1. 背景
PolarDB 的云原生存算分离架构, 具备低廉的数据存储、高效扩展弹性、高速多机并行计算能力、高速数据搜索和处理; PolarDB与计算算法结合, 将实现双剑合璧, 推动业务数据的 价值产出, 将数据变成生产力.
本文将介绍PolarDB 开源版通过 parray_gin 实现高效率 数组、JSON 内元素的模糊搜索
测试环境为macos+docker, polardb部署请参考如何用 PolarDB 证明巴菲特的投资理念 - 包括PolarDB简单部署。
2. 测试
PG 模糊搜索采用GIN倒排索引, 使用pg_trgm插件将字符串前1后2加上空格后, 按连续3个字符切分, 并对切分后的token建立token,ctid的倒排索引.
在模糊搜索时, 可以将搜索字符串按同样方式切分, 根据倒排搜索快速的定位到对应的ctid.
《PostgreSQL 模糊查询、相似查询 (like '%xxx%') pg_bigm 比 pg_trgm 优势在哪?》
《PostgreSQL 模糊查询增强插件pgroonga , pgbigm (含单字、双字、多字、多字节字符) - 支持JSON模糊查询等》
《PostgreSQL ghtree实现的海明距离排序索引, 性能不错(模糊图像) - pg-knn_hamming - bit string 比特字符串 相似度搜索》
《PostgreSQL VagueGeometry vague spatial data - VASA (Vague Spatial Algebra) for PG - 模糊空间数据》
《PostgreSQL bktree 索引using gist例子 - 海明距离检索 - 短文相似、模糊图像搜索 - bit string 比特字符串 相似度搜索》
《PostgreSQL 一复合查询SQL优化例子 - (多个exists , 范围检索 , IN检索 , 模糊检索 组合)》
《从难缠的模糊查询聊开 - PostgreSQL独门绝招之一 GIN , GiST , SP-GiST , RUM 索引原理与技术背景》
即使没有parray_gin, 我们也可以将数组或JSON格式化处理后, 用大字符串和pg_trgm来实现元素模糊搜索. 例如
array['abc','aaa','hello']
把元素内容的sep char和quote char转义, 然后直接把 'abc','aaa','hello'当成字符串处理. 建立pg_trgm gin索引.
搜索元素时如果需要指定元素前缀或后缀搜索, 那么带上sep char和quote char即可. 使用parray_gin就简单多了, 不需要处理那么多.
下面测试PolarDB+parray_gin 实现数组内元素的模糊搜索.
git clone --depth 1 http://github.com/theirix/parray_gin/
cd parray_gin/
USE_PGXS=1 make
USE_PGXS=1 make install
export PGHOST=localhost
[postgres@1bbb8082aa60 parray_gin]$ psql
psql (11.9)
Type "help" for help.
postgres=# \q
[postgres@1bbb8082aa60 parray_gin]$ USE_PGXS=1 make installcheck
/home/postgres/tmp_basedir_polardb_pg_1100_bld/lib/pgxs/src/makefiles/../../src/test/regress/pg_regress --inputdir=./ --bindir='/home/postgres/tmp_basedir_polardb_pg_1100_bld/bin' --inputdir=test --dbname=contrib_regression op index
(using postmaster on localhost, default port)
============== dropping database "contrib_regression" ==============
NOTICE: database "contrib_regression" does not exist, skipping
DROP DATABASE
============== creating database "contrib_regression" ==============
CREATE DATABASE
ALTER DATABASE
============== running regression test queries ==============
test op ... ok
test index ... ok
==========================================================
All 2 tests passed.
POLARDB:
All 2 tests, 0 tests in ignore, 0 tests in polar ignore.
========================================================== create table t (id int, info text[]);
create or replace function gen_text_arr(int) returns text[] as $$
select array(select md5(random()::text) from generate_series(1,$1));
$$ language sql strict;
postgres=# select gen_text_arr(10);
-[ RECORD 1 ]+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
gen_text_arr | {4134ee81fcdc29da486df37a1725e1cc,d0bb424307f93a6374d1af5a4b1c0451,def4b4bc24bc6aefb084df8a1571d773,aff17d39b2c3e8ccebf1059c2cd466dc,3988cb3f89372081c6444b7f8a825cf6,77d3a12d9a5159bd2e11fac1782eaf90,0ecac2cd508f60221b31934ea1128223,622819cfa7c3e3e600f70ed90265edaa,e9311e8d6f23be74b2e73eae4408aaa8,207eb23a50212cb101f83a6041211d90}
postgres=# insert into t select id , gen_text_arr(10) from generate_series(1,1000) id;
INSERT 0 1000
postgres=# select * from t where info @@> array['%4b1%'];
id |
info
-----+----------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------
14 | {745b761d7145edb79904c5217c0ec0b4,eab9d9d4de9afc8c7a2bc4cdcd3bcb2a,3116cd48046936709c56e952f5d50380,642eec5d3c17721dadb89759ac116821,49ba14c3c71b73c0a3b8
6aa6f20a4f9c,01632c5889d4ae642422fea8620187e1,078ea7bf29a6f8bf53c6abcec98df5ad,2548e08ad3cb87dfcfe55a86e47cc60f,0c7002203e72d854f9c0643bec6c59b7,cfdd57d32f4bcee
8b4b1adfe11a08a81}
33 | {639e7f990ef271b24b1ac1a1f154476b,5c0dd44f87821cf555fb579f2dd9871d,b3118d34a6f788ad9c9d3343743900bc,798abd4aece1cbe604e608294227dde6,f08757d02fd0db9d08c9
2240c55ec14b,54f206220cf2097f0e2a6f630a7871be,585d04664a022ab49607d0d6ff18fc89,f5681d20b2b923973652f9952df6b71d,1d204241c105c78ba0514bdf1dba6bbb,5f427b5c2b65e0d
e41b70e804dfcc41d}
...
postgres=# select * from t where info @@> array['%4b1ac%'];
id |
info
----+-----------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------
33 | {639e7f990ef271b24b1ac1a1f154476b,5c0dd44f87821cf555fb579f2dd9871d,b3118d34a6f788ad9c9d3343743900bc,798abd4aece1cbe604e608294227dde6,f08757d02fd0db9d08c92
240c55ec14b,54f206220cf2097f0e2a6f630a7871be,585d04664a022ab49607d0d6ff18fc89,f5681d20b2b923973652f9952df6b71d,1d204241c105c78ba0514bdf1dba6bbb,5f427b5c2b65e0de
41b70e804dfcc41d}
(1 row)
postgres=# select * from t where info @@> array['%4b1acd%'];
id | info
----+------
(0 rows)
postgres=# explain select * from t where info @@> array['%4b1ac%'];
QUERY PLAN
--------------------------------------------------------------------------
Bitmap Heap Scan on t (cost=28.01..32.02 rows=1 width=36)
Recheck Cond: (info @@> '{%4b1ac%}'::text[])
-> Bitmap Index Scan on t_info_idx (cost=0.00..28.01 rows=1 width=0)
Index Cond: (info @@> '{%4b1ac%}'::text[])
(4 rows)
postgres=# explain (analyze,verbose,timing,costs,buffers) select * from t where info @@> array['%4b1ac%', '%8fc89'];
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on public.t (cost=60.01..64.02 rows=1 width=36) (actual time=0.121..0.122 rows=1 loops=1)
Output: id, info
Recheck Cond: (t.info @@> '{%4b1ac%,%8fc89}'::text[])
Heap Blocks: exact=1
Buffers: shared hit=16
-> Bitmap Index Scan on t_info_idx (cost=0.00..60.01 rows=1 width=0) (actual time=0.109..0.109 rows=1 loops=1)
Index Cond: (t.info @@> '{%4b1ac%,%8fc89}'::text[])
Buffers: shared hit=15
Planning Time: 0.075 ms
Execution Time: 0.144 ms
(10 rows)
Time: 0.699 ms
postgres=# select * from t where info @@> array['%4b1ac%', '%8fc89'];
id |
info
----+-----------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------
33 | {639e7f990ef271b24b1ac1a1f154476b,5c0dd44f87821cf555fb579f2dd9871d,b3118d34a6f788ad9c9d3343743900bc,798abd4aece1cbe604e608294227dde6,f08757d02fd0db9d08c92
240c55ec14b,54f206220cf2097f0e2a6f630a7871be,585d04664a022ab49607d0d6ff18fc89,f5681d20b2b923973652f9952df6b71d,1d204241c105c78ba0514bdf1dba6bbb,5f427b5c2b65e0de
41b70e804dfcc41d}
(1 row)
Time: 0.733 ms
postgres=# insert into t select id , gen_text_arr(10) from generate_series(1,120000) id;
INSERT 0 100000
Time: 9242.877 ms (00:09.243)
postgres=# \dt+
List of relations
Schema | Name | Type | Owner | Size | Description
--------+------+-------+----------+-------+-------------
public | t | table | postgres | 50 MB |
(1 row)
postgres=# select * from t where info @@> array['%4b1ac%', '%8fc89'];
id |
info
----+-----------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------
33 | {639e7f990ef271b24b1ac1a1f154476b,5c0dd44f87821cf555fb579f2dd9871d,b3118d34a6f788ad9c9d3343743900bc,798abd4aece1cbe604e608294227dde6,f08757d02fd0db9d08c92
240c55ec14b,54f206220cf2097f0e2a6f630a7871be,585d04664a022ab49607d0d6ff18fc89,f5681d20b2b923973652f9952df6b71d,1d204241c105c78ba0514bdf1dba6bbb,5f427b5c2b65e0de
41b70e804dfcc41d}
(1 row)
Time: 4.783 ms
postgres=# explain (analyze,timing,costs,buffers,verbose) select * from t where info @@> array['%4b1ac%', '%8fc89'];
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on public.t (cost=96.94..529.04 rows=121 width=36) (actual time=4.114..4.115 rows=1 loops=1)
Output: id, info
Recheck Cond: (t.info @@> '{%4b1ac%,%8fc89}'::text[])
Heap Blocks: exact=1
Buffers: shared hit=48
-> Bitmap Index Scan on t_info_idx (cost=0.00..96.91 rows=121 width=0) (actual time=4.103..4.103 rows=1 loops=1)
Index Cond: (t.info @@> '{%4b1ac%,%8fc89}'::text[])
Buffers: shared hit=47
Planning Time: 0.090 ms
Execution Time: 4.170 ms
(10 rows) 全表扫描性能差了几十倍
postgres=# set enable_bitmapscan =off;
SET
Time: 0.473 ms
postgres=# explain (analyze,timing,costs,buffers,verbose) select * from t where info @@> array['%4b1ac%', '%8fc89'];
QUERY PLAN
----------------------------------------------------------------------------------------------------------
Seq Scan on public.t (cost=0.00..7881.50 rows=121 width=36) (actual time=0.632..193.929 rows=1 loops=1)
Output: id, info
Filter: (t.info @@> '{%4b1ac%,%8fc89}'::text[])
Rows Removed by Filter: 120999
Buffers: shared hit=6229 read=140
Planning Time: 0.081 ms
Execution Time: 193.947 ms
(7 rows)
Time: 194.697 ms
postgres=# select * from t where info @@> array['%4b1ac%', '%8fc89'];
id |
info
----+-----------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------
33 | {639e7f990ef271b24b1ac1a1f154476b,5c0dd44f87821cf555fb579f2dd9871d,b3118d34a6f788ad9c9d3343743900bc,798abd4aece1cbe604e608294227dde6,f08757d02fd0db9d08c92
240c55ec14b,54f206220cf2097f0e2a6f630a7871be,585d04664a022ab49607d0d6ff18fc89,f5681d20b2b923973652f9952df6b71d,1d204241c105c78ba0514bdf1dba6bbb,5f427b5c2b65e0de
41b70e804dfcc41d}
(1 row)
Time: 199.342 ms 有了parray_gin, 在设计数据结构时, 可以更加灵活, 例如将“一个时间段、一个组、一个对象”的“多个标签、多个信息”打包成1行数组存储, 对数组进行元素搜索, 则可以快速匹配到符合条件的“一个时间段、一个组、一个对象”.