一、索引列存在多个or连接
当查询条件中存在多个OR连接时,PostgreSQL需要将所有条件的结果集进行合并,而这个合并操作可能会导致索引失效。
01
模拟环境
postgres=# create table idxidx as select * from pg_class;
SELECT 445
postgres=# create index idx_11 on idxidx(oid);
CREATE INDEX
02
测试情况
一个or连接两个索引列(走索引)
postgres=# explain analyze select oid,relname,relnamespace from idxidx where oid =17726 or oid=17743;
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on idxidx (cost=8.56..14.14 rows=2 width=72) (actual time=0.018..0.019 rows=1 loops=1)
Recheck Cond: ((oid = '17726'::oid) OR (oid = '17743'::oid))
Heap Blocks: exact=1
-> BitmapOr (cost=8.56..8.56 rows=2 width=0) (actual time=0.012..0.013 rows=0 loops=1)
-> Bitmap Index Scan on idx_11 (cost=0.00..4.28 rows=1 width=0) (actual time=0.011..0.011 rows=1 loops=1)
Index Cond: (oid = '17726'::oid)
-> Bitmap Index Scan on idx_11 (cost=0.00..4.28 rows=1 width=0) (actual time=0.001..0.001 rows=0 loops=1)
Index Cond: (oid = '17743'::oid)
Planning Time: 0.061 ms
Execution Time: 0.038 ms
(10 rows)
两个or连接三个索引列(走全表扫描)
postgres=# explain analyze select oid,relname,relnamespace from idxidx where oid =17726 or oid=17765 or oid=17743;
QUERY PLAN
--------------------------------------------------------------------------------------------------
Seq Scan on idxidx (cost=0.00..19.79 rows=3 width=72) (actual time=0.012..0.064 rows=1 loops=1)
Filter: ((oid = '17726'::oid) OR (oid = '17765'::oid) OR (oid = '17743'::oid))
Rows Removed by Filter: 444
Planning Time: 0.059 ms
Execution Time: 0.079 ms
(5 rows)
要避免这种情况,可以尝试对查询条件进行重写,例如使用UNION ALL连接多个查询条件,例如如如下这种方式
postgres=# explain analyze select oid,relname,relnamespace from idxidx where oid =17726 union all select oid,relname,relnamespace from idxidx where oid=17765 union all select oid,relname,relnamespace from idxidx where oid=17743;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------
Append (cost=0.27..24.92 rows=3 width=72) (actual time=0.041..0.046 rows=1 loops=1)
-> Index Scan using idx_11 on idxidx (cost=0.27..8.29 rows=1 width=72) (actual time=0.041..0.042 rows=1 loops=1)
Index Cond: (oid = '17726'::oid)
-> Index Scan using idx_11 on idxidx idxidx_1 (cost=0.27..8.29 rows=1 width=72) (actual time=0.002..0.002 rows=0 loops=1)
Index Cond: (oid = '17765'::oid)
-> Index Scan using idx_11 on idxidx idxidx_2 (cost=0.27..8.29 rows=1 width=72) (actual time=0.001..0.001 rows=0 loops=1)
Index Cond: (oid = '17743'::oid)
Planning Time: 0.169 ms
Execution Time: 0.082 ms
(9 rows)
二、数据量太小
对于非常小的表或者索引,使用索引可能会比全表扫描更慢。这是因为使用索引需要进行额外的 I/O 操作,而这些操作可能比直接扫描表更慢。
01
模拟环境
postgres=# create table tn(id int,name varchar);
CREATE TABLE
postgres=# insert into tn values(1,'ysl');
INSERT 0 1
postgres=# insert into tn values(2,'ysl');
INSERT 0 1
postgres=# insert into tn values(2,'ysll');
INSERT 0 1
postgres=# insert into tn values(2,'ysll');
INSERT 0 1
postgres=# create index idx_tn on tn(id);
CREATE INDEX
postgres=# \d tn
Table "public.tn"
Column | Type | Collation | Nullable | Default
--------+-------------------+-----------+----------+---------
id | integer | | |
name | character varying | | |
Indexes:
"idx_tn" btree (id)
postgres=# select * from tn;
id | name
----+------
1 | ysl
2 | ysl
2 | ysll
2 | ysll
(4 rows)
02
测试
postgres=# explain analyze select * from tn where id=2;
QUERY PLAN
---------------------------------------------------------------------------------------------
Seq Scan on tn (cost=0.00..1.05 rows=1 width=36) (actual time=0.007..0.007 rows=3 loops=1)
Filter: (id = 2)
Rows Removed by Filter: 1
Planning Time: 0.053 ms
Execution Time: 0.021 ms
(5 rows)
postgres=# explain analyze select * from tn where id=1;
QUERY PLAN
---------------------------------------------------------------------------------------------
Seq Scan on tn (cost=0.00..1.05 rows=1 width=36) (actual time=0.011..0.012 rows=1 loops=1)
Filter: (id = 1)
Rows Removed by Filter: 3
Planning Time: 0.057 ms
Execution Time: 0.026 ms
(5 rows
三、选择性不好
如果索引列中有大量重复的数据,或者一个字段全是一个值,这个时候,索引可能并不能发挥它的作用,起到加快检索的作用,因为这个索引并不能显著地减少需要扫描的行数,所以计算的代价可能远远大于走别的执行计划的代价。
基数:数据库基数是指数据库中不同值的数量
select count(distinct column_name) from table_name;
选择性:基数和总行数的比值再乘以100%就是某个列的选择性。
select count(distinct column_name) /count(column_name)* 100% from table_name;
01
模拟环境
postgres=# create table tb_t1 as select * from pg_class;
SELECT 465
postgres=# create index idx_tb_t1 on tb_t1(oid);
CREATE INDEX
02
测试
可以看到,原本oid这一列,选择性较好,分布较均匀的时候,可以正常使用到索引。而选择性不好的情况下,则
postgres=# explain analyze select * from tb_t1 where oid=17726;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on tb_t1 (cost=4.29..9.86 rows=2 width=236) (actual time=0.024..0.025 rows=1 loops=1)
Recheck Cond: (oid = '17726'::oid)
Heap Blocks: exact=1
-> Bitmap Index Scan on idx_tb_t1 (cost=0.00..4.29 rows=2 width=0) (actual time=0.021..0.021 rows=1 loops=1)
Index Cond: (oid = '17726'::oid)
Planning Time: 0.220 ms
Execution Time: 0.059 ms
(7 rows)
postgres=# update tb_t1 set oid=1;
UPDATE 465
postgres=# reindex index idx_tb_t1;
REINDEX
postgres=# explain analyze select * from tb_t1 where oid=1;
QUERY PLAN
------------------------------------------------------------------------------------------------------
Seq Scan on tb_t1 (cost=0.00..29.81 rows=465 width=274) (actual time=0.013..0.080 rows=465 loops=1)
Filter: (oid = '1'::oid)
Planning Time: 0.344 ms
Execution Time: 0.111 ms
(4 rows)
上边的这个例子,在我做完update后,列的基数是select count(distinct oid) from tb_t1;也就是1。而选择性是select count(distinct oid)/count(oid) 100% from tb_t1;也就是1/465 100% 选择性特别低。索引不能起到减少扫描的行数,反而在原本的基础上多了回表的动作,代价就增多了。因此CBO没有选择走这个索引的执行计划。
四、查询条件模糊
如果查询条件模糊,例如使用了不等于(<>)、LIKE等运算符或者使用了函数等,那么索引可能无法被使用。
因为正常情况下,等于(=)操作符可以直接利用B-tree或哈希索引进行查找。这是因为,这些操作符只需要在索引树中查找与给定值相等的项,就可以快速地定位到符合条件的记录。
而不等于(<>)操作符则需要查找所有不符合条件的记录,这会导致需要遍历整个索引树来找到匹配的记录,因此使用索引的成本比全表扫描更高。
LIKE操作符也可能导致不使用索引。这是因为,LIKE操作符通常需要执行模糊匹配,即查找包含你给的关键字的记录。虽然可以使用B-tree索引进行模糊匹配,但是如果模式以通配符开头(例如’%abc’),则索引将不会被使用,因为这种情况下需要遍历整个索引树来查找符合条件的记录。
这两种方式在列上有索引的时候,都是不能显著地减少需要扫描的行数。甚至加大SQL执行的代价,因此可能上边的索引不会被CBO选择为最后最优的执行计划。
01
模拟环境
postgres=# create table tb_l1 as select * from pg_class;
SELECT 465
postgres=# create index idx_tb_l1 on tb_l1(oid);
CREATE INDEX
02
测试
postgres=# explain analyze select * from tb_l1 where oid=17726;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
Index Scan using idx_tb_l1 on tb_l1 (cost=0.27..8.29 rows=1 width=274) (actual time=0.029..0.030 rows=1 loops=1)
Index Cond: (oid = '17726'::oid)
Planning Time: 0.473 ms
Execution Time: 0.083 ms
(4 rows)
postgres=# explain analyze select * from tb_l1 where oid<>17726;
QUERY PLAN
------------------------------------------------------------------------------------------------------
Seq Scan on tb_l1 (cost=0.00..17.81 rows=464 width=274) (actual time=0.007..0.103 rows=464 loops=1)
Filter: (oid <> '17726'::oid)
Rows Removed by Filter: 1
Planning Time: 0.069 ms
Execution Time: 0.132 ms
(5 rows)
五、表的一个列上有重复索引
在PostgreSQL里,是允许在一列上建立多个索引的,也就是如下这种方式,是不会报错说索引重复的,这也就导致了,使用过程中表上可能存在多余的重复索引,索引不会全部被使用到,而且可能引起性能问题。
postgres=# create index idx_tb_l1 on tb_l1(oid);
CREATE INDEX
postgres=# create index idx_tb_l2 on tb_l1(oid);
CREATE INDEX
01
模拟环境
postgres=# create table tb_l1 as select * from pg_class;
SELECT 465
postgres=# create index idx_tb_l1 on tb_l1(oid);
CREATE INDEX
postgres=# create index idx_tb_l2 on tb_l1(oid);
CREATE INDEX
postgres=# \d tb_l1
Table "public.tb_l1"
Column | Type | Collation | Nullable | Default
---------------------+--------------+-----------+----------+---------
oid | oid | | |
relname | name | | |
relnamespace | oid | | |
reltype | oid | | |
reloftype | oid | | |
relowner | oid | | |
... ...
... ...
Indexes:
"idx_tb_l1" btree (oid)
"idx_tb_l2" btree (oid)
02
测试
postgres=# explain analyze select * from tb_l1 where oid=17726; QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
Index Scan using idx_tb_l2 on tb_l1 (cost=0.27..8.29 rows=1 width=274) (actual time=0.025..0.025 rows=1 loops=1)
Index Cond: (oid = '17726'::oid)
Planning Time: 0.364 ms
Execution Time: 0.043 ms
(4 rows)
测试可以看到,在一个表的同一列上的两个索引其实作用是一样的,仅仅名字不一样,属于重复索引,这种情况下,就算用到索引,同一时刻也就会使用到一个索引。
使用如下的SQL可以找到数据库里的重复索引,可以定期巡检的时候进行检查,并在确认后合理优化掉重复的索引
SELECT
indrelid :: regclass AS table_name,
array_agg(indexrelid :: regclass) AS indexes
FROM pg_index
GROUP BY
indrelid, indkey
HAVING COUNT(*) > 1;
一个执行的结果如下所示:
postgres=# SELECT
indrelid :: regclass AS table_name,
array_agg(indexrelid :: regclass) AS indexes
FROM pg_index
GROUP BY
indrelid, indkey
HAVING COUNT(*) > 1;
table_name | indexes
------------+-----------------------
tb_l1 | {idx_tb_l1,idx_tb_l2}
t1 | {ind1,idx2}
(2 rows)
postgres=# \di+ idx_tb_l1
List of relations
Schema | Name | Type | Owner | Table | Persistence | Access method | Size | Description
--------+-----------+-------+---------+-------+-------------+---------------+-------+-------------
public | idx_tb_l1 | index | xmaster | tb_l1 | permanent | btree | 32 kB |
(1 row)
postgres=# \di+ idx_tb_l2
List of relations
Schema | Name | Type | Owner | Table | Persistence | Access method | Size | Description
--------+-----------+-------+---------+-------+-------------+---------------+-------+-------------
public | idx_tb_l2 | index | xmaster | tb_l1 | permanent | btree | 32 kB |
(1 row)
六、优化器选项关闭了索引扫描
PostgreSQL里有着很多的可以影响优化器的参数,例如enable_indexscan,enable_bitmapscan,enable_hashjoin,enable_sort等等,这些参数可以在session,用户,数据库级别进行设置。可以通过设置这些参数的值,来改变相关SQL执行时的执行计划。但是需要注意的是,为了个别的SQL,去盲目改变这些参数的值,往往是得不偿失的,操作的时候需要严谨并且仔细考虑,否则,这些类型的参数的改变,对于数据库的性能影响可能是巨大的。
01
模拟环境
postgres=# create table tb_l1 as select * from pg_class;
SELECT 465
postgres=# create index idx_tb_l1 on tb_l1(oid);
CREATE INDEX
02
测试
开启了对应优化器选项
postgres=# show enable_indexscan ;
enable_indexscan
------------------
on
(1 row)
postgres=# show enable_bitmapscan ;
enable_bitmapscan
-------------------
on
(1 row)
postgres=# explain analyze select * from tb_l1 where oid=17721;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
Index Scan using idx_tb_l2 on tb_l1 (cost=0.27..8.29 rows=1 width=274) (actual time=0.017..0.018 rows=1 loops=1)
Index Cond: (oid = '17721'::oid)
Planning Time: 0.088 ms
Execution Time: 0.038 ms
(4 rows)
关闭对应的优化器选项,可以看到CBO受到设置的参数的影响,选择了seq scan的执行计划,而没有用到字段上的索引。
postgres=# set enable_indexscan=off;
SET
postgres=# set enable_bitmapscan=off;
SET
postgres=# explain analyze select * from tb_l1 where oid=17721;
QUERY PLAN
--------------------------------------------------------------------------------------------------
Seq Scan on tb_l1 (cost=0.00..17.81 rows=1 width=274) (actual time=0.024..0.137 rows=1 loops=1)
Filter: (oid = '17721'::oid)
Rows Removed by Filter: 464
Planning Time: 0.079 ms
Execution Time: 0.192 ms
(5 rows)
七、统计信息不准确
因为CBO本身是基于代价的优化器,而计算代价要根据统计信息去做计算,统计信息不准确,得到的执行计划可能不是最优,这一点不做具体的举例。
八、Hints影响执行计划
PostgreSQL数据库里有着像ORACLE里类似的Hints功能,即pg_hint_plan工具,用Hints能够改变sql语句的执行计划,hint就是优化器的一种指示。虽然功能上和效果是类似的,但是PostgreSQL和ORACLE的Hints并不完全一致的,例如全表扫描等的关键字是不同的,需要进行区分。
01
准备环境
数据库需安装pg_hint_plan插件
postgres=# create table test_hint(id int,c varchar(100));
CREATE TABLE
postgres=# insert into test_hint select i,'test'||i from generate_series(1,10000) i;
INSERT 0 10000
postgres=# create index idx_test_hint_id on test_hint(id);
CREATE INDEX
02
测试
默认会走索引扫描,但是使用了hint,让其走了seqscan,没有使用到对应的字段上的索引。
postgres=# explain analyze select * from test_hint where id=10;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------
Index Scan using idx_test_hint_id on test_hint (cost=0.29..8.30 rows=1 width=12) (actual time=0.008..0.008 rows=1 loops=1)
Index Cond: (id = 10)
Planning Time: 0.111 ms
Execution Time: 0.024 ms
(4 rows)
postgres=# explain analyze select /*+seqscan(t) */ * from test_hint t where id=10;
QUERY PLAN
--------------------------------------------------------------------------------------------------------
Seq Scan on test_hint t (cost=0.00..180.00 rows=1 width=12) (actual time=0.022..2.691 rows=1 loops=1)
Filter: (id = 10)
Rows Removed by Filter: 9999
Planning Time: 0.311 ms
Execution Time: 2.712 ms
(5 rows)
九、查询条件中使用函数
当查询条件中包含函数调用时,PostgreSQL里可能无法使用索引,因为它需要对所有数据进行计算,而不是只计算索引值。
01
准备环境
postgres=# CREATE TABLE test_table (
postgres(# id SERIAL PRIMARY KEY,
postgres(# name TEXT,
postgres(# age INTEGER
postgres(# );
CREATE TABLE
postgres=#
postgres=# CREATE INDEX age_index ON test_table(age);
CREATE INDEX
postgres=# INSERT INTO test_table (name, age) VALUES
postgres-# ('Alice', 25),
postgres-# ('Bob', 30),
postgres-# ('Charlie', 35),
postgres-# ('David', 40),
postgres-# ('Eve', 45),
postgres-# ('Frank', 50);
INSERT 0 6
postgres=# CREATE OR REPLACE FUNCTION search_age(p_age INTEGER) RETURNS SETOF test_table AS $$
postgres$# BEGIN
postgres$# RETURN QUERY SELECT * FROM test_table WHERE age > p_age;
postgres$# END;
postgres$# $$ LANGUAGE plpgsql;
CREATE FUNCTION
02
测试
可以看到,当查询条件中包含函数调用时,没有使用到索引,而是使用了一个 Function Scan。这个Function Scan也是一种特殊的扫描方式,是从函数中获取数据。PostgreSQL会调用指定的函数来处理查询结果,并且会为函数的输出结果创建一个虚拟的关系表,以便后续的节点可以使用这个关系表继续执行查询。
postgres=# EXPLAIN ANALYZE SELECT * FROM test_table WHERE age > 35;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_table (cost=7.25..22.25 rows=400 width=40) (actual time=0.008..0.009 rows=3 loops=1)
Recheck Cond: (age > 35)
Heap Blocks: exact=1
-> Bitmap Index Scan on age_index (cost=0.00..7.15 rows=400 width=0) (actual time=0.004..0.004 rows=3 loops=1)
Index Cond: (age > 35)
Planning Time: 0.075 ms
Execution Time: 0.027 ms
(7 rows)
postgres=# EXPLAIN ANALYZE SELECT * FROM search_age(35);
QUERY PLAN
--------------------------------------------------------------------------------------------------------------
Function Scan on search_age (cost=0.25..10.25 rows=1000 width=40) (actual time=0.147..0.147 rows=3 loops=1)
Planning Time: 0.027 ms
Execution Time: 0.162 ms
(3 rows)
十、查询条件中有不等于运算符
因为在索引扫描期间,不等于运算符会导致索引中的每一行都需要进行比较,因此需要走全表扫描,不会走索引。
01
环境准备
postgres=# create table tb_l1 as select * from pg_class;
SELECT 465
postgres=# create index idx_tb_l1 on tb_l1(oid);
CREATE INDEX
02
测试
postgres=# explain analyze select * from tb_l1 where oid<>17721;
QUERY PLAN
------------------------------------------------------------------------------------------------------
Seq Scan on tb_l1 (cost=0.00..17.81 rows=464 width=274) (actual time=0.007..0.063 rows=464 loops=1)
Filter: (oid <> '17721'::oid)
Rows Removed by Filter: 1
Planning Time: 0.064 ms
Execution Time: 0.091 ms
(5 rows)
postgres=# explain analyze select * from tb_l1 where oid =17721;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
Index Scan using idx_tb_l2 on tb_l1 (cost=0.27..8.29 rows=1 width=274) (actual time=0.019..0.021 rows=1 loops=1)
Index Cond: (oid = '17721'::oid)
Planning Time: 0.107 ms
Execution Time: 0.051 ms
(4 rows)
postgres=# explain analyze select * from tb_l1 where oid !=17721;
QUERY PLAN
------------------------------------------------------------------------------------------------------
Seq Scan on tb_l1 (cost=0.00..17.81 rows=464 width=274) (actual time=0.012..0.072 rows=464 loops=1)
Filter: (oid <> '17721'::oid)
Rows Removed by Filter: 1
Planning Time: 0.071 ms
Execution Time: 0.102 ms
(5 rows)