浅谈pg_hint_plan定制执行计划

有的时候pg给出的执行计划由于很多原因并不是最优的，需要手动指定执行路径时我们可以加载pg_hint_plan这个插件。

1 安装插件

预先安装postgresql10.7

cd postgresql-10.7/contrib/
wget https://github.com/ossc-db/pg_hint_plan/archive/rel10_1_3_3.tar.gz
tar xzvf pg_hint_plan-rel10_1_3_3.tar.gz
cd pg_hint_plan-rel10_1_3_3
make
make install

检查文件

cd $pghome
ls lib/pg_hint_plan.so
lib/pg_hint_plan.so
ls share/extension/
pg_hint_plan--1.3.0--1.3.1.sql pg_hint_plan--1.3.2--1.3.3.sql pg_hint_plan.control plpgsql.control
pg_hint_plan--1.3.1--1.3.2.sql pg_hint_plan--1.3.3.sql   plpgsql--1.0.sql  plpgsql--unpackaged--1.0.sql

2 加载插件

2.1 当前会话加载

load 'pg_hint_plan';

注意这样加载只在当前回话生效。

2.2 用户、库级自动加载

alter user postgres set session_preload_libraries='pg_hint_plan';
alter database postgres set session_preload_libraries='pg_hint_plan';

配置错了的话就连不上数据库了！

如果配置错了，连接template1库执行

alter database postgres reset session_preload_libraries;
alter user postgres reset session_preload_libraries;

2.3 cluster级自动加载

在postgresql.conf中修改shared_preload_libraries=‘pg_hint_plan'

重启数据库

3 检查是否已经加载

pg_hint_plan加载后在extension里面是看不到的，所以需要确认插件是否已经加载

show session_preload_libraries;
 session_preload_libraries
---------------------------
 pg_hint_plan

或者

show shared_preload_libraries;

如果使用load方式加载不需要检查。

4 使用插件定制执行计划

4.1 初始化测试数据

create table t1 (id int, t int, name varchar(255));
create table t2 (id int , salary int);
create table t3 (id int , age int);
insert into t1 values (1,200,'jack');
insert into t1 values (2,300,'tom');
insert into t1 values (3,400,'john');
insert into t2 values (1,40000);
insert into t2 values (2,38000);
insert into t2 values (3,18000);
insert into t3 values (3,38);
insert into t3 values (2,55);
insert into t3 values (1,12);
explain analyze select * from t1 left join t2 on t1.id=t2.id left join t3 on t1.id=t3.id;
              query plan
-------------------------------------------------------------------------------------------------------------------------
 hash right join (cost=89.82..337.92 rows=17877 width=540) (actual time=0.053..0.059 rows=3 loops=1)
 hash cond: (t3.id = t1.id)
 -> seq scan on t3 (cost=0.00..32.60 rows=2260 width=8) (actual time=0.002..0.002 rows=3 loops=1)
 -> hash (cost=70.05..70.05 rows=1582 width=532) (actual time=0.042..0.043 rows=3 loops=1)
   buckets: 2048 batches: 1 memory usage: 17kb
   -> hash right join (cost=13.15..70.05 rows=1582 width=532) (actual time=0.034..0.039 rows=3 loops=1)
    hash cond: (t2.id = t1.id)
    -> seq scan on t2 (cost=0.00..32.60 rows=2260 width=8) (actual time=0.002..0.002 rows=3 loops=1)
    -> hash (cost=11.40..11.40 rows=140 width=524) (actual time=0.017..0.017 rows=3 loops=1)
      buckets: 1024 batches: 1 memory usage: 9kb
      -> seq scan on t1 (cost=0.00..11.40 rows=140 width=524) (actual time=0.010..0.011 rows=3 loops=1)
 planning time: 0.154 ms
 execution time: 0.133 ms

创建索引

create index idx_t1_id on t1(id);
create index idx_t2_id on t2(id);
create index idx_t3_id on t3(id);
explain analyze select * from t1 left join t2 on t1.id=t2.id left join t3 on t1.id=t3.id;
             query plan
--------------------------------------------------------------------------------------------------------------
 hash left join (cost=2.14..3.25 rows=3 width=540) (actual time=0.045..0.047 rows=3 loops=1)
 hash cond: (t1.id = t3.id)
 -> hash left join (cost=1.07..2.14 rows=3 width=532) (actual time=0.030..0.032 rows=3 loops=1)
   hash cond: (t1.id = t2.id)
   -> seq scan on t1 (cost=0.00..1.03 rows=3 width=524) (actual time=0.005..0.006 rows=3 loops=1)
   -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.007..0.007 rows=3 loops=1)
    buckets: 1024 batches: 1 memory usage: 9kb
    -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)
 -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.005..0.005 rows=3 loops=1)
   buckets: 1024 batches: 1 memory usage: 9kb
   -> seq scan on t3 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.002 rows=3 loops=1)
 planning time: 0.305 ms
 execution time: 0.128 ms

4.2 强制走index scan

/*+ indexscan(t1 idx_d)
/*+ indexscan(t1 idx_t1_id)
explain (analyze,buffers) select * from t1 where id=2;
           query plan
----------------------------------------------------------------------------------------------
 seq scan on t1 (cost=0.00..1.04 rows=1 width=524) (actual time=0.011..0.013 rows=1 loops=1)
 filter: (id = 2)
 rows removed by filter: 2
 buffers: shared hit=1
 planning time: 0.058 ms
 execution time: 0.028 ms
explain (analyze,buffers) /*+ indexscan(t1) */select * from t1 where id=2;
             query plan
----------------------------------------------------------------------------------------------------------------
 index scan using idx_t1_id on t1 (cost=0.13..8.15 rows=1 width=524) (actual time=0.044..0.046 rows=1 loops=1)
 index cond: (id = 2)
 buffers: shared hit=1 read=1
 planning time: 0.145 ms
 execution time: 0.072 ms
explain (analyze,buffers) /*+ indexscan(t1 idx_t1_id) */select * from t1 where id=2;
             query plan
----------------------------------------------------------------------------------------------------------------
 index scan using idx_t1_id on t1 (cost=0.13..8.15 rows=1 width=524) (actual time=0.016..0.017 rows=1 loops=1)
 index cond: (id = 2)
 buffers: shared hit=2
 planning time: 0.079 ms
 execution time: 0.035 ms

4.3 强制多条件组合

/*+ indexscan(t2) indexscan(t1 idx_t1_id) */
/*+ seqscan(t2) indexscan(t1 idx_t1_id) */
explain analyze select * from t1 join t2 on (t1.id = t2.id);
            query plan
--------------------------------------------------------------------------------------------------------
 hash join (cost=1.07..2.14 rows=3 width=532) (actual time=0.018..0.020 rows=3 loops=1)
 hash cond: (t1.id = t2.id)
 -> seq scan on t1 (cost=0.00..1.03 rows=3 width=524) (actual time=0.006..0.007 rows=3 loops=1)
 -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.005..0.005 rows=3 loops=1)
   buckets: 1024 batches: 1 memory usage: 9kb
   -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.001..0.003 rows=3 loops=1)
 planning time: 0.114 ms
 execution time: 0.055 ms
(8 rows)

组合两个条件走indexscan

/*+ indexscan(t2) indexscan(t1 idx_t1_id) */explain analyze select * from t1 join t2 on (t1.id = t2.id);
              query plan
-----------------------------------------------------------------------------------------------------------------------
 merge join (cost=0.26..24.40 rows=3 width=532) (actual time=0.047..0.053 rows=3 loops=1)
 merge cond: (t1.id = t2.id)
 -> index scan using idx_t1_id on t1 (cost=0.13..12.18 rows=3 width=524) (actual time=0.014..0.015 rows=3 loops=1)
 -> index scan using idx_t2_id on t2 (cost=0.13..12.18 rows=3 width=8) (actual time=0.026..0.028 rows=3 loops=1)

组合两个条件走indexscan+seqscan

/*+ seqscan(t2) indexscan(t1 idx_t1_id) */explain analyze select * from t1 join t2 on (t1.id = t2.id);
              query plan
-----------------------------------------------------------------------------------------------------------------------
 nested loop (cost=0.13..13.35 rows=3 width=532) (actual time=0.025..0.032 rows=3 loops=1)
 join filter: (t1.id = t2.id)
 rows removed by join filter: 6
 -> index scan using idx_t1_id on t1 (cost=0.13..12.18 rows=3 width=524) (actual time=0.016..0.018 rows=3 loops=1)
 -> materialize (cost=0.00..1.04 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=3)
   -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.004..0.005 rows=3 loops=1)

4.4 强制指定join method

/*+ nestloop(t1 t2) mergejoin(t1 t2 t3) leading(t1 t2 t3) */
/*+ nestloop(t1 t2 t3) mergejoin(t2 t3) leading(t1 (t2 t3)) */
explain analyze select * from t1 left join t2 on t1.id=t2.id left join t3 on t1.id=t3.id;
             query plan
--------------------------------------------------------------------------------------------------------------
 hash left join (cost=2.14..3.25 rows=3 width=540) (actual time=0.053..0.056 rows=3 loops=1)
 hash cond: (t1.id = t3.id)
 -> hash left join (cost=1.07..2.14 rows=3 width=532) (actual time=0.036..0.038 rows=3 loops=1)
   hash cond: (t1.id = t2.id)
   -> seq scan on t1 (cost=0.00..1.03 rows=3 width=524) (actual time=0.007..0.007 rows=3 loops=1)
   -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.009..0.009 rows=3 loops=1)
    buckets: 1024 batches: 1 memory usage: 9kb
    -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)
 -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.006..0.006 rows=3 loops=1)
   buckets: 1024 batches: 1 memory usage: 9kb
   -> seq scan on t3 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)

强制走循环嵌套连接

/*+ nestloop(t1 t2) mergejoin(t1 t2 t3) leading(t1 t2 t3) */
explain analyze select * from t1 left join t2 on t1.id=t2.id left join t3 on t1.id=t3.id;
              query plan
--------------------------------------------------------------------------------------------------------------------
 merge left join (cost=3.28..3.34 rows=3 width=540) (actual time=0.093..0.096 rows=3 loops=1)
 merge cond: (t1.id = t3.id)
 -> sort (cost=2.23..2.23 rows=3 width=532) (actual time=0.077..0.078 rows=3 loops=1)
   sort key: t1.id
   sort method: quicksort memory: 25kb
   -> nested loop left join (cost=0.00..2.20 rows=3 width=532) (actual time=0.015..0.020 rows=3 loops=1)
    join filter: (t1.id = t2.id)
    rows removed by join filter: 6
    -> seq scan on t1 (cost=0.00..1.03 rows=3 width=524) (actual time=0.005..0.005 rows=3 loops=1)
    -> materialize (cost=0.00..1.04 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=3)
      -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)
 -> sort (cost=1.05..1.06 rows=3 width=8) (actual time=0.012..0.013 rows=3 loops=1)
   sort key: t3.id
   sort method: quicksort memory: 25kb
   -> seq scan on t3 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)

控制连接顺序

/*+ nestloop(t1 t2 t3) mergejoin(t2 t3) leading(t1 (t2 t3)) */
explain analyze select * from t1 left join t2 on t1.id=t2.id left join t3 on t1.id=t3.id;
query plan
--------------------------------------------------------------------------------------------------------------
 nested loop left join (cost=1.07..3.31 rows=3 width=540) (actual time=0.036..0.041 rows=3 loops=1)
 join filter: (t1.id = t3.id)
 rows removed by join filter: 6
 -> hash left join (cost=1.07..2.14 rows=3 width=532) (actual time=0.030..0.032 rows=3 loops=1)
   hash cond: (t1.id = t2.id)
   -> seq scan on t1 (cost=0.00..1.03 rows=3 width=524) (actual time=0.008..0.009 rows=3 loops=1)
   -> hash (cost=1.03..1.03 rows=3 width=8) (actual time=0.007..0.007 rows=3 loops=1)
    buckets: 1024 batches: 1 memory usage: 9kb
    -> seq scan on t2 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.004 rows=3 loops=1)
 -> materialize (cost=0.00..1.04 rows=3 width=8) (actual time=0.001..0.002 rows=3 loops=3)
   -> seq scan on t3 (cost=0.00..1.03 rows=3 width=8) (actual time=0.002..0.003 rows=3 loops=1)

4.5 控制单条sql的cost

/*+ set(seq_page_cost 20.0) seqscan(t1) */
/*+ set(seq_page_cost 20.0) seqscan(t1) */explain analyze select * from t1 where id > 1;
           query plan
-----------------------------------------------------------------------------------------------
 seq scan on t1 (cost=0.00..20.04 rows=1 width=524) (actual time=0.011..0.013 rows=2 loops=1)
 filter: (id > 1)
 rows removed by filter: 1

set seq_page_cost 200，注意下面的cost已经变成了200.04

/*+ set(seq_page_cost 200.0) seqscan(t1) */explain analyze select * from t1 where id > 1;
           query plan
------------------------------------------------------------------------------------------------
 seq scan on t1 (cost=0.00..200.04 rows=1 width=524) (actual time=0.010..0.011 rows=2 loops=1)
 filter: (id > 1)
 rows removed by filter: 1

以上为个人经验，希望能给大家一个参考，也希望大家多多支持。如有错误或未考虑完全的地方，望不吝赐教。