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
|
DO $$
DECLARE
v_version numeric;
BEGIN
SELECT substring(setting from '\d+')::numeric INTO v_version FROM pg_settings WHERE name = 'server_version';
-- GUCs common for all versions
PERFORM set_config('extra_float_digits', '0', false);
PERFORM set_config('parallel_setup_cost', '0', false);
PERFORM set_config('parallel_tuple_cost', '0', false);
PERFORM set_config('max_parallel_workers_per_gather', '2', false);
-- 9.6 used somewhat different GUC name for relation size
IF v_version < 10 THEN
PERFORM set_config('min_parallel_relation_size', '1kB', false);
ELSE
PERFORM set_config('min_parallel_table_scan_size', '1kB', false);
END IF;
-- in 14 disable Memoize nodes, to make explain more consistent
IF v_version >= 14 THEN
PERFORM set_config('enable_memoize', 'off', false);
END IF;
END;
$$ LANGUAGE plpgsql;
-- <value,count> API
select tdigest_percentile(value, count, 100, 0.95)
from (values
(47325940488,1),
(15457695432,2),
(6889790700,3),
(4188763788,4),
(2882932224,5),
(2114815860,6),
(1615194324,7),
(2342114568,9),
(1626471924,11),
(1660755408,14),
(1143728292,17),
(1082582424,21),
(911488284,26),
(728863908,32),
(654898692,40),
(530198076,50),
(417883440,62),
(341452344,77),
(274579584,95),
(231921120,118),
(184091820,146),
(152469828,181),
(125634972,224),
(107059704,278),
(88746120,345),
(73135668,428),
(61035756,531),
(50683320,658),
(42331824,816),
(35234400,1012),
(29341356,1255),
(24290928,1556),
(20284668,1929),
(17215908,2391),
(14737488,2964),
(12692772,3674),
(11220732,4555),
(9787584,5647),
(8148420,7000),
(6918612,8678),
(6015000,10758),
(5480316,13336),
(5443356,16532),
(4535616,20494),
(3962316,25406),
(3914484,31495),
(3828108,39043),
(3583536,48400),
(4104120,60000),
(166024740,2147483647)) foo (count, value);
----------------------------------------------
-- nice data set with random data (uniform) --
----------------------------------------------
-- 10 centroids (tiny)
WITH
data AS (SELECT prng(1000) x, prng(1000, 29823218) cnt),
data_expanded AS (SELECT x FROM (SELECT x, generate_series(1, (10 + 100 * cnt)::int) FROM data) foo ORDER BY random())
SELECT
p,
abs(a - b) < 0.1, -- arbitrary threshold of 10%
(CASE WHEN abs(a - b) < 0.1 THEN NULL ELSE (a - b) END) AS err
FROM (
SELECT
unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
unnest(a) AS a,
unnest(b) AS b
FROM
(SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) a FROM data_expanded) foo,
(SELECT tdigest_percentile(x, (10 + 100 * cnt)::int, 10, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) b FROM data) bar
) baz;
-- 100 centroids (okay-ish)
WITH
data AS (SELECT prng(1000) x, prng(1000, 29823218) cnt),
data_expanded AS (SELECT x FROM (SELECT x, generate_series(1, (10 + 100 * cnt)::int) FROM data) foo ORDER BY random())
SELECT
p,
abs(a - b) < 0.01, -- arbitrary threshold of 1%
(CASE WHEN abs(a - b) < 0.1 THEN NULL ELSE (a - b) END) AS err
FROM (
SELECT
unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
unnest(a) AS a,
unnest(b) AS b
FROM
(SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) a FROM data_expanded) foo,
(SELECT tdigest_percentile(x, (10 + 100 * cnt)::int, 100, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) b FROM data) bar
) baz;
-- 1000 centroids (very accurate)
WITH
data AS (SELECT prng(1000) x, prng(1000, 29823218) cnt),
data_expanded AS (SELECT x FROM (SELECT x, generate_series(1, (10 + 100 * cnt)::int) FROM data) foo ORDER BY random())
SELECT
p,
abs(a - b) < 0.01, -- arbitrary threshold of 1%
(CASE WHEN abs(a - b) < 0.1 THEN NULL ELSE (a - b) END) AS err
FROM (
SELECT
unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
unnest(a) AS a,
unnest(b) AS b
FROM
(SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) a FROM data_expanded) foo,
(SELECT tdigest_percentile(x, (10 + 100 * cnt)::int, 1000, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) b FROM data) bar
) baz;
-- <value,count> API
EXPLAIN (COSTS OFF)
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percentile_disc(0.95) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
0.95,
abs(a - b) / 1000 < 0.01
FROM (
SELECT
(SELECT p FROM x) AS a,
tdigest_percentile(v, c, 100, 0.95) AS b
FROM t) foo;
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percentile_disc(0.95) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
0.95,
abs(a - b) / 1000 < 0.01
FROM (
SELECT
(SELECT p FROM x) AS a,
tdigest_percentile(v, c, 100, 0.95) AS b
FROM t) foo;
EXPLAIN (COSTS OFF)
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percent_rank(950) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
950,
abs(a - b) < 0.01
FROM (
SELECT
(SELECT p FROM x) AS a,
tdigest_percentile_of(v, c, 100, 950) AS b
FROM t) foo;
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percent_rank(950) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
950,
abs(a - b) < 0.01
FROM (
SELECT
(SELECT p FROM x) AS a,
tdigest_percentile_of(v, c, 100, 950) AS b
FROM t) foo;
-- array of percentiles / values
EXPLAIN (COSTS OFF)
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percentile_disc(ARRAY[0.0, 0.95, 0.99, 1.0]) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
p,
abs(a - b) / 1000 < 0.01
FROM (
SELECT
unnest(ARRAY[0.0, 0.95, 0.99, 1.0]) p,
unnest((SELECT p FROM x)) AS a,
unnest(tdigest_percentile(v, c, 100, ARRAY[0.0, 0.95, 0.99, 1.0])) AS b
FROM t) foo;
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT percentile_disc(ARRAY[0.0, 0.95, 0.99, 1.0]) WITHIN GROUP (ORDER BY v) AS p FROM d)
SELECT
p,
abs(a - b) / 1000 < 0.01
FROM (
SELECT
unnest(ARRAY[0.0, 0.95, 0.99, 1.0]) p,
unnest((SELECT p FROM x)) AS a,
unnest(tdigest_percentile(v, c, 100, ARRAY[0.0, 0.95, 0.99, 1.0])) AS b
FROM t) foo;
EXPLAIN (COSTS OFF)
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT array_agg((SELECT percent_rank(f) WITHIN GROUP (ORDER BY v) AS p FROM d)) p FROM unnest(ARRAY[950, 990]) f)
SELECT
p,
abs(a - b) < 0.01
FROM (
SELECT
unnest(ARRAY[950, 990]) AS p,
unnest((select x.p from x)) AS a,
unnest(tdigest_percentile_of(v, c, 100, ARRAY[950, 990])) AS b
FROM t) foo;
WITH
d AS (SELECT t.* FROM t, LATERAL generate_series(1,t.c)),
x AS (SELECT array_agg((SELECT percent_rank(f) WITHIN GROUP (ORDER BY v) AS p FROM d)) p FROM unnest(ARRAY[950, 990]) f)
SELECT
p,
abs(a - b) < 0.01
FROM (
SELECT
unnest(ARRAY[950, 990]) AS p,
unnest((select x.p from x)) AS a,
unnest(tdigest_percentile_of(v, c, 100, ARRAY[950, 990])) AS b
FROM t) foo;
|
