如何让in/exists 子查询(半连接)作为驱动表？
Posted robinson1988
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了如何让in/exists 子查询(半连接)作为驱动表？相关的知识，希望对你有一定的参考价值。
一哥们问我，怎么才能让子查询作为驱动表？ SQL如下：
select rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/执行计划如下：
Execution Plan
----------------------------------------------------------
Plan hash value: 3634433140
--------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
--------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 2 (50)| 00:00:01 | | |
| 1 | SORT ORDER BY | | 1 | 111 | 2 (50)| 00:00:01 | | |
| 2 | NESTED LOOPS | | | | | | | |
| 3 | NESTED LOOPS | | 1 | 111 | 1 (0)| 00:00:01 | | |
| 4 | PARTITION RANGE SINGLE | | 1 | 39 | 1 (0)| 00:00:01 | 1284 | 1284 |
|* 5 | INDEX SKIP SCAN | IDX_VT7_DEVICEID | 1 | 39 | 1 (0)| 00:00:01 | 1284 | 1284 |
|* 6 | INDEX UNIQUE SCAN | PK_ITS_BASE_DEVICE | 1 | | 0 (0)| 00:00:01 | | |
|* 7 | TABLE ACCESS BY INDEX ROWID| ITS_BASE_DEVICE | 1 | 72 | 0 (0)| 00:00:01 | | |
--------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
5 - access("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
filter("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
6 - access("V"."PASS_DEVICE_UNID"="UNID")
7 - filter("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393' AND "DEV_TYPE"='1' AND
"DEV_DELETE_FLAG"='N' AND "DEV_CHK_FLAG"='1')
Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
110973 consistent gets
0 physical reads
0 redo size
47861 bytes sent via SQL*Net to client
1656 bytes received via SQL*Net from client
105 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
1560 rows processed这里我们就不管统计信息是否准确了，也不管SQL优化的问题，就单单讨论哥们问的问题吧。
那哥们说，怎么才能让子查询作为驱动表呢？他自己试了很多方法就是搞不定。 那我们来亲自搞搞吧
explain plan for select rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/执行计划如下
SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY(NULL, NULL, 'ADVANCED -PROJECTION'));
-----------------------------------------------------------
Plan hash value: 2191740724
---------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
---------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 2092K (1)| 06:58:26 | | |
| 1 | NESTED LOOPS | | | | | | | |
| 2 | NESTED LOOPS | | 1 | 111 | 2092K (1)| 06:58:26 | | |
| 3 | PARTITION RANGE SINGLE | | 1 | 39 | 2092K (1)| 06:58:26 | 1284 | 1284 |
| 4 | TABLE ACCESS BY LOCAL INDEX ROWID| ITS_CAR_PASS7 | 1 | 39 | 2092K (1)| 06:58:26 | 1284 | 1284 |
|* 5 | INDEX RANGE SCAN | IDX_VT7_DATETIME | 1 | | 6029 (1)| 00:01:13 | 1284 | 1284 |
|* 6 | INDEX UNIQUE SCAN | PK_ITS_BASE_DEVICE | 1 | | 0 (0)| 00:00:01 | | |
|* 7 | TABLE ACCESS BY INDEX ROWID | ITS_BASE_DEVICE | 1 | 72 | 0 (0)| 00:00:01 | | |
---------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$5DA710D3
4 - SEL$5DA710D3 / V@SEL$1
5 - SEL$5DA710D3 / V@SEL$1
6 - SEL$5DA710D3 / ITS_BASE_DEVICE@SEL$2
7 - SEL$5DA710D3 / ITS_BASE_DEVICE@SEL$2
Outline Data
-------------
/*+
BEGIN_OUTLINE_DATA
NLJ_BATCHING(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2")
USE_NL(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2")
LEADING(@"SEL$5DA710D3" "V"@"SEL$1" "ITS_BASE_DEVICE"@"SEL$2")
INDEX(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2" ("ITS_BASE_DEVICE"."UNID"))
INDEX_RS_ASC(@"SEL$5DA710D3" "V"@"SEL$1" ("ITS_CAR_PASS7"."PASS_DATETIME"))
OUTLINE(@"SEL$2")
OUTLINE(@"SEL$1")
UNNEST(@"SEL$2")
OUTLINE_LEAF(@"SEL$5DA710D3")
FIRST_ROWS
DB_VERSION('11.2.0.3')
OPTIMIZER_FEATURES_ENABLE('11.2.0.3')
IGNORE_OPTIM_EMBEDDED_HINTS
END_OUTLINE_DATA
*/
Predicate Information (identified by operation id):
---------------------------------------------------
5 - access("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
6 - access("V"."PASS_DEVICE_UNID"="UNID")
7 - filter("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393' AND "DEV_TYPE"='1' AND "DEV_DELETE_FLAG"='N' AND
"DEV_CHK_FLAG"='1')
TMD 执行计划又变了，我们也先别管执行计划为啥变了，驱动表仍然是 ITS_CAR_PASS7，现在我们来改变驱动表
select /*+ leading(ITS_BASE_DEVICE@SEL$2) */ rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/
Execution Plan
----------------------------------------------------------
Plan hash value: 712001411
-------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
-------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 25 (4)| 00:00:01 | | |
| 1 | SORT ORDER BY | | 1 | 111 | 25 (4)| 00:00:01 | | |
|* 2 | TABLE ACCESS BY LOCAL INDEX ROWID| ITS_CAR_PASS7 | 1 | 39 | 2 (0)| 00:00:01 | 1284 | 1284 |
| 3 | NESTED LOOPS | | 1 | 111 | 24 (0)| 00:00:01 | | |
|* 4 | TABLE ACCESS BY INDEX ROWID | ITS_BASE_DEVICE | 6 | 432 | 12 (0)| 00:00:01 | | |
|* 5 | INDEX RANGE SCAN | IDX_DEVICE_DEV_BAY_UNID | 7 | | 1 (0)| 00:00:01 | | |
| 6 | PARTITION RANGE SINGLE | | 44M| | 2 (0)| 00:00:01 | 1284 | 1284 |
|* 7 | INDEX RANGE SCAN | IDX_VT7_PASS_DEVICE_UNID | 44M| | 2 (0)| 00:00:01 | 1284 | 1284 |
-------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
4 - filter("DEV_TYPE"='1' AND "DEV_DELETE_FLAG"='N' AND "DEV_CHK_FLAG"='1')
5 - access("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393')
7 - access("V"."PASS_DEVICE_UNID"="UNID")
Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
18645 consistent gets
130 physical reads
0 redo size
47861 bytes sent via SQL*Net to client
1657 bytes received via SQL*Net from client
105 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
1560 rows processed
驱动表改了之后，逻辑读从11W降低为1.8W
以上是关于如何让in/exists 子查询(半连接)作为驱动表？的主要内容，如果未能解决你的问题，请参考以下文章
HiveSql&SparkSql —— 使用left semi join做inexists类型子查询优化
sql中的in与not in,exists与not exists的区别
sql中的in与not in,exists与not exists的区别
sql中的in与not in,exists与not exists的区别