x, workers=[w2.address])
await wait(y)
assert any(d["who"] == w2.address for d in w1.transfer_outgoing_log)
assert not any(d["who"] == w2.address for d in w3.transfer_outgoing_log)
@gen_cluster(
client=True,
nthreads=[("127.0.0.1", 1)] * 20,
config={"distributed.worker.connections.incoming": 1},
)
async def test_avoid_oversubscription(c, s, *workers):
np = pytest.importorskip("numpy")
x = c.submit(np.random.random, 1000000, workers=[workers[0].address])
await wait(x)
futures = [c.submit(len, x, pure=False, workers=[w.address]) for w in workers[1:]]
await wait(futures)
# Original worker not responsible for all transfers
assert len(workers[0].transfer_outgoing_log) < len(workers) - 2
# Some other workers did some work
assert len([w for w in workers if len(w.transfer_outgoing_log) > 0]) >= 3
@gen_cluster(client=True, worker_kwargs={"metrics": {"my_port": lambda w: w.port}})
async def test_custom_metrics(c, s, a, b):
assert s.workers[a.address].metrics["my_port"] == a.port
assert s.workers[b.address].metrics["my_port"] == b.port
@gen_cluster(client=True)
async def test_register_worker_callbacks(c, s, a, b):
# plugin function to run
def mystartup(dask_worker):
dask_worker.init_variable = 1
def mystartup2():
import os
os.environ["MY_ENV_VALUE"] = "WORKER_ENV_VALUE"
return "Env set."
# Check that plugin function has been run
def test_import(dask_worker):
return hasattr(dask_worker, "init_variable")
# and dask_worker.init_variable == 1
def test_startup2():
import os
return os.getenv("MY_ENV_VALUE", None) == "WORKER_ENV_VALUE"
# Nothing has been run yet
result = await c.run(test_import)
assert list(result.values()) == [False] * 2
result = await c.run(test_startup2)
assert list(result.values()) == [False] * 2
# Start a worker and check that startup is not run
async with Worker(s.address) as worker:
result = await c.run(test_import, workers=[worker.address])
assert list(result.values()) == [False]
# Add a plugin function
response = await c.register_worker_callbacks(setup=mystartup)
assert len(response) == 2
# Check it has been ran on existing worker
result = await c.run(test_import)
assert list(result.values()) == [True] * 2
# Start a worker and check it is run on it
async with Worker(s.address) as worker:
result = await c.run(test_import, workers=[worker.address])
assert list(result.values()) == [True]
# Register another plugin function
response = await c.register_worker_callbacks(setup=mystartup2)
assert len(response) == 2
# Check it has been run
result = await c.run(test_startup2)
assert list(result.values()) == [True] * 2
# Start a worker and check it is run on it
async with Worker(s.address) as worker:
result = await c.run(test_import, workers=[worker.address])
assert list(result.values()) == [True]
result = await c.run(test_startup2, workers=[worker.address])
assert list(result.values()) == [True]
@gen_cluster(client=True)
async def test_register_worker_callbacks_err(c, s, a, b):
with pytest.raises(ZeroDivisionError):
await c.register_worker_callbacks(setup=lambda: 1 / 0)
@gen_cluster(nthreads=[])
async def test_local_directory(s, tmp_path):
with dask.config.set(temporary_directory=str(tmp_path)):
async with Worker(s.address) as w:
assert w.local_directory.startswith(str(tmp_path))
assert "dask-worker-space" in w.local_directory
@gen_cluster(nthreads=[])
async def test_local_directory_make_new_directory(s, tmp_path):
async with Worker(s.address, local_directory=str(tmp_path / "foo" / "bar")) as w:
assert w.local_directory.startswith(str(tmp_path))
assert "foo" in w.local_directory
assert "dask-worker-space" in w.local_directory
@pytest.mark.skipif(not LINUX, reason="Need 127.0.0.2 to mean localhost")
@gen_cluster(nthreads=[], client=True)
async def test_host_address(c, s):
async with Worker(s.address, host="127.0.0.2") as w:
assert "127.0.0.2" in w.address
async with Nanny(s.address, host="127.0.0.3") as n:
assert "127.0.0.3" in n.address
assert "127.0.0.3" in n.worker_address
@pytest.mark.parametrize("Worker", [Worker, Nanny])
@gen_test()
async def test_interface_async(Worker):
from distributed.utils import get_ip_interface
psutil = pytest.importorskip("psutil")
if_names = sorted(psutil.net_if_addrs())
for if_name in if_names:
try:
ipv4_addr = get_ip_interface(if_name)
except ValueError:
pass
else:
if ipv4_addr == "127.0.0.1":
break
else:
pytest.skip(
"Could not find loopback interface. "
"Available interfaces are: %s." % (if_names,)
)
async with Scheduler(dashboard_address=":0", interface=if_name) as s:
assert s.address.startswith("tcp://127.0.0.1")
async with Worker(s.address, interface=if_name) as w:
assert w.address.startswith("tcp://127.0.0.1")
assert w.ip == "127.0.0.1"
async with Client(s.address, asynchronous=True) as c:
info = c.scheduler_info()
assert "tcp://127.0.0.1" in info["address"]
assert all("127.0.0.1" == d["host"] for d in info["workers"].values())
@pytest.mark.gpu
@pytest.mark.parametrize("Worker", [Worker, Nanny])
@gen_test()
async def test_protocol_from_scheduler_address(ucx_loop, Worker):
pytest.importorskip("ucp")
async with Scheduler(protocol="ucx", dashboard_address=":0") as s:
assert s.address.startswith("ucx://")
async with Worker(s.address) as w:
assert w.address.startswith("ucx://")
async with Client(s.address, asynchronous=True) as c:
info = c.scheduler_info()
assert info["address"].startswith("ucx://")
@gen_test()
async def test_host_uses_scheduler_protocol(monkeypatch):
# Ensure worker uses scheduler's protocol to determine host address, not the default scheme
# See https://github.com/dask/distributed/pull/4883
from distributed.comm.tcp import TCPBackend
class BadBackend(TCPBackend):
def get_address_host(self, loc):
raise ValueError("asdf")
monkeypatch.setitem(backends, "foo", BadBackend())
with dask.config.set({"distributed.comm.default-scheme": "foo"}):
async with Scheduler(protocol="tcp", dashboard_address=":0") as s:
async with Worker(s.address):
# Ensure that worker is able to properly start up
# without BadBackend.get_address_host raising a ValueError
pass
@pytest.mark.parametrize("Worker", [Worker, Nanny])
@gen_test()
async def test_worker_listens_on_same_interface_by_default(Worker):
async with Scheduler(host="localhost", dashboard_address=":0") as s:
assert s.ip in {"127.0.0.1", "localhost"}
async with Worker(s.address) as w:
assert s.ip == w.ip
def assert_amm_transfer_story(key: str, w_from: Worker, w_to: Worker) -> None:
"""Test that an in-memory key was transferred from worker w_from to worker w_to by
the Active Memory Manager and it was not recalculated on w_to
"""
assert_story(
w_to.state.story(key),
[
(key, "fetch", "flight", "flight", {}),
("request-dep", w_from.address, lambda set_: key in set_),
("receive-dep", w_from.address, lambda set_: key in set_),
(key, "put-in-memory"),
(key, "flight", "memory", "memory", {}),
],
strict=False,
)
assert key in w_to.data
# The key may or may not still be in w_from.data, depending if the AMM had the
# chance to run a second time after the copy was successful.
@pytest.mark.slow
@gen_cluster(client=True)
async def test_close_gracefully(c, s, a, b):
futures = c.map(slowinc, range(200), delay=0.1, workers=[b.address])
# Note: keys will appear in b.data several milliseconds before they switch to
# status=memory in s.tasks. It's important to sample the in-memory keys from the
# scheduler side, because those that the scheduler thinks are still processing won't
# be replicated by retire_workers().
while True:
mem = {k for k, ts in s.tasks.items() if ts.state == "memory"}
if len(mem) >= 8 and any(
ts.state == "executing" for ts in b.state.tasks.values()
):
break
await asyncio.sleep(0.01)
assert any(ts for ts in b.state.tasks.values() if ts.state == "executing")
with captured_logger("distributed.worker") as logger:
await b.close_gracefully(reason="foo")
assert "Reason: foo" in logger.getvalue()
assert b.status == Status.closed
assert b.address not in s.workers
# All tasks that were in memory in b have been copied over to a;
# they have not been recomputed
for key in mem:
assert_amm_transfer_story(key, b, a)
@pytest.mark.parametrize("sync", [False, pytest.param(True, marks=[pytest.mark.slow])])
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_close_while_executing(c, s, a, sync):
ev = Event()
if sync:
def f(ev):
ev.set()
sleep(2)
else:
async def f(ev):
await ev.set()
await asyncio.Event().wait() # Block indefinitely
f1 = c.submit(f, ev, key="f1")
await ev.wait()
task = next(
task for task in asyncio.all_tasks() if "execute(f1)" in task.get_name()
)
await a.close()
assert task.done()
assert s.tasks["f1"].state in ("queued", "no-worker")
@pytest.mark.slow
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_close_async_task_handles_cancellation(c, s, a):
ev = Event()
async def f(ev):
await ev.set()
try:
await asyncio.Event().wait() # Block indefinitely
except asyncio.CancelledError:
await asyncio.Event().wait() # Ignore the first cancel()
f1 = c.submit(f, ev, key="f1")
await ev.wait()
task = next(
task for task in asyncio.all_tasks() if "execute(f1)" in task.get_name()
)
start = time()
with captured_logger(
"distributed.worker.state_machine", level=logging.ERROR
) as logger:
await a.close(timeout=1)
assert "Failed to cancel asyncio task" in logger.getvalue()
assert time() - start < 5
assert not task.cancelled()
assert s.tasks["f1"].state in ("queued", "no-worker")
task.cancel()
await asyncio.wait({task})
@pytest.mark.slow
@gen_cluster(client=True, nthreads=[("", 1)], timeout=10)
async def test_lifetime(c, s, a):
# Note: test was occasionally failing with lifetime="1 seconds"
async with Worker(s.address, lifetime="2 seconds") as b:
futures = c.map(slowinc, range(200), delay=0.1, workers=[b.address])
# Note: keys will appear in b.data several milliseconds before they switch to
# status=memory in s.tasks. It's important to sample the in-memory keys from the
# scheduler side, because those that the scheduler thinks are still processing
# won't be replicated by retire_workers().
while True:
mem = {k for k, ts in s.tasks.items() if ts.state == "memory"}
if len(mem) >= 8:
break
await asyncio.sleep(0.01)
assert b.status == Status.running
assert not a.data
while b.status != Status.closed:
await asyncio.sleep(0.01)
# All tasks that were in memory in b have been copied over to a;
# they have not been recomputed
for key in mem:
assert_amm_transfer_story(key, b, a)
@gen_cluster(worker_kwargs={"lifetime": "10s", "lifetime_stagger": "2s"})
async def test_lifetime_stagger(s, a, b):
assert a.lifetime != b.lifetime
assert 8 <= a.lifetime <= 12
assert 8 <= b.lifetime <= 12
@gen_cluster(nthreads=[])
async def test_bad_metrics(s):
def bad_metric(w):
raise Exception("Hello")
async with Worker(s.address, metrics={"bad": bad_metric}) as w:
assert "bad" not in s.workers[w.address].metrics
@gen_cluster(nthreads=[])
async def test_bad_startup(s):
"""Bad startup functions do not cause the Worker to fail"""
def bad_startup(w):
raise Exception("Hello")
async with Worker(s.address, startup_information={"bad": bad_startup}):
pass
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_pip_install(c, s, a):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.INFO
) as logger:
mocked = mock.Mock()
mocked.configure_mock(
**{"communicate.return_value": (b"", b""), "wait.return_value": 0}
)
with mock.patch(
"distributed.diagnostics.plugin.subprocess.Popen", return_value=mocked
) as Popen:
await c.register_worker_plugin(
PipInstall(packages=["requests"], pip_options=["--upgrade"])
)
assert Popen.call_count == 1
args = Popen.call_args[0][0]
assert "python" in args[0]
assert args[1:] == ["-m", "pip", "install", "--upgrade", "requests"]
logs = logger.getvalue()
assert "pip installing" in logs
assert "failed" not in logs
assert "restart" not in logs
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_conda_install(c, s, a):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.INFO
) as logger:
run_command_mock = mock.Mock(name="run_command_mock")
run_command_mock.configure_mock(return_value=(b"", b"", 0))
module_mock = mock.Mock(name="conda_cli_python_api_mock")
module_mock.run_command = run_command_mock
module_mock.Commands.INSTALL = "INSTALL"
with mock.patch.dict("sys.modules", {"conda.cli.python_api": module_mock}):
await c.register_worker_plugin(
CondaInstall(packages=["requests"], conda_options=["--update-deps"])
)
assert run_command_mock.call_count == 1
command = run_command_mock.call_args[0][0]
assert command == "INSTALL"
arguments = run_command_mock.call_args[0][1]
assert arguments == ["--update-deps", "requests"]
logs = logger.getvalue()
assert "conda installing" in logs
assert "failed" not in logs
assert "restart" not in logs
@gen_cluster(client=True, nthreads=[("", 2), ("", 2)])
async def test_pip_install_fails(c, s, a, b):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.ERROR
) as logger:
mocked = mock.Mock()
mocked.configure_mock(
**{
"communicate.return_value": (
b"",
b"Could not find a version that satisfies the requirement not-a-package",
),
"wait.return_value": 1,
}
)
with mock.patch(
"distributed.diagnostics.plugin.subprocess.Popen", return_value=mocked
) as Popen:
with pytest.raises(RuntimeError):
await c.register_worker_plugin(PipInstall(packages=["not-a-package"]))
assert Popen.call_count == 1
logs = logger.getvalue()
assert "install failed" in logs
assert "not-a-package" in logs
@gen_cluster(client=True, nthreads=[("", 2), ("", 2)])
async def test_conda_install_fails_when_conda_not_found(c, s, a, b):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.ERROR
) as logger:
with mock.patch.dict("sys.modules", {"conda": None}):
with pytest.raises(RuntimeError):
await c.register_worker_plugin(CondaInstall(packages=["not-a-package"]))
logs = logger.getvalue()
assert "install failed" in logs
assert "conda could not be found" in logs
@gen_cluster(client=True, nthreads=[("", 2), ("", 2)])
async def test_conda_install_fails_when_conda_raises(c, s, a, b):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.ERROR
) as logger:
run_command_mock = mock.Mock(name="run_command_mock")
run_command_mock.configure_mock(side_effect=RuntimeError)
module_mock = mock.Mock(name="conda_cli_python_api_mock")
module_mock.run_command = run_command_mock
module_mock.Commands.INSTALL = "INSTALL"
with mock.patch.dict("sys.modules", {"conda.cli.python_api": module_mock}):
with pytest.raises(RuntimeError):
await c.register_worker_plugin(CondaInstall(packages=["not-a-package"]))
assert run_command_mock.call_count == 1
logs = logger.getvalue()
assert "install failed" in logs
@gen_cluster(client=True, nthreads=[("", 2), ("", 2)])
async def test_conda_install_fails_on_returncode(c, s, a, b):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.ERROR
) as logger:
run_command_mock = mock.Mock(name="run_command_mock")
run_command_mock.configure_mock(return_value=(b"", b"", 1))
module_mock = mock.Mock(name="conda_cli_python_api_mock")
module_mock.run_command = run_command_mock
module_mock.Commands.INSTALL = "INSTALL"
with mock.patch.dict("sys.modules", {"conda.cli.python_api": module_mock}):
with pytest.raises(RuntimeError):
await c.register_worker_plugin(CondaInstall(packages=["not-a-package"]))
assert run_command_mock.call_count == 1
logs = logger.getvalue()
assert "install failed" in logs
class StubInstall(PackageInstall):
INSTALLER = "stub"
def __init__(self, packages: list[str], restart: bool = False):
super().__init__(packages=packages, restart=restart)
def install(self) -> None:
pass
@gen_cluster(client=True, nthreads=[("", 1), ("", 1)])
async def test_package_install_installs_once_with_multiple_workers(c, s, a, b):
with captured_logger(
"distributed.diagnostics.plugin", level=logging.INFO
) as logger:
install_mock = mock.Mock(name="install")
with mock.patch.object(StubInstall, "install", install_mock):
await c.register_worker_plugin(
StubInstall(
packages=["requests"],
)
)
assert install_mock.call_count == 1
logs = logger.getvalue()
assert "already been installed" in logs
@gen_cluster(client=True, nthreads=[("", 1)], Worker=Nanny)
async def test_package_install_restarts_on_nanny(c, s, a):
(addr,) = s.workers
await c.register_worker_plugin(
StubInstall(
packages=["requests"],
restart=True,
)
)
# Wait until the worker is restarted
while len(s.workers) != 1 or set(s.workers) == {addr}:
await asyncio.sleep(0.01)
class FailingInstall(PackageInstall):
INSTALLER = "fail"
def __init__(self, packages: list[str], restart: bool = False):
super().__init__(packages=packages, restart=restart)
def install(self) -> None:
raise RuntimeError()
@gen_cluster(client=True, nthreads=[("", 1)], Worker=Nanny)
async def test_package_install_failing_does_not_restart_on_nanny(c, s, a):
(addr,) = s.workers
with pytest.raises(RuntimeError):
await c.register_worker_plugin(
FailingInstall(
packages=["requests"],
restart=True,
)
)
# Nanny does not restart
assert a.status is Status.running
assert set(s.workers) == {addr}
@gen_cluster(nthreads=[])
async def test_update_latency(s):
async with Worker(s.address) as w:
original = w.latency
await w.heartbeat()
assert original != w.latency
if w.digests is not None:
assert w.digests["latency"].size() > 0
@pytest.mark.slow
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_workerstate_executing(c, s, a):
ws = s.workers[a.address]
# Initially there are no active tasks
assert not ws.executing
# Submit a task and ensure the WorkerState is updated with the task
# it's executing
f = c.submit(slowinc, 1, delay=3)
while not ws.executing:
assert f.status == "pending"
await asyncio.sleep(0.01)
assert s.tasks[f.key] in ws.executing
await f
@gen_cluster(nthreads=[("", 1)])
async def test_shutdown_on_scheduler_comm_closed(s, a):
with captured_logger("distributed.core", level=logging.INFO) as logger:
# Temporary network disconnect
s.stream_comms[a.address].abort()
await a.finished()
assert a.status == Status.closed
assert not s.workers
assert not s.stream_comms
assert f"Connection to {s.address} has been closed" in logger.getvalue()
@gen_cluster(nthreads=[])
async def test_heartbeat_comm_closed(s, monkeypatch):
with captured_logger("distributed.worker", level=logging.WARNING) as logger:
def bad_heartbeat_worker(*args, **kwargs):
raise CommClosedError()
async with Worker(s.address) as w:
# Trigger CommClosedError during worker heartbeat
monkeypatch.setattr(w.scheduler, "heartbeat_worker", bad_heartbeat_worker)
await w.heartbeat()
assert w.status == Status.running
logs = logger.getvalue()
assert "Failed to communicate with scheduler during heartbeat" in logs
assert "Traceback" in logs
@gen_cluster(nthreads=[("", 1)], worker_kwargs={"heartbeat_interval": "100s"})
async def test_heartbeat_missing(s, a, monkeypatch):
async def missing_heartbeat_worker(*args, **kwargs):
return {"status": "missing"}
with captured_logger("distributed.worker", level=logging.WARNING) as wlogger:
monkeypatch.setattr(a.scheduler, "heartbeat_worker", missing_heartbeat_worker)
await a.heartbeat()
assert a.status == Status.closed
assert "Scheduler was unaware of this worker" in wlogger.getvalue()
while s.workers:
await asyncio.sleep(0.01)
@gen_cluster(nthreads=[("", 1)], worker_kwargs={"heartbeat_interval": "100s"})
async def test_heartbeat_missing_real_cluster(s, a):
# The idea here is to create a situation where `s.workers[a.address]`,
# doesn't exist, but the worker is not yet closed and can still heartbeat.
# Ideally, this state would be impossible, and this test would be removed,
# and the `status: missing` handling would be replaced with an assertion error.
# However, `Scheduler.remove_worker` and `Worker.close` both currently leave things
# in degenerate, half-closed states while they're running (and yielding control
# via `await`).
# When https://github.com/dask/distributed/issues/6390 is fixed, this should no
# longer be possible.
assumption_msg = "Test assumptions have changed. Race condition may have been fixed; this test may be removable."
with captured_logger(
"distributed.worker", level=logging.WARNING
) as wlogger, captured_logger(
"distributed.scheduler", level=logging.WARNING
) as slogger:
with freeze_batched_send(s.stream_comms[a.address]):
await s.remove_worker(a.address, stimulus_id="foo")
assert not s.workers
# The scheduler has removed the worker state, but the close message has
# not reached the worker yet.
assert a.status == Status.running, assumption_msg
assert a.periodic_callbacks["heartbeat"].is_running(), assumption_msg
# The heartbeat PeriodicCallback is still running, so one _could_ fire
# before the `op: close` message reaches the worker. We simulate that explicitly.
await a.heartbeat()
# The heartbeat receives a `status: missing` from the scheduler, so it
# closes the worker. Heartbeats aren't sent over batched comms, so
# `freeze_batched_send` doesn't affect them.
assert a.status == Status.closed
assert "Scheduler was unaware of this worker" in wlogger.getvalue()
assert "Received heartbeat from unregistered worker" in slogger.getvalue()
assert not s.workers
@gen_cluster(
client=True,
nthreads=[("", 1)],
Worker=Nanny,
worker_kwargs={"heartbeat_interval": "1ms"},
)
async def test_heartbeat_missing_restarts(c, s, n):
old_heartbeat_handler = s.handlers["heartbeat_worker"]
s.handlers["heartbeat_worker"] = lambda *args, **kwargs: {"status": "missing"}
assert n.process
await n.process.stopped.wait()
assert not s.workers
s.handlers["heartbeat_worker"] = old_heartbeat_handler
await n.process.running.wait()
assert n.status == Status.running
await c.wait_for_workers(1)
@pytest.mark.skipif(not os.access("/", os.W_OK),
reason="skip if we have elevated privileges")
@gen_cluster(nthreads=[])
async def test_bad_local_directory(s):
try:
async with Worker(s.address, local_directory="/not/a/valid-directory"):
pass
except OSError:
# On Linux: [Errno 13] Permission denied: '/not'
# On MacOSX: [Errno 30] Read-only file system: '/not'
pass
else:
assert WINDOWS
assert not any("error" in log for log in s.get_logs())
@gen_cluster(client=True, nthreads=[])
async def test_taskstate_metadata(c, s):
async with Worker(s.address) as a:
await c.register_worker_plugin(TaskStateMetadataPlugin())
f = c.submit(inc, 1)
await f
ts = a.state.tasks[f.key]
assert "start_time" in ts.metadata
assert "stop_time" in ts.metadata
assert ts.metadata["stop_time"] > ts.metadata["start_time"]
# Check that Scheduler TaskState.metadata was also updated
assert s.tasks[f.key].metadata == ts.metadata
@gen_cluster(client=True, nthreads=[])
async def test_executor_offload(c, s, monkeypatch):
class SameThreadClass:
def __getstate__(self):
return ()
def __setstate__(self, state):
self._thread_ident = threading.get_ident()
return self
monkeypatch.setattr("distributed.worker.OFFLOAD_THRESHOLD", 1)
async with Worker(s.address, executor="offload") as w:
from distributed.utils import _offload_executor
assert w.executor is _offload_executor
x = SameThreadClass()
def f(x):
return threading.get_ident() == x._thread_ident
assert await c.submit(f, x)
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_story(c, s, w):
future = c.submit(inc, 1)
await future
ts = w.state.tasks[future.key]
assert ts.state in str(w.state.story(ts))
assert w.state.story(ts) == w.state.story(ts.key)
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_stimulus_story(c, s, a):
# Test that substrings aren't matched by stimulus_story()
f = c.submit(inc, 1, key="f")
f1 = c.submit(lambda: "foo", key="f1")
f2 = c.submit(inc, f1, key="f2") # This will fail
await wait([f, f1, f2])
story = a.state.stimulus_story("f1", "f2")
assert len(story) == 4
assert isinstance(story[0], ComputeTaskEvent)
assert story[0].key == "f1"
assert story[0].run_spec == SerializedTask(task=None) # Not logged
assert isinstance(story[1], ExecuteSuccessEvent)
assert story[1].key == "f1"
assert story[1].value is None # Not logged
assert story[1].handled >= story[0].handled
assert isinstance(story[2], ComputeTaskEvent)
assert story[2].key == "f2"
assert story[2].who_has == {"f1": (a.address,)}
assert story[2].run_spec == SerializedTask(task=None) # Not logged
assert story[2].handled >= story[1].handled
assert isinstance(story[3], ExecuteFailureEvent)
assert story[3].key == "f2"
assert story[3].handled >= story[2].handled
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_worker_descopes_data(c, s, a):
"""Test that data is released on the worker:
1. when it's the output of a successful task
2. when it's the input of a failed task
3. when it's a local variable in the frame of a failed task
4. when it's embedded in the exception of a failed task
"""
class C:
instances = weakref.WeakSet()
def __init__(self):
C.instances.add(self)
def f(x):
y = C()
raise Exception(x, y)
f1 = c.submit(C, key="f1")
f2 = c.submit(f, f1, key="f2")
await wait([f2])
assert type(a.data["f1"]) is C
del f1
del f2
while a.data:
await asyncio.sleep(0.01)
with profile.lock:
gc.collect()
assert not C.instances
# @pytest.mark.slow
@gen_cluster(client=True, config=NO_AMM)
async def test_gather_dep_one_worker_always_busy(c, s, a, b):
# Ensure that both dependencies for H are on another worker than H itself.
# The worker where the dependencies are on is then later blocked such that
# the data cannot be fetched
# In the past it was important that there is more than one key on the
# worker. This should be kept to avoid any edge case specific to one
f = c.submit(inc, 1, key="f", workers=[a.address])
g = c.submit(inc, 2, key="g", workers=[a.address])
await wait([f, g])
assert set(a.state.tasks) == {"f", "g"}
# We will block A for any outgoing communication. This simulates an
# overloaded worker which will always return "busy" for get_data requests,
# effectively blocking H indefinitely
a.transfer_outgoing_count = 10000000
h = c.submit(add, f, g, key="h", workers=[b.address])
await wait_for_state(h.key, "waiting", b)
assert b.state.tasks[f.key].state in ("flight", "fetch")
assert b.state.tasks[g.key].state in ("flight", "fetch")
with pytest.raises(asyncio.TimeoutError):
await h.result(timeout=0.8)
story = b.state.story("busy-gather")
# 1 busy response straight away, followed by 1 retry every 150ms for 800ms.
# The requests for b and g are clustered together in single messages.
# We need to be very lax in measuring as PeriodicCallback+network comms have been
# observed on CI to occasionally lag behind by several hundreds of ms.
assert 2 <= len(story) <= 8
async with Worker(s.address, name="x") as x:
# We "scatter" the data to another worker which is able to serve this data.
# In reality this could be another worker which fetched this dependency and got
# through to A or another worker executed the task using work stealing or any
# other. To avoid cross effects, we'll just put the data onto the worker
# ourselves. Note that doing so means that B won't know about the existence of
# the extra replicas until it takes the initiative to invoke scheduler.who_has.
x.update_data({"f": 2, "g": 3})
assert await h == 5
@pytest.mark.skipif(not LINUX, reason="Need 127.0.0.2 to mean localhost")
@gen_cluster(
client=True,
nthreads=[("127.0.0.1", 1)] * 2 + [("127.0.0.2", 2)] * 10, # type: ignore
config=NO_AMM,
)
async def test_gather_dep_local_workers_first(c, s, a, lw, *rws):
f = (
await c.scatter(
{"f": 1}, workers=[w.address for w in (lw,) + rws], broadcast=True
)
)["f"]
g = c.submit(inc, f, key="g", workers=[a.address])
assert await g == 2
assert_story(a.state.story("f"), [("receive-dep", lw.address, {"f"})])
@pytest.mark.skipif(not LINUX, reason="Need 127.0.0.2 to mean localhost")
@gen_cluster(
client=True,
nthreads=[("127.0.0.2", 1)] + [("127.0.0.1", 1)] * 10, # type: ignore
config=NO_AMM,
)
async def test_gather_dep_from_remote_workers_if_all_local_workers_are_busy(
c, s, rw, a, *lws
):
f = (
await c.scatter(
{"f": 1}, workers=[w.address for w in (rw,) + lws], broadcast=True
)
)["f"]
for w in lws:
w.transfer_outgoing_count = 10000000
g = c.submit(inc, f, key="g", workers=[a.address])
assert await g == 2
# Tried fetching from each local worker exactly once before falling back to the
# remote worker
assert sorted(ev[1] for ev in a.state.story("busy-gather")) == sorted(
w.address for w in lws
)
assert_story(a.state.story("receive-dep"), [("receive-dep", rw.address, {"f"})])
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_worker_client_uses_default_no_close(c, s, a):
"""
If a default client is available in the process, the worker will pick this
one and will not close it if it is closed
"""
assert not Worker._initialized_clients
assert default_client() is c
existing_client = c.id
def get_worker_client_id():
def_client = get_client()
return def_client.id
worker_client = await c.submit(get_worker_client_id)
assert worker_client == existing_client
assert not Worker._initialized_clients
await a.close()
assert len(Client._instances) == 1
assert c.status == "running"
c_def = default_client()
assert c is c_def
@gen_cluster(nthreads=[("127.0.0.1", 1)])
async def test_worker_client_closes_if_created_on_worker_one_worker(s, a):
async with Client(s.address, set_as_default=False, asynchronous=True) as c:
with pytest.raises(ValueError):
default_client()
def get_worker_client_id():
def_client = get_client()
return def_client.id
new_client_id = await c.submit(get_worker_client_id)
default_client_id = await c.submit(get_worker_client_id)
assert new_client_id != c.id
assert new_client_id == default_client_id
new_client = default_client()
assert new_client_id == new_client.id
assert new_client.status == "running"
# If a worker closes, all clients created on it should close as well
await a.close()
assert new_client.status == "closed"
assert len(Client._instances) == 2
assert c.status == "running"
with pytest.raises(ValueError):
default_client()
@gen_cluster()
async def test_worker_client_closes_if_created_on_worker_last_worker_alive(s, a, b):
async with Client(s.address, set_as_default=False, asynchronous=True) as c:
with pytest.raises(ValueError):
default_client()
def get_worker_client_id():
def_client = get_client()
return def_client.id
new_client_id = await c.submit(get_worker_client_id, workers=[a.address])
default_client_id = await c.submit(get_worker_client_id, workers=[a.address])
default_client_id_b = await c.submit(get_worker_client_id, workers=[b.address])
assert not b._comms
assert new_client_id != c.id
assert new_client_id == default_client_id
assert new_client_id == default_client_id_b
new_client = default_client()
assert new_client_id == new_client.id
assert new_client.status == "running"
# We'll close A. This should *not* close the client since the client is also used by B
await a.close()
assert new_client.status == "running"
client_id_b_after = await c.submit(get_worker_client_id, workers=[b.address])
assert client_id_b_after == default_client_id_b
assert len(Client._instances) == 2
await b.close()
assert new_client.status == "closed"
assert c.status == "running"
with pytest.raises(ValueError):
default_client()
@gen_cluster(client=True, nthreads=[])
async def test_multiple_executors(c, s):
def get_thread_name():
return threading.current_thread().name
async with Worker(
s.address,
nthreads=2,
executor={"foo": ThreadPoolExecutor(1, thread_name_prefix="Dask-Foo-Threads")},
):
futures = []
with dask.annotate(executor="default"):
futures.append(c.submit(get_thread_name, pure=False))
with dask.annotate(executor="foo"):
futures.append(c.submit(get_thread_name, pure=False))
default_result, gpu_result = await c.gather(futures)
assert "Dask-Default-Threads" in default_result
assert "Dask-Foo-Threads" in gpu_result
@gen_cluster(client=True)
async def test_bad_executor_annotation(c, s, a, b):
with dask.annotate(executor="bad"):
future = c.submit(inc, 1)
with pytest.raises(ValueError, match="Invalid executor 'bad'; expected one of: "):
await future
assert future.status == "error"
@gen_cluster(client=True)
async def test_process_executor(c, s, a, b):
with ProcessPoolExecutor() as e:
a.executors["processes"] = e
b.executors["processes"] = e
future = c.submit(os.getpid, pure=False)
assert (await future) == os.getpid()
with dask.annotate(executor="processes"):
future = c.submit(os.getpid, pure=False)
assert (await future) != os.getpid()
def kill_process():
import os
import signal
if WINDOWS:
# There's no SIGKILL on Windows
sig = signal.SIGTERM
else:
# With SIGTERM there may be several seconds worth of delay before the worker
# actually shuts down - particularly on slow CI. Use SIGKILL for instant
# termination.
sig = signal.SIGKILL
os.kill(os.getpid(), sig)
sleep(60) # Cope with non-instantaneous termination
@gen_cluster(nthreads=[("127.0.0.1", 1)], client=True)
async def test_process_executor_kills_process(c, s, a):
with ProcessPoolExecutor() as e:
a.executors["processes"] = e
with dask.annotate(executor="processes", retries=1):
future = c.submit(kill_process)
msg = "A child process terminated abruptly, the process pool is not usable anymore"
with pytest.raises(BrokenProcessPool, match=msg):
await future
with dask.annotate(executor="processes", retries=1):
future = c.submit(inc, 1)
# The process pool is now unusable and the worker is effectively dead
with pytest.raises(BrokenProcessPool, match=msg):
await future
def raise_exc():
raise RuntimeError("foo")
@gen_cluster(client=True)
async def test_process_executor_raise_exception(c, s, a, b):
with ProcessPoolExecutor() as e:
a.executors["processes"] = e
b.executors["processes"] = e
with dask.annotate(executor="processes", retries=1):
future = c.submit(raise_exc)
with pytest.raises(RuntimeError, match="foo"):
await future
@pytest.mark.gpu
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_gpu_executor(c, s, w):
if nvml.device_get_count() > 0:
e = w.executors["gpu"]
assert isinstance(e, distributed.threadpoolexecutor.ThreadPoolExecutor)
assert e._max_workers == 1
else:
assert "gpu" not in w.executors
async def assert_task_states_on_worker(
expected: dict[str, str], worker: Worker
) -> None:
active_exc = None
for _ in range(10):
try:
for dep_key, expected_state in expected.items():
assert dep_key in worker.state.tasks, (
worker.name,
dep_key,
worker.state.tasks,
)
dep_ts = worker.state.tasks[dep_key]
assert dep_ts.state == expected_state, (
worker.name,
dep_ts,
expected_state,
)
assert set(expected) == set(worker.state.tasks)
return
except AssertionError as exc:
active_exc = exc
await asyncio.sleep(0.1)
# If after a second the workers are not in equilibrium, they are broken
assert active_exc
raise active_exc
@gen_cluster(client=True, config=NO_AMM)
async def test_worker_state_error_release_error_last(c, s, a, b):
"""
Create a chain of tasks and err one of them. Then release tasks in a certain
order and ensure the tasks are released and/or kept in memory as appropriate
F -- RES (error)
/
/
G
Free error last
"""
def raise_exc(*args):
raise RuntimeError()
f = c.submit(inc, 1, workers=[a.address], key="f")
g = c.submit(inc, 1, workers=[b.address], key="g")
res = c.submit(raise_exc, f, g, workers=[a.address])
with pytest.raises(RuntimeError):
await res
# Nothing bad happened on B, therefore B should hold on to G
assert len(b.state.tasks) == 1
assert g.key in b.state.tasks
# A raised the exception therefore we should hold on to the erroneous task
assert res.key in a.state.tasks
ts = a.state.tasks[res.key]
assert ts.state == "error"
expected_states = {
# A was instructed to compute this result and we're still holding a ref via `f`
f.key: "memory",
# This was fetched from another worker. While we hold a ref via `g`, the
# scheduler only instructed to compute this on B
g.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states, a)
# Expected states after we release references to the futures
f.release()
g.release()
# We no longer hold any refs to f or g and B didn't have any errors. It
# releases everything as expected
while b.state.tasks:
await asyncio.sleep(0.01)
expected_states = {
f.key: "released",
g.key: "released",
res.key: "error",
}
await assert_task_states_on_worker(expected_states, a)
res.release()
# We no longer hold any refs. Cluster should reset completely
# This is not happening
while any((s.tasks, a.state.tasks, b.state.tasks)):
await asyncio.sleep(0.01)
@gen_cluster(client=True, config=NO_AMM)
async def test_worker_state_error_release_error_first(c, s, a, b):
"""
Create a chain of tasks and err one of them. Then release tasks in a certain
order and ensure the tasks are released and/or kept in memory as appropriate
F -- RES (error)
/
/
G
Free error first
"""
def raise_exc(*args):
raise RuntimeError()
f = c.submit(inc, 1, workers=[a.address], key="f")
g = c.submit(inc, 1, workers=[b.address], key="g")
res = c.submit(raise_exc, f, g, workers=[a.address])
with pytest.raises(RuntimeError):
await res
# Nothing bad happened on B, therefore B should hold on to G
assert len(b.state.tasks) == 1
assert g.key in b.state.tasks
# A raised the exception therefore we should hold on to the erroneous task
assert res.key in a.state.tasks
ts = a.state.tasks[res.key]
assert ts.state == "error"
expected_states = {
# A was instructed to compute this result and we're still holding a ref
# via `f`
f.key: "memory",
# This was fetched from another worker. While we hold a ref via `g`, the
# scheduler only instructed to compute this on B
g.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states, a)
# Expected states after we release references to the futures
res.release()
# We no longer hold any refs to f or g and B didn't have any errors. It
# releases everything as expected
while res.key in a.state.tasks:
await asyncio.sleep(0.01)
expected_states = {
f.key: "memory",
g.key: "memory",
}
await assert_task_states_on_worker(expected_states, a)
f.release()
g.release()
while any((s.tasks, a.state.tasks, b.state.tasks)):
await asyncio.sleep(0.01)
@gen_cluster(client=True, config=NO_AMM)
async def test_worker_state_error_release_error_int(c, s, a, b):
"""
Create a chain of tasks and err one of them. Then release tasks in a certain
order and ensure the tasks are released and/or kept in memory as appropriate
F -- RES (error)
/
/
G
Free one successful task, then error, then last task
"""
def raise_exc(*args):
raise RuntimeError()
f = c.submit(inc, 1, workers=[a.address], key="f")
g = c.submit(inc, 1, workers=[b.address], key="g")
res = c.submit(raise_exc, f, g, workers=[a.address])
with pytest.raises(RuntimeError):
await res
# Nothing bad happened on B, therefore B should hold on to G
assert len(b.state.tasks) == 1
assert g.key in b.state.tasks
# A raised the exception therefore we should hold on to the erroneous task
assert res.key in a.state.tasks
ts = a.state.tasks[res.key]
assert ts.state == "error"
expected_states = {
# A was instructed to compute this result and we're still holding a ref via `f`
f.key: "memory",
# This was fetched from another worker. While we hold a ref via `g`, the
# scheduler only instructed to compute this on B
g.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states, a)
# Expected states after we release references to the futures
f.release()
res.release()
# We no longer hold any refs to f or g and B didn't have any errors. It
# releases everything as expected
while len(a.state.tasks) > 1:
await asyncio.sleep(0.01)
expected_states = {
g.key: "memory",
}
await assert_task_states_on_worker(expected_states, a)
await assert_task_states_on_worker(expected_states, b)
g.release()
# We no longer hold any refs. Cluster should reset completely
while any((s.tasks, a.state.tasks, b.state.tasks)):
await asyncio.sleep(0.01)
@gen_cluster(client=True, config=NO_AMM)
async def test_worker_state_error_long_chain(c, s, a, b):
def raise_exc(*args):
raise RuntimeError()
# f (A) --------> res (B)
# /
# g (B) -> h (A)
f = c.submit(inc, 1, workers=[a.address], key="f", allow_other_workers=False)
g = c.submit(inc, 1, workers=[b.address], key="g", allow_other_workers=False)
h = c.submit(inc, g, workers=[a.address], key="h", allow_other_workers=False)
res = c.submit(
raise_exc, f, h, workers=[b.address], allow_other_workers=False, key="res"
)
with pytest.raises(RuntimeError):
await res
expected_states_A = {
f.key: "memory",
g.key: "memory",
h.key: "memory",
}
await assert_task_states_on_worker(expected_states_A, a)
expected_states_B = {
f.key: "memory",
g.key: "memory",
h.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states_B, b)
f.release()
expected_states_A = {
g.key: "memory",
h.key: "memory",
}
await assert_task_states_on_worker(expected_states_A, a)
expected_states_B = {
f.key: "released",
g.key: "memory",
h.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states_B, b)
g.release()
expected_states_A = {
g.key: "released",
h.key: "memory",
}
await assert_task_states_on_worker(expected_states_A, a)
# B must not forget a task since all have a still valid dependent
expected_states_B = {
f.key: "released",
h.key: "memory",
res.key: "error",
}
await assert_task_states_on_worker(expected_states_B, b)
h.release()
expected_states_A = {}
await assert_task_states_on_worker(expected_states_A, a)
expected_states_B = {
f.key: "released",
h.key: "released",
res.key: "error",
}
await assert_task_states_on_worker(expected_states_B, b)
res.release()
# We no longer hold any refs. Cluster should reset completely
while any((s.tasks, a.state.tasks, b.state.tasks)):
await asyncio.sleep(0.01)
@gen_cluster(client=True, nthreads=[("", x) for x in (1, 2, 3, 4)], config=NO_AMM)
async def test_hold_on_to_replicas(c, s, *workers):
f1 = c.submit(inc, 1, workers=[workers[0].address], key="f1")
f2 = c.submit(inc, 2, workers=[workers[1].address], key="f2")
sum_1 = c.submit(
slowsum, [f1, f2], delay=0.1, workers=[workers[2].address], key="sum"
)
sum_2 = c.submit(
slowsum, [f1, sum_1], delay=0.2, workers=[workers[3].address], key="sum_2"
)
f1.release()
f2.release()
while sum_2.key not in workers[3].state.tasks:
await asyncio.sleep(0.01)
while not workers[3].state.tasks[sum_2.key].state == "memory":
assert len(s.tasks[f1.key].who_has) >= 2
assert s.tasks[f2.key].state == "released"
await asyncio.sleep(0.01)
while len(workers[2].data) > 1:
await asyncio.sleep(0.01)
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)])
async def test_forget_dependents_after_release(c, s, a):
fut = c.submit(inc, 1, key="f-1")
fut2 = c.submit(inc, fut, key="f-2")
await asyncio.gather(fut, fut2)
assert fut.key in a.state.tasks
assert fut2.key in a.state.tasks
assert fut2.key in {d.key for d in a.state.tasks[fut.key].dependents}
fut2.release()
while fut2.key in a.state.tasks:
await asyncio.sleep(0.001)
assert fut2.key not in {d.key for d in a.state.tasks[fut.key].dependents}
@gen_cluster(client=True)
async def test_steal_during_task_deserialization(c, s, a, b, monkeypatch):
stealing_ext = s.extensions["stealing"]
await stealing_ext.stop()
from distributed.utils import ThreadPoolExecutor
class CountingThreadPool(ThreadPoolExecutor):
counter = 0
def submit(self, *args, **kwargs):
CountingThreadPool.counter += 1
return super().submit(*args, **kwargs)
# Ensure we're always offloading
monkeypatch.setattr("distributed.worker.OFFLOAD_THRESHOLD", 1)
threadpool = CountingThreadPool(
max_workers=1, thread_name_prefix="Counting-Offload-Threadpool"
)
try:
monkeypatch.setattr("distributed.utils._offload_executor", threadpool)
class SlowDeserializeCallable:
def __init__(self, delay=0.1):
self.delay = delay
def __getstate__(self):
return self.delay
def __setstate__(self, state):
delay = state
import time
time.sleep(delay)
return SlowDeserializeCallable(delay)
def __call__(self, *args, **kwargs):
return 41
slow_deserialized_func = SlowDeserializeCallable()
fut = c.submit(
slow_deserialized_func, 1, workers=[a.address], allow_other_workers=True
)
while CountingThreadPool.counter == 0:
await asyncio.sleep(0)
ts = s.tasks[fut.key]
a.handle_stimulus(StealRequestEvent(key=fut.key, stimulus_id="test"))
stealing_ext.scheduler.send_task_to_worker(b.address, ts)
fut2 = c.submit(inc, fut, workers=[a.address])
fut3 = c.submit(inc, fut2, workers=[a.address])
assert await fut2 == 42
await fut3
finally:
threadpool.shutdown()
@gen_cluster(client=True)
async def test_run_spec_deserialize_fail(c, s, a, b):
class F:
def __call__(self):
pass
def __reduce__(self):
return lambda: 1 / 0, ()
with captured_logger("distributed.worker") as logger:
fut = c.submit(F())
assert isinstance(await fut.exception(), ZeroDivisionError)
logvalue = logger.getvalue()
assert "Could not deserialize task" in logvalue
assert "return lambda: 1 / 0, ()" in logvalue
@gen_cluster(client=True, config=NO_AMM)
async def test_acquire_replicas(c, s, a, b):
fut = c.submit(inc, 1, workers=[a.address])
await fut
s.request_acquire_replicas(b.address, [fut.key], stimulus_id=f"test-{time()}")
while len(s.tasks[fut.key].who_has) != 2:
await asyncio.sleep(0.005)
for w in (a, b):
assert w.data[fut.key] == 2
assert w.state.tasks[fut.key].state == "memory"
fut.release()
while b.state.tasks or a.state.tasks:
await asyncio.sleep(0.005)
@gen_cluster(client=True, config=NO_AMM)
async def test_acquire_replicas_same_channel(c, s, a, b):
futA = c.submit(inc, 1, workers=[a.address], key="f-A")
futB = c.submit(inc, 2, workers=[a.address], key="f-B")
futC = c.submit(inc, futB, workers=[b.address], key="f-C")
await futA
s.request_acquire_replicas(b.address, [futA.key], stimulus_id=f"test-{time()}")
await futC
while (
futA.key not in b.state.tasks or not b.state.tasks[futA.key].state == "memory"
):
await asyncio.sleep(0.005)
while len(s.tasks[futA.key].who_has) != 2:
await asyncio.sleep(0.005)
# Ensure that both the replica and an ordinary dependency pass through the
# same communication channel
for fut in (futA, futB):
assert_story(
b.state.story(fut.key),
[
("gather-dependencies", a.address, {fut.key}),
("request-dep", a.address, {fut.key}),
],
)
assert any(fut.key in msg["keys"] for msg in b.transfer_incoming_log)
@gen_cluster(client=True, nthreads=[("127.0.0.1", 1)] * 3, config=NO_AMM)
async def test_acquire_replicas_many(c, s, w1, w2, w3):
futs = c.map(inc, range(10), workers=[w1.address])
res = c.submit(sum, futs, workers=[w2.address])
final = c.submit(slowinc, res, delay=0.5, workers=[w2.address])
await wait(futs)
s.request_acquire_replicas(
w3.address, [fut.key for fut in futs], stimulus_id=f"test-{time()}"
)
# Worker 2 should normally not even be involved if there was no replication
while not all(
f.key in w3.state.tasks and w3.state.tasks[f.key].state == "memory"
for f in futs
):
await asyncio.sleep(0.01)
assert all(ts.state == "memory" for ts in w3.state.tasks.values())
assert await final == sum(map(inc, range(10))) + 1
# All workers have a replica
assert all(len(s.tasks[f.key].who_has) == 3 for f in futs)
del futs, res, final
while any(w.state.tasks for w in (w1, w2, w3)):
await asyncio.sleep(0.001)
@gen_cluster(client=True, nthreads=[("", 1)], config=NO_AMM)
async def test_acquire_replicas_already_in_flight(c, s, a):
"""Trying to acquire a replica that is already in flight is a no-op"""
async with BlockedGatherDep(s.address) as b:
x = c.submit(inc, 1, workers=[a.address], key="x")
y = c.submit(inc, x, workers=[b.address], key="y")
await b.in_gather_dep.wait()
assert b.state.tasks["x"].state == "flight"
b.handle_stimulus(
AcquireReplicasEvent(
who_has={"x": a.address}, nbytes={"x": 1}, stimulus_id="test"
)
)
assert b.state.tasks["x"].state == "flight"
b.block_gather_dep.set()
assert await y == 3
assert_story(
b.state.story("x"),
[
("x", "fetch", "flight", "flight", {}),
("x", "flight", "fetch", "flight", {}),
],
)
@pytest.mark.slow
@gen_cluster(client=True, config=NO_AMM)
async def test_forget_acquire_replicas(c, s, a, b):
"""
1. The scheduler sends acquire-replicas to the worker
2. Before the task can reach the worker, the task is forgotten on the scheduler
and on the peer workers holding the replicas.
This is unlike in a compute-task command, where the workers that are fetching
the key are tracked on the scheduler side and receive a release-key command too.
3. The task eventually transitions to missing
4. At the next run of find_missing, the worker sends a request-refresh-who-has
message to the scheduler
5. The scheduler responds with free-keys
6. The task is forgotten everywhere.
"""
x = c.submit(inc, 2, key="x", workers=[a.address])
await x
with freeze_data_fetching(b, jump_start=True):
s.request_acquire_replicas(b.address, ["x"], stimulus_id="test")
await wait_for_state("x", "fetch", b)
x.release()
while "x" in s.tasks or "x" in a.state.tasks:
await asyncio.sleep(0.01)
while "x" in b.state.tasks:
await asyncio.sleep(0.01)
assert "x" not in s.tasks
@gen_cluster(client=True, config=NO_AMM)
async def test_remove_replicas_simple(c, s, a, b):
futs = c.map(inc, range(10), workers=[a.address])
await wait(futs)
s.request_acquire_replicas(
b.address, [fut.key for fut in futs], stimulus_id=f"test-{time()}"
)
while not all(len(s.tasks[f.key].who_has) == 2 for f in futs):
await asyncio.sleep(0.01)
s.request_remove_replicas(
b.address, [fut.key for fut in futs], stimulus_id=f"test-{time()}"
)
assert all(len(s.tasks[f.key].who_has) == 1 for f in futs)
while b.state.tasks:
await asyncio.sleep(0.01)
# Ensure there is no delayed reply to re-register the key
await asyncio.sleep(0.01)
assert all(s.tasks[f.key].who_has == {s.workers[a.address]} for f in futs)
@gen_cluster(
client=True,
nthreads=[("", 1), ("", 6)], # Up to 5 threads of b will get stuck; read below
config=merge(NO_AMM, {"distributed.comm.recent-messages-log-length": 1_000}),
)
async def test_remove_replicas_while_computing(c, s, a, b):
futs = c.map(inc, range(10), workers=[a.address])
dependents_event = distributed.Event()
def some_slow(x, event):
if x % 2:
event.wait()
return x + 1
# All interesting things will happen on b
dependents = c.map(some_slow, futs, event=dependents_event, workers=[b.address])
while not any(f.key in b.state.tasks for f in dependents):
await asyncio.sleep(0.01)
# The scheduler removes keys from who_has/has_what immediately
# Make sure the worker responds to the rejection and the scheduler corrects
# the state
ws = s.workers[b.address]
def ws_has_futs(aggr_func):
nonlocal futs
return aggr_func(s.tasks[fut.key] in ws.has_what for fut in futs)
# Wait for all futs to transfer over
while not ws_has_futs(all):
await asyncio.sleep(0.01)
# Wait for some dependent tasks to be done. No more than half of the dependents can
# finish, as the others are blocked on dependents_event.
# Note: for this to work reliably regardless of scheduling order, we need to have 6+
# threads. At the moment of writing it works with 2 because futures of Client.map
# are always scheduled from left to right, but we'd rather not rely on this
# assumption.
while not any(fut.status == "finished" for fut in dependents):
await asyncio.sleep(0.01)
assert not all(fut.status == "finished" for fut in dependents)
# Try removing the initial keys
s.request_remove_replicas(
b.address, [fut.key for fut in futs], stimulus_id=f"test-{time()}"
)
# Scheduler removed all keys immediately...
assert not ws_has_futs(any)
# ... but the state is properly restored for all tasks for which the dependent task
# isn't done yet
while not ws_has_futs(any):
await asyncio.sleep(0.01)
# Let the remaining dependent tasks complete
await dependents_event.set()
await wait(dependents)
assert ws_has_futs(any) and not ws_has_futs(all)
# If a request is rejected, the worker responds with an add-keys message to
# reenlist the key in the schedulers state system to avoid race conditions,
# see also https://github.com/dask/distributed/issues/5265
rejections = set()
for msg in b.state.log:
if msg[0] == "remove-replica-rejected":
rejections.update(msg[1])
assert rejections
def answer_sent(key):
for batch in b.batched_stream.recent_message_log:
for msg in batch:
if "op" in msg and msg["op"] == "add-keys" and key in msg["keys"]:
return True
return False
for rejected_key in rejections:
assert answer_sent(rejected_key)
# Now that all dependent tasks are done, futs replicas may be removed.
# They might be already gone due to the above remove replica calls
s.request_remove_replicas(
b.address,
[fut.key for fut in futs if ws in s.tasks[fut.key].who_has],
stimulus_id=f"test-{time()}",
)
while any(
b.state.tasks[f.key].state != "released" for f in futs if f.key in b.state.tasks
):
await asyncio.sleep(0.01)
# The scheduler actually gets notified about the removed replica
while ws_has_futs(any):
await asyncio.sleep(0.01)
# A replica is still on workers[0]
assert all(len(s.tasks[f.key].who_has) == 1 for f in futs)
del dependents, futs
while any(w.state.tasks for w in (a, b)):
await asyncio.sleep(0.01)
@gen_cluster(client=True, nthreads=[("", 1)] * 3, config=NO_AMM)
async def test_who_has_consistent_remove_replicas(c, s, *workers):
a = workers[0]
other_workers = {w for w in workers if w != a}
f1 = c.submit(inc, 1, key="f1", workers=[w.address for w in other_workers])
await wait(f1)
for w in other_workers:
s.request_acquire_replicas(w.address, [f1.key], stimulus_id=f"test-{time()}")
while not len(s.tasks[f1.key].who_has) == len(other_workers):
await asyncio.sleep(0)
f2 = c.submit(inc, f1, workers=[a.address])
# Wait just until the moment the worker received the task and scheduled the
# task to be fetched, then remove the replica from the worker this one is
# trying to get the data from. Ensure this is handled gracefully and no
# suspicious counters are raised since this is expected behaviour when
# removing replicas
while f1.key not in a.state.tasks or a.state.tasks[f1.key].state != "flight":
await asyncio.sleep(0)
coming_from = None
for w in other_workers:
coming_from = w
if w.address == a.state.tasks[f1.key].coming_from:
break
coming_from.handle_stimulus(RemoveReplicasEvent(keys=[f1.key], stimulus_id="test"))
await f2
assert a.state.tasks[f1.key].suspicious_count == 0
assert s.tasks[f1.key].suspicious == 0
@gen_cluster(client=True, config=NO_AMM)
async def test_acquire_replicas_with_no_priority(c, s, a, b):
"""Scattered tasks have no priority. When they transit to another worker through
acquire-replicas, they end up in the Worker.data_needed heap together with tasks
with a priority; they must gain a priority to become sortable.
"""
x = (await c.scatter({"x": 1}, workers=[a.address]))["x"]
y = c.submit(lambda: 2, key="y", workers=[a.address])
await y
assert s.tasks["x"].priority is None
assert a.state.tasks["x"].priority is None
assert s.tasks["y"].priority is not None
assert a.state.tasks["y"].priority is not None
s.request_acquire_replicas(b.address, [x.key, y.key], stimulus_id=f"test-{time()}")
while b.data != {"x": 1, "y": 2}:
await asyncio.sleep(0.01)
assert s.tasks["x"].priority is None
assert a.state.tasks["x"].priority is None
assert b.state.tasks["x"].priority is not None
@gen_cluster(client=True, nthreads=[("", 1)], config=NO_AMM)
async def test_acquire_replicas_large_data(c, s, a):
"""When acquire-replicas is used to acquire multiple sizeable tasks, it respects
transfer_message_bytes_limit and acquires them over multiple iterations.
"""
size = a.state.transfer_message_bytes_limit // 5 - 10_000
class C:
def __sizeof__(self):
return size
futs = [c.submit(C, key=f"x{i}", workers=[a.address]) for i in range(10)]
await wait(futs)
async with BlockedGatherDep(s.address) as b:
s.request_acquire_replicas(
b.address, [f"x{i}" for i in range(10)], stimulus_id="test"
)
await b.in_gather_dep.wait()
assert len(b.state.in_flight_tasks) == 5
assert len(b.state.data_needed[a.address]) == 5
b.block_gather_dep.set()
while len(b.data) < 10:
await asyncio.sleep(0.01)
@gen_cluster(client=True)
async def test_missing_released_zombie_tasks(c, s, a, b):
"""
Ensure that no fetch/flight tasks are left in the task dict of a
worker after everything was released
"""
a.transfer_outgoing_count_limit = 0
f1 = c.submit(inc, 1, key="f1", workers=[a.address])
f2 = c.submit(inc, f1, key="f2", workers=[b.address])
key = f1.key
while key not in b.state.tasks or b.state.tasks[key].state != "fetch":
await asyncio.sleep(0.01)
await a.close()
del f1, f2
while b.state.tasks:
await asyncio.sleep(0.01)
class BrokenWorker(Worker):
async def get_data(self, comm, *args, **kwargs):
comm.abort()
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_missing_released_zombie_tasks_2(c, s, b):
# If get_data_from_worker raises this will suggest a dead worker to B and it
# will transition the task to missing. We want to make sure that a missing
# task is properly released and not left as a zombie
async with BrokenWorker(s.address) as a:
f1 = c.submit(inc, 1, key="f1", workers=[a.address])
f2 = c.submit(inc, f1, key="f2", workers=[b.address])
while f1.key not in b.state.tasks:
await asyncio.sleep(0)
ts = b.state.tasks[f1.key]
assert ts.state == "flight"
while ts.state != "missing":
# If we sleep for a longer time, the worker will spin into an
# endless loop of asking the scheduler who_has and trying to connect
# to A
await asyncio.sleep(0)
del f1, f2
while b.state.tasks:
await asyncio.sleep(0.01)
assert_story(
b.state.story(ts),
[("f1", "missing", "released", "released", {"f1": "forgotten"})],
)
@gen_cluster(nthreads=[("", 1)], config={"distributed.worker.memory.pause": False})
async def test_worker_status_sync(s, a):
ws = s.workers[a.address]
a.status = Status.paused
while ws.status != Status.paused:
await asyncio.sleep(0.01)
a.status = Status.running
while ws.status != Status.running:
await asyncio.sleep(0.01)
await s.retire_workers()
while ws.status != Status.closed:
await asyncio.sleep(0.01)
events = [ev for _, ev in s.events[ws.address] if ev["action"] != "heartbeat"]
assert events == [
{"action": "add-worker"},
{
"action": "worker-status-change",
"prev-status": "init",
"status": "running",
},
{
"action": "worker-status-change",
"prev-status": "running",
"status": "paused",
},
{
"action": "worker-status-change",
"prev-status": "paused",
"status": "running",
},
{
"action": "worker-status-change",
"prev-status": "running",
"status": "closing_gracefully",
},
{"action": "remove-worker", "processing-tasks": set()},
{"action": "retired"},
]
@gen_cluster(client=True)
async def test_task_flight_compute_oserror(c, s, a, b):
"""If the remote worker dies while a task is in flight, the task may be
rescheduled to be computed on the worker trying to fetch the data.
However, the OSError caused by the dead remote would try to transition the
task to missing which is not what we want. This test ensures that the task
is properly transitioned to executing and the scheduler doesn't reschedule
anything and rejects the "false missing" signal from the worker, if there is
any.
"""
write_queue = asyncio.Queue()
write_event = asyncio.Event()
b.rpc = _LockedCommPool(
b.rpc,
write_queue=write_queue,
write_event=write_event,
)
futs = c.submit(map, inc, range(10), workers=[a.address], allow_other_workers=True)
await wait(futs)
assert a.data
assert write_queue.empty()
f1 = c.submit(sum, futs, workers=[b.address], key="f1")
peer, msg = await write_queue.get()
assert peer == a.address
assert msg["op"] == "get_data"
in_flight_tasks = [ts for ts in b.state.tasks.values() if ts.key != "f1"]
assert all(ts.state == "flight" for ts in in_flight_tasks)
await a.close()
write_event.set()
await f1
# If the above doesn't deadlock the behavior should be OK. We're still
# asserting a few internals to make sure that if things change this is done
# deliberately
sum_story = b.state.story("f1")
expected_sum_story = [
("f1", "compute-task", "released"),
(
"f1",
"released",
"waiting",
"waiting",
{ts.key: "fetch" for ts in in_flight_tasks},
),
# inc is lost and needs to be recomputed. Therefore, sum is released
("free-keys", ("f1",)),
# The recommendations here are hard to predict. Whatever key is
# currently scheduled to be fetched, if any, will be recommended to be
# released.
("f1", "waiting", "released", "released", lambda msg: msg["f1"] == "forgotten"),
("f1", "released", "forgotten", "forgotten", {}),
# Now, we actually compute the task *once*. This must not cycle back
("f1", "compute-task", "released"),
("f1", "released", "waiting", "waiting", {"f1": "ready"}),
("f1", "waiting", "ready", "ready", {"f1": "executing"}),
("f1", "ready", "executing", "executing", {}),
("f1", "put-in-memory"),
("f1", "executing", "memory", "memory", {}),
]
assert_story(sum_story, expected_sum_story, strict=True)
@gen_cluster(client=True, nthreads=[])
async def test_gather_dep_cancelled_rescheduled(c, s):
"""At time of writing, the gather_dep implementation filtered tasks again
for in-flight state. The response parser, however, did not distinguish
resulting in unwanted missing-data signals to the scheduler, causing
potential rescheduling or data leaks.
(Note: missing-data was removed in #6445).
If a cancelled key is rescheduled for fetching while gather_dep waits
internally for get_data, the response parser would misclassify this key and
cause the key to be recommended for a release causing deadlocks and/or lost
keys.
At time of writing, this transition was implemented wrongly and caused a
flight->cancelled transition which should be recoverable but the cancelled
state was corrupted by this transition since ts.done==True. This attribute
setting would cause a cancelled->fetch transition to actually drop the key
instead, causing https://github.com/dask/distributed/issues/5366
See also test_gather_dep_do_not_handle_response_of_not_requested_tasks
"""
async with BlockedGetData(s.address) as a:
async with BlockedGatherDep(s.address) as b:
fut1 = c.submit(inc, 1, workers=[a.address], key="f1")
fut2 = c.submit(inc, fut1, workers=[a.address], key="f2")
await wait(fut2)
fut4 = c.submit(sum, fut1, fut2, workers=[b.address], key="f4")
fut3 = c.submit(inc, fut1, workers=[b.address], key="f3")
await wait_for_state(fut2.key, "flight", b)
await b.in_gather_dep.wait()
fut4.release()
while fut4.key in b.state.tasks:
await asyncio.sleep(0)
assert b.state.tasks[fut2.key].state == "cancelled"
b.block_gather_dep.set()
await a.in_get_data.wait()
fut4 = c.submit(sum, [fut1, fut2], workers=[b.address], key="f4")
await wait_for_state(fut2.key, "flight", b)
a.block_get_data.set()
await wait([fut3, fut4])
f2_story = b.state.story(fut2.key)
assert f2_story
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_gather_dep_do_not_handle_response_of_not_requested_tasks(c, s, a):
"""At time of writing, the gather_dep implementation filtered tasks again
for in-flight state. The response parser, however, did not distinguish
resulting in unwanted missing-data signals to the scheduler, causing
potential rescheduling or data leaks.
(Note: missing-data was removed in #6445).
This test may become obsolete if the implementation changes significantly.
"""
async with BlockedGatherDep(s.address) as b:
fut1 = c.submit(inc, 1, workers=[a.address], key="f1")
fut2 = c.submit(inc, fut1, workers=[a.address], key="f2")
await fut2
fut4 = c.submit(sum, fut1, fut2, workers=[b.address], key="f4")
fut3 = c.submit(inc, fut1, workers=[b.address], key="f3")
await b.in_gather_dep.wait()
assert b.state.tasks[fut2.key].state == "flight"
fut4.release()
while fut4.key in b.state.tasks:
await asyncio.sleep(0.01)
assert b.state.tasks[fut2.key].state == "cancelled"
b.block_gather_dep.set()
await fut3
assert fut2.key not in b.state.tasks
f2_story = b.state.story(fut2.key)
assert f2_story
assert not any("missing-dep" in msg for msg in f2_story)
@gen_cluster(
client=True,
nthreads=[("", 1)],
config={"distributed.comm.recent-messages-log-length": 1000},
)
async def test_gather_dep_no_longer_in_flight_tasks(c, s, a):
async with BlockedGatherDep(s.address) as b:
fut1 = c.submit(inc, 1, workers=[a.address], key="f1")
fut2 = c.submit(sum, fut1, fut1, workers=[b.address], key="f2")
await wait_for_state(fut1.key, "flight", b)
await b.in_gather_dep.wait()
fut2.release()
while fut2.key in b.state.tasks:
await asyncio.sleep(0.01)
assert b.state.tasks[fut1.key].state == "cancelled"
b.block_gather_dep.set()
while fut2.key in b.state.tasks:
await asyncio.sleep(0.01)
f1_story = b.state.story(fut1.key)
f2_story = b.state.story(fut2.key)
assert f1_story
assert f2_story
assert not any("missing-dep" in msg for msg in f2_story)
@gen_cluster(client=True, nthreads=[("", 1)])
async def test_Worker__to_dict(c, s, a):
x = c.submit(inc, 1, key="x")
await wait(x)
d = a._to_dict()
assert set(d) == {
"type",
"id",
"scheduler",
"address",
"status",
"thread_id",
"logs",
"config",
"transfer_incoming_log",
"transfer_outgoing_log",
# Attributes of WorkerMemoryManager
"data",
"max_spill",
"memory_limit",
"memory_monitor_interval",
"memory_pause_fraction",
"memory_spill_fraction",
"memory_target_fraction",
# Attributes of WorkerState
"nthreads",
"running",
"ready",
"has_what",
"constrained",
"executing",
"long_running",
"missing_dep_flight",
"in_flight_tasks",
"in_flight_workers",
"busy_workers",
"log",
"stimulus_log",
"transition_counter",
"tasks",
"data_needed",
}
assert d["tasks"]["x"]["key"] == "x"
assert d["data"] == {"x": None}
@gen_cluster(nthreads=[])
async def test_extension_methods(s):
flag = False
shutdown = False
class WorkerExtension:
def __init__(self, worker):
pass
def heartbeat(self):
return {"data": 123}
async def close(self):
nonlocal shutdown
shutdown = True
class SchedulerExtension:
def __init__(self, scheduler):
self.scheduler = scheduler
pass
def heartbeat(self, ws, data):
nonlocal flag
assert ws in self.scheduler.workers.values()
assert data == {"data": 123}
flag = True
s.extensions["test"] = SchedulerExtension(s)
async with Worker(s.address, extensions={"test": WorkerExtension}) as w:
assert not shutdown
await w.heartbeat()
assert flag
assert shutdown
@gen_cluster()
async def test_benchmark_hardware(s, a, b):
sizes = ["1 kiB", "10 kiB"]
disk = benchmark_disk(sizes=sizes, duration="1 ms")
memory = benchmark_memory(sizes=sizes, duration="1 ms")
network = await benchmark_network(
address=a.address, rpc=b.rpc, sizes=sizes, duration="1 ms"
)
assert set(disk) == set(memory) == set(network) == set(sizes)
@gen_cluster(
client=True,
config={
"distributed.admin.tick.interval": "5ms",
"distributed.admin.tick.cycle": "100ms",
},
)
async def test_tick_interval(c, s, a, b):
import time
await a.heartbeat()
x = s.workers[a.address].metrics["event_loop_interval"]
while s.workers[a.address].metrics["event_loop_interval"] > 0.050:
await asyncio.sleep(0.01)
while s.workers[a.address].metrics["event_loop_interval"] < 0.100:
await asyncio.sleep(0.01)
time.sleep(0.200)
class BreakingWorker(Worker):
broke_once = False
def get_data(self, comm, **kwargs):
if not self.broke_once:
self.broke_once = True
raise OSError("fake error")
return super().get_data(comm, **kwargs)
@pytest.mark.slow
@gen_cluster(client=True, Worker=BreakingWorker)
async def test_broken_comm(c, s, a, b):
df = dask.datasets.timeseries(
start="2000-01-01",
end="2000-01-10",
)
s = df.shuffle("id", shuffle="tasks")
await c.compute(s.size)
@gen_cluster(nthreads=[])
async def test_do_not_block_event_loop_during_shutdown(s):
loop = asyncio.get_running_loop()
called_handler = threading.Event()
block_handler = threading.Event()
w = await Worker(s.address)
executor = w.executors["default"]
# The block wait must be smaller than the test timeout and smaller than the
# default value for timeout in `Worker.close``
async def block():
def fn():
called_handler.set()
assert block_handler.wait(20)
await loop.run_in_executor(executor, fn)
async def set_future():
while True:
try:
await loop.run_in_executor(executor, sleep, 0.1)
except RuntimeError: # executor has started shutting down
block_handler.set()
return
async def close():
called_handler.wait()
# executor_wait is True by default but we want to be explicit here
await w.close(executor_wait=True)
await asyncio.gather(block(), close(), set_future())
@gen_cluster(nthreads=[])
async def test_reconnect_argument_deprecated(s):
with pytest.deprecated_call(match="`reconnect` argument"):
async with Worker(s.address, reconnect=False):
pass
with pytest.raises(ValueError, match="reconnect=True"):
async with Worker(s.address, reconnect=True):
pass
with warnings.catch_warnings():
# No argument should not warn or raise
warnings.simplefilter("error")
async with Worker(s.address):
pass
@gen_cluster(client=True, nthreads=[])
async def test_worker_running_before_running_plugins(c, s, caplog):
class InitWorkerNewThread(WorkerPlugin):
name = "init_worker_new_thread"
setup_status = None
def setup(self, worker):
self.setup_status = worker.status
def teardown(self, worker):
pass
await c.register_worker_plugin(InitWorkerNewThread())
async with Worker(s.address) as worker:
assert await c.submit(inc, 1) == 2
assert worker.plugins[InitWorkerNewThread.name].setup_status is Status.running
@gen_cluster(
client=True,
nthreads=[],
config={
# This is just to make Scheduler.retire_worker more reactive to changes
"distributed.scheduler.active-memory-manager.start": True,
"distributed.scheduler.active-memory-manager.interval": "50ms",
},
)
async def test_execute_preamble_abort_retirement(c, s):
"""Test race condition in the preamble of Worker.execute(), which used to cause a
task to remain permanently in executing state in case of very tight timings when
exiting the closing_gracefully status.
See also
--------
https://github.com/dask/distributed/issues/6867
test_cancelled_state.py::test_execute_preamble_early_cancel
"""
async with BlockedExecute(s.address) as a:
await c.wait_for_workers(1)
a.block_deserialize_task.set() # Uninteresting in this test
x = await c.scatter({"x": 1}, workers=[a.address])
y = c.submit(inc, 1, key="y", workers=[a.address])
await a.in_execute.wait()
async with BlockedGatherDep(s.address) as b:
await c.wait_for_workers(2)
retire_fut = asyncio.create_task(c.retire_workers([a.address]))
while a.status != Status.closing_gracefully:
await asyncio.sleep(0.01)
# The Active Memory Manager will send to b the message
# {op: acquire-replicas, who_has: {x: [a.address]}}
await b.in_gather_dep.wait()
# Run Worker.execute. At the moment of writing this test, the method would
# detect the closing_gracefully status and perform an early exit.
a.block_execute.set()
await a.in_execute_exit.wait()
# b has shut down. There's nowhere to replicate x to anymore, so retire_workers
# will give up and reinstate a to running status.
assert await retire_fut == {}
while a.status != Status.running:
await asyncio.sleep(0.01)
# Finally let the done_callback of Worker.execute run
a.block_execute_exit.set()
# Test that y does not get stuck.
assert await y == 2
@gen_cluster()
async def test_deprecation_of_renamed_worker_attributes(s, a, b):
msg = (
"The `Worker.outgoing_count` attribute has been renamed to "
"`Worker.transfer_outgoing_count_total`"
)
with pytest.warns(DeprecationWarning, match=msg):
assert a.outgoing_count == a.transfer_outgoing_count_total
msg = (
"The `Worker.outgoing_current_count` attribute has been renamed to "
"`Worker.transfer_outgoing_count`"
)
with pytest.warns(DeprecationWarning, match=msg):
assert a.outgoing_current_count == a.transfer_outgoing_count
@gen_cluster(nthreads=[])
async def test_worker_log_memory_limit_too_high(s):
with captured_logger("distributed.worker.memory") as caplog:
async with Worker(s.address, memory_limit="1PB"):
pass
expected_snippets = [
("ignore", "ignoring"),
("memory limit", "memory_limit"),
("system"),
("1PB"),
]
for snippets in expected_snippets:
# assert any(snip in caplog.text for snip in snippets)
assert any(snip in caplog.getvalue().lower() for snip in snippets)
@gen_cluster(client=True, Worker=Nanny)
async def test_forward_output(c, s, a, b, capsys):
def print_stdout(*args, **kwargs):
print(*args, file=sys.stdout, **kwargs)
def print_stderr(*args, **kwargs):
print(*args, file=sys.stderr, **kwargs)
plugin = ForwardOutput()
out, err = capsys.readouterr()
counter = itertools.count()
# Without the plugin installed, we should not see any output
# Note that we use nannies so workers run in subprocesses
await c.submit(print_stdout, "foo", key=next(counter))
await c.submit(print_stdout, "bar\n", key=next(counter))
await c.submit(print_stdout, "baz", end="", flush=True, key=next(counter))
await c.submit(print_stdout, 1, 2, end="\n", sep="\n", key=next(counter))
out, err = capsys.readouterr()
assert "" == out
assert "" == err
await c.submit(print_stderr, "foo", key=next(counter))
await c.submit(print_stderr, "bar\n", key=next(counter))
await c.submit(print_stderr, "baz", flush=True, key=next(counter))
await c.submit(print_stderr, 1, 2, end="\n", sep="\n", key=next(counter))
out, err = capsys.readouterr()
assert "" == out
assert "" == err
# After installing, output should be forwarded
await c.register_worker_plugin(plugin, "forward")
await asyncio.sleep(0.1) # Let setup messages come in
capsys.readouterr()
await c.submit(print_stdout, "foo", key=next(counter))
out, err = capsys.readouterr()
assert "foo\n" == out
assert "" == err
await c.submit(print_stdout, "bar\n", key=next(counter))
out, err = capsys.readouterr()
assert "bar\n\n" == out
assert "" == err
await c.submit(print_stdout, "baz", end="", flush=True, key=next(counter))
out, err = capsys.readouterr()
assert "baz" == out
assert "" == err
await c.submit(print_stdout, "first\nsecond", end="", key=next(counter))
out, err = capsys.readouterr()
assert "first\nsecond" == out
assert "" == err
await c.submit(print_stdout, 1, 2, sep=":", key=next(counter))
out, err = capsys.readouterr()
assert "1:2\n" == out
assert err == ""
await c.submit(print_stderr, "fatal", key=next(counter))
out, err = capsys.readouterr()
assert "" == out
assert "fatal\n" == err
# Registering the plugin is idempotent
other_plugin = ForwardOutput()
await c.register_worker_plugin(other_plugin, "forward")
await asyncio.sleep(0.1) # Let teardown/setup messages come in
out, err = capsys.readouterr()
await c.submit(print_stdout, "foo", key=next(counter))
out, err = capsys.readouterr()
assert "foo\n" == out
assert "" == err
# After unregistering the plugin, we should once again not see any output
await c.unregister_worker_plugin("forward")
await asyncio.sleep(0.1) # Let teardown messages come in
capsys.readouterr()
await c.submit(print_stdout, "foo", key=next(counter))
out, err = capsys.readouterr()
assert "" == out
assert "" == err
await c.submit(print_stderr, "fatal", key=next(counter))
out, err = capsys.readouterr()
assert "" == out
assert "" == err
|