两个左连接SQL执行计划解析（Oracle和PGSQL对比）：
上一篇解析链接如下：
https://www.cnblogs.com/wcwen1990/p/9325968.html
1、SQL示例1：
SQL> select *
from (
select * from tmp1 where c >= 1
) t1 left join (
select * from tmp2 where b < 30
) t2 on t1.a = t2.a
and t2.d > 1 and t1.e >= 2
where t1.b < 50
;
A B C E A B D E
---------- ---------- ---------- ---------- ---------- ---------- ---------- ----------
2 20 2 2 2 20 2 2
4 40 4 4
3 30 3 3
1 10 1 1
Execution Plan
----------------------------------------------------------
Plan hash value: 2592321047
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 4 | 416 | 7 (15)| 00:00:01 |
|* 1 | HASH JOIN OUTER | | 4 | 416 | 7 (15)| 00:00:01 |
|* 2 | TABLE ACCESS FULL| TMP1 | 4 | 208 | 3 (0)| 00:00:01 |
|* 3 | TABLE ACCESS FULL| TMP2 | 1 | 52 | 3 (0)| 00:00:01 |
---------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("TMP1"."A"="TMP2"."A"(+))
filter("TMP1"."E">=CASE WHEN ("TMP2"."A"(+) IS NOT NULL) THEN 2
ELSE 2 END )
2 - filter("TMP1"."B"<50 AND "C">=1)
3 - filter("TMP2"."D"(+)>1 AND "B"(+)<30)
Note
-----
- dynamic sampling used for this statement (level=2)
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
7 consistent gets
0 physical reads
0 redo size
1082 bytes sent via SQL*Net to client
524 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
4 rows processed
postgres=# explain analyze select *
postgres-# from (
postgres(# select * from tmp1 where c >= 1
postgres(# ) t1 left join (
postgres(# select * from tmp2 where b < 30
postgres(# ) t2 on t1.a = t2.a
postgres-# and t2.d > 1 and t1.e >= 2
postgres-# where t1.b < 50
postgres-# ;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------
Hash Left Join (cost=34.90..80.00 rows=181 width=32) (actual time=0.021..0.035 rows=4 loops=1)
Hash Cond: ("outer".a = "inner".a)
Join Filter: ("outer".e >= 2)
-> Seq Scan on tmp1 (cost=0.00..34.45 rows=181 width=16) (actual time=0.006..0.011 rows=4 loops=1)
Filter: ((c >= 1) AND (b < 50))
-> Hash (cost=34.45..34.45 rows=181 width=16) (actual time=0.007..0.007 rows=1 loops=1)
-> Seq Scan on tmp2 (cost=0.00..34.45 rows=181 width=16) (actual time=0.002..0.003 rows=1 loops=1)
Filter: ((b < 30) AND (d > 1))
Total runtime: 0.063 ms
(9 rows)
SQL执行计划的分析：
1) 全表扫描左表TMP1，同时根据TMP1表子查询条件"C">=1和where过滤条件"T1"."B"<50联合过滤，即filter("TMP1"."B"<50 AND "C">=1)，计算结果临时表记为tmp1；
2) 全表扫描右表TMP2，同时根据TMP2表子查询条件"B"(+)<30和on子句"T2"."D"(+)>1联合过滤，即filter("TMP2"."D"(+)>1 AND "B"(+)<30)，计算结果临时表记为tmp2；
3) 左表TMP1及右表TMP2处理后临时表tmp1和tmp2通过access("TMP1"."A"="TMP2"."A"(+))连接条件进行Hash Left Join操作，左临时表结果集全量返回，右表不匹配行置为null，返回结果临时表记为tmp3；
4) 返回结果集。
2、SQL示例2：
SQL> select *
from (
select * from tmp1 where c >= 1
) t1 left join (
select * from tmp2 where b < 30
) t2 on t1.a = t2.a
and t2.d > 1 and t1.e >= 2
where t1.b < 50 and t2.e <= 3
;
A B C E A B D E
---------- ---------- ---------- ---------- ---------- ---------- ---------- ----------
2 20 2 2 2 20 2 2
Execution Plan
----------------------------------------------------------
Plan hash value: 1630095649
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 104 | 7 (15)| 00:00:01 |
|* 1 | HASH JOIN | | 1 | 104 | 7 (15)| 00:00:01 |
|* 2 | TABLE ACCESS FULL| TMP2 | 1 | 52 | 3 (0)| 00:00:01 |
|* 3 | TABLE ACCESS FULL| TMP1 | 3 | 156 | 3 (0)| 00:00:01 |
---------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("TMP1"."A"="TMP2"."A")
2 - filter("TMP2"."E"<=3 AND "TMP2"."D">1 AND "B"<30)
3 - filter("TMP1"."B"<50 AND "TMP1"."E">=2 AND "C">=1)
Note
-----
- dynamic sampling used for this statement (level=2)
Statistics
----------------------------------------------------------
9 recursive calls
0 db block gets
15 consistent gets
0 physical reads
0 redo size
981 bytes sent via SQL*Net to client
524 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1 rows processed
SQL>
postgres=# select *
postgres-# from (
postgres(# select * from tmp1 where c >= 1
postgres(# ) t1 left join (
postgres(# select * from tmp2 where b < 30
postgres(# ) t2 on t1.a = t2.a
postgres-# and t2.d > 1 and t1.e >= 2
postgres-# where t1.b < 50 and t2.e <= 3
postgres-# ;
a | b | c | e | a | b | d | e
---+----+---+---+---+----+---+---
2 | 20 | 2 | 2 | 2 | 20 | 2 | 2
(1 row)
postgres=# explain analyze select *
postgres-# from (
postgres(# select * from tmp1 where c >= 1
postgres(# ) t1 left join (
postgres(# select * from tmp2 where b < 30
postgres(# ) t2 on t1.a = t2.a
postgres-# and t2.d > 1 and t1.e >= 2
postgres-# where t1.b < 50 and t2.e <= 3
postgres-# ;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Hash Join (cost=38.68..78.43 rows=18 width=32) (actual time=0.033..0.041 rows=1 loops=1)
Hash Cond: ("outer".a = "inner".a)
-> Seq Scan on tmp1 (cost=0.00..38.53 rows=60 width=16) (actual time=0.007..0.011 rows=3 loops=1)
Filter: ((c >= 1) AND (e >= 2) AND (b < 50))
-> Hash (cost=38.53..38.53 rows=60 width=16) (actual time=0.008..0.008 rows=1 loops=1)
-> Seq Scan on tmp2 (cost=0.00..38.53 rows=60 width=16) (actual time=0.003..0.005 rows=1 loops=1)
Filter: ((b < 30) AND (d > 1) AND (e <= 3))
Total runtime: 0.070 ms
(8 rows)
postgres=#
SQL执行计划的分析：
1) 全表扫描左表TMP2，同时根据TMP2表子查询条件"B"<30和where过滤条件"TMP2"."E"<=3及ON子句过滤条件"TMP2"."D">1联合过滤，即filter("TMP2"."E"<=3 AND "TMP2"."D">1 AND "B"<30)，计算结果临时表记为tmp1；
2) 全表扫描右表TMP1，同时根据TMP1表子查询条件"C">=1和where子句过滤条件"TMP1"."B"<50及ON子句"TMP1"."E">=2联合过滤，即filter("TMP1"."B"<50 AND "TMP1"."E">=2 AND "C">=1)，计算结果临时表记为tmp2；
3) 临时表tmp1和tmp2通过access("TMP1"."A"="TMP2"."A")连接条件进行Hash Join连接操作（此处left join写法已经被转换为内链接），返回匹配结果临时表记为tmp3；
4) 返回结果集。
