postgres=# create table t_hash as select id,md5(id::text) from generate_series(1,5000000) as id;
SELECT 5000000
postgres=# vacuum ANALYZE t_hash;
VACUUM
postgres=# \timing
Timing is on.
postgres=# select * from t_hash limit 10;
id | md5
----+----------------------------------
1 | c4ca4238a0b923820dcc509a6f75849b
2 | c81e728d9d4c2f636f067f89cc14862c
3 | eccbc87e4b5ce2fe28308fd9f2a7baf3
4 | a87ff679a2f3e71d9181a67b7542122c
5 | e4da3b7fbbce2345d7772b0674a318d5
6 | 1679091c5a880faf6fb5e6087eb1b2dc
7 | 8f14e45fceea167a5a36dedd4bea2543
8 | c9f0f895fb98ab9159f51fd0297e236d
9 | 45c48cce2e2d7fbdea1afc51c7c6ad26
10 | d3d9446802a44259755d38e6d163e820
(10 rows)
Time: 1.430 ms
postgres=# explain analyze select * from t_hash where md5 like '%923820dc%';
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------
Gather (cost=1000.00..68758.88 rows=500 width=37) (actual time=1.998..753.217 rows=1 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Parallel Seq Scan on t_hash (cost=0.00..67708.88 rows=208 width=37) (actual time=492.740..742.780 rows=0 loops=3)
Filter: (md5 ~~ '%923820dc%'::text)
Rows Removed by Filter: 1666666
Planning Time: 0.115 ms
Execution Time: 753.275 ms
(8 rows)
Time: 754.916 ms
postgres=# create extension pg_trgm ;
CREATE EXTENSION
postgres=# select show_trgm('c4ca4238a0b923820dcc509a6f75849b');
show_trgm
-----------------------------------------------------------------------------------------------------------------------------------------
{" c"," c4",09a,0b9,0dc,20d,238,382,38a,423,49b,4ca,509,584,6f7,758,820,849,8a0,923,9a6,"9b ",a0b,a42,a6f,b92,c4c,c50,ca4,cc5,dcc,f75}
(1 row)
Time: 12.006 ms
#创建gin索引
postgres=# create index idx_gin on t_hash using gin(md5 gin_trgm_ops);
CREATE INDEX
Time: 177973.977 ms (02:57.974)
postgres=# explain analyze select * from t_hash where md5 like '%ce2345d%';
QUERY PLAN
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on t_hash (cost=239.87..2074.79 rows=500 width=37) (actual time=9.299..9.358 rows=2 loops=1)
Recheck Cond: (md5 ~~ '%ce2345d%'::text)
Heap Blocks: exact=2
-> Bitmap Index Scan on idx_gin (cost=0.00..239.75 rows=500 width=0) (actual time=9.256..9.258 rows=2 loops=1)
Index Cond: (md5 ~~ '%ce2345d%'::text)
Planning Time: 0.710 ms
Execution Time: 9.394 ms
(7 rows)
postgres=# explain analyze select * from t_hash where md5 like '%9b%';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
Seq Scan on t_hash (cost=0.00..104167.00 rows=808081 width=37) (actual time=0.035..6246.231 rows=574238 loops=1)
Filter: (md5 ~~ '%9b%'::text)
Rows Removed by Filter: 4425762
Planning Time: 6.721 ms
Execution Time: 9816.262 ms
如果碰到Like 小于两个字符的时候,无法使用gin索引。比如like '%ab%'无法使用索引。但是如果‘%abc%’就可以使用索引。
postgres=# CREATE INDEX idx_gist ON t_hash USING gist (md5 gist_trgm_ops);
CREATE INDEX
postgres=# drop index idx_gin;
DROP INDEX
postgres=# DISCARD all;
DISCARD ALL
postgres=# explain analyze select * from t_hash where md5 like '%ce2345d%';
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on t_hash (cost=52.29..1887.21 rows=500 width=37) (actual time=808.728..808.738 rows=2 loops=1)
Recheck Cond: (md5 ~~ '%ce2345d%'::text)
Heap Blocks: exact=2
-> Bitmap Index Scan on idx_gist (cost=0.00..52.16 rows=500 width=0) (actual time=808.707..808.708 rows=2 loops=1)
Index Cond: (md5 ~~ '%ce2345d%'::text)
Planning Time: 0.220 ms
Execution Time: 808.855 ms
(7 rows)
测试发现,上述测试条件下,gin的效率要高很多。
对于上面gin索引两个字符无法使索引的问题,gist可以使用索引。
#gist索引情况
postgres=# explain analyze select * from t_hash where md5 ='1679091c5a880faf6fb5e6087eb1b2dc';
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------
Index Scan using idx_gist on t_hash (cost=0.41..8.43 rows=1 width=37) (actual time=36.534..77.858 rows=1 loops=1)
Index Cond: (md5 = '1679091c5a880faf6fb5e6087eb1b2dc'::text)
Planning Time: 0.117 ms
Execution Time: 77.885 ms
(4 rows)
postgres=# drop index idx_gist;
DROP INDEX
postgres=# create index idx_gin on t_hash using gin(md5 gin_trgm_ops);
CREATE INDEX
postgres=# discard all;
DISCARD ALL
#gin索引情况
postgres=# explain analyze select * from t_hash where md5 ='1679091c5a880faf6fb5e6087eb1b2dc';
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on t_hash (cost=1560.01..1564.02 rows=1 width=37) (actual time=28.292..28.293 rows=1 loops=1)
Recheck Cond: (md5 = '1679091c5a880faf6fb5e6087eb1b2dc'::text)
Heap Blocks: exact=1
-> Bitmap Index Scan on idx_gin (cost=0.00..1560.01 rows=1 width=0) (actual time=28.275..28.276 rows=1 loops=1)
Index Cond: (md5 = '1679091c5a880faf6fb5e6087eb1b2dc'::text)
Planning Time: 0.374 ms
Execution Time: 28.323 ms
(7 rows)
# btree索引情况
postgres=# create index idx_dx on t_hash(md5);
CREATE INDEX
postgres=# discard all;
DISCARD ALL
postgres=# explain analyze select * from t_hash where md5 ='1679091c5a880faf6fb5e6087eb1b2dc';
QUERY PLAN
----------------------------------------------------------------------------------------------------------------
Index Scan using idx_dx on t_hash (cost=0.56..8.57 rows=1 width=37) (actual time=0.034..0.038 rows=1 loops=1)
Index Cond: (md5 = '1679091c5a880faf6fb5e6087eb1b2dc'::text)
Planning Time: 0.127 ms
Execution Time: 0.060 ms
(4 rows)
测试情况:
gist:77.885 ms
gin:28.323 ms
btree:0.060 ms
测试结果:在=的测试中btree索引吊打。
postgres=# select pg_size_pretty(pg_total_relation_size('idx_dx'));
pg_size_pretty
----------------
282 MB
(1 row)
postgres=# select pg_size_pretty(pg_total_relation_size('idx_gin'));
pg_size_pretty
----------------
332 MB
postgres=# select pg_size_pretty(pg_total_relation_size('idx_gist'));
pg_size_pretty
----------------
885 MB
(1 row)
结论:gist索引更大。
#创建表
postgres=# CREATE TABLE t_fti (payload tsvector) WITH (autovacuum_enabled = off);
CREATE TABLE
#插入数据
postgres=# INSERT INTO t_fti
SELECT to_tsvector('english', md5('dummy' || id))
FROM generate_series(1, 2000000) AS id;
INSERT 0 2000000
postgres=# select * from t_fti limit 5;
payload
--------------------------------------
'8c2753548775b4161e531c323ea24c08':1
'c0c40e7a94eea7e2c238b75273087710':1
'ffdc12d8d601ae40f258acf3d6e7e1fb':1
'abc5fc01b06bef661bbd671bde23aa39':1
'20b70cebcb94b1c9ba30d17ab542a6dc':1
(5 rows)
#创建索引
postgres=# CREATE INDEX idx_fti ON t_fti USING gin(payload);
CREATE INDEX
#使用插件观察索引
postgres=# CREATE EXTENSION pgstattuple;
CREATE EXTENSION
#首次没有pending list
postgres=# SELECT * FROM pgstatginindex('idx_fti');
version | pending_pages | pending_tuples
---------+---------------+----------------
2 | 0 | 0
(1 row)
#再次插入数据
postgres=# INSERT INTO t_fti
SELECT to_tsvector('english', md5('dummy' || id))
FROM generate_series(2000001, 3000000) AS id;
INSERT 0 1000000
#pendling有数据,说明fastupate有效
postgres=# SELECT * FROM pgstatginindex('idx_fti');
version | pending_pages | pending_tuples
---------+---------------+----------------
2 | 326 | 50141
(1 row)
#vacuum后写入gin树中
postgres=# vacuum t_fti ;
VACUUM
postgres=# SELECT * FROM pgstatginindex('idx_fti');
version | pending_pages | pending_tuples
---------+---------------+----------------
2 | 0 | 0