共计 6199 个字符，预计需要花费 16 分钟才能阅读完成。
生产环境中有一支 SQL 消耗 cpu 资源很大，逻辑读为 299361，相关问题描述:
Rationale
The SQL spent 100% of its database time on CPU, I/O and Cluster waits.
This part of database time may be improved by the SQL Tuning Advisor.
Rationale
Database time for this SQL was divided as follows: 100% for SQL
execution, 0% for parsing, 0% for PL/SQL execution and 0% for Java
execution.
Rationale
SQL statement with SQL_ID “7ptu1y0d95s2r” was executed 3209 times and
had an average elapsed time of 20 seconds.
原 SQL:
update WWH
set WWH.STATUS = (case
when (select count(1)
from WWD
where
…………………….
and exists (select 1
from WWD
where 1 = 1
and WWH.WH = WWD.WH
and WWH.WAVE = WWD.WAVE
and WWD.SO in (‘2349212263452’)
and rownum = 1)
set line 1000
set pagesize 1000
set timing on
set autotrace traceonly
执行 SQL………
查看执行计划如下：
Execution Plan
———————————————————-
Plan hash value: 3044987130
————————————————————————————————–
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
————————————————————————————————–
| 0 | UPDATE STATEMENT | | 1 | 45 | 191K (1)| 00:38:18 |
| 1 | UPDATE | WWH | | | | |
|* 2 | FILTER | | | | | |
|* 3 | TABLE ACCESS FULL | WWH | 98070 | 4309K| 449 (1)| 00:00:06 |
|* 4 | COUNT STOPKEY | | | | | |
|* 5 | INDEX UNIQUE SCAN | UNQ_WWD | 1 | 40 | 2 (0)| 00:00:01 |
| 6 | SORT AGGREGATE | | 1 | 24 | | |
|* 7 | COUNT STOPKEY | | | | | |
|* 8 | FILTER | | | | | |
|* 9 | INDEX RANGE SCAN | UNQ_WWD| 1 | 24 | 3 (0)| 00:00:01 |
| 10 | SORT AGGREGATE | | 1 | 27 | | |
|* 11 | TABLE ACCESS BY INDEX ROWID| WWD | 1 | 27 | 4 (0)| 00:00:01 |
|* 12 | INDEX RANGE SCAN | UNQ_WWD| 1 | | 3 (0)| 00:00:01 |
————————————————————————————————–
Predicate Information (identified by operation id):
—————————————————
2 – filter(EXISTS (SELECT 0 FROM “WWD” WHERE ROWNUM=1 AND
“WWD”.”WH”=:B1 AND “WWD”.”SO”=’2349212263452′ AND “WWD”.”WAVE”=:B2))
-……………………………………..
………………………………………
Statistics
———————————————————-
1 recursive calls
0 db block gets
263080 consistent gets
0 physical reads
72 redo size
829 bytes sent via SQL*Net to client
3583 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
0 rows processed
上面显示表没走索引，逻辑读为 263080，查看表的字段 (WAVE,WH) 存在索引，我们再建一个选择性更好的索引：
增加索引：
create index IND_MODIFY_TIME_IX on WWH(Modify_Time,WAVE,WH) tablespace wsx;
再查看执行计划依然没有走索引，逻辑读同样是 263080 大。
强制加入索引后，虽然执行计划中有显示走了全索引扫描，且驱动表由 WWH 表变成了 WWD 表，但逻辑读变得更大为 299361，走索引前后性能都差的最主要原因是执行计划中看到有
“filter(EXISTS………..” 这行，表示子查询没有展开。
SQL> update /*+index(WWH IND_MODIFY_TIME_IX) */ WWH …………….
Elapsed: 00:00:01.11
Execution Plan
———————————————————-
Plan hash value: 389457693
—————————————————————————————————-
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
—————————————————————————————————-
| 0 | UPDATE STATEMENT | | 1 | 45 | 196K (1)| 00:39:16 |
| 1 | UPDATE | WWH | | | | |
|* 2 | TABLE ACCESS BY INDEX ROWID | WWD | 4903 | 215K| 5302 (1)| 00:01:04 |
|* 3 | INDEX FULL SCAN | IND_MODIFY_TIME_IX | 4996 | | 602 (1)| 00:00:08 |
|* 4 | COUNT STOPKEY | | | | | |
|* 5 | INDEX UNIQUE SCAN | UNQ_WWD | 1 | 40 | 2 (0)| 00:00:01 |
| 6 | SORT AGGREGATE | | 1 | 24 | | |
|* 7 | COUNT STOPKEY | | | | | |
|* 8 | FILTER | | | | | |
|* 9 | INDEX RANGE SCAN | UNQ_WWD | 1 | 24 | 3 (0)| 00:00:01 |
| 10 | SORT AGGREGATE | | 1 | 27 | | |
|* 11 | TABLE ACCESS BY INDEX ROWID| WWD | 1 | 27 | 4 (0)| 00:00:01 |
|* 12 | INDEX RANGE SCAN | UNQ_WWD | 1 | | 3 (0)| 00:00:01 |
—————————————————————————————————-
Predicate Information (identified by operation id):
—————————————————
2 – filter(EXISTS (SELECT 0 FROM “WWD” WHERE ROWNUM=1 AND
“WWD”.”WH”=:B1 AND “WWD”.”SO”=’2349212263452′ AND “WWD”.”WAVE”=:B2))
-……………………………………..
………………………………………
Statistics
———————————————————-
1 recursive calls
0 db block gets
299361 consistent gets
2 physical reads
0 redo size
840 bytes sent via SQL*Net to client
3619 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
2 sorts (memory)
0 sorts (disk)
0 rows processed
手工加入 /*+unnest*/ 让子查询展开，但从执行计划查看，依然没有展开。
exists 无法展开子查询，我们将 SQL 等价改成 in 子查询，如下：
SQL 略
Execution Plan
———————————————————-
Plan hash value: 1922347980
——————————————————————————————————–
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
——————————————————————————————————–
| 0 | UPDATE STATEMENT | | 1 | 81 | 616 (1)| 00:00:08 |
| 1 | UPDATE | WWH | | | | |
| 2 | NESTED LOOPS | | 1 | 81 | 607 (1)| 00:00:08 |
| 3 | NESTED LOOPS | | 1 | 81 | 607 (1)| 00:00:08 |
| 4 | VIEW | VW_NSO_1 | 1 | 36 | 4 (0)| 00:00:01 |
| 5 | SORT UNIQUE | | 1 | 40 | | |
|* 6 | COUNT STOPKEY | | | | | |
| 7 | TABLE ACCESS BY INDEX ROWID| WWD | 1 | 40 | 4 (0)| 00:00:01 |
|* 8 | INDEX RANGE SCAN | UNQ_WWD | 1 | | 3 (0)| 00:00:01 |
|* 9 | INDEX FULL SCAN | IND_MODIFY_TIME_IX | 1 | | 601 (1)| 00:00:08 |
|* 10 | TABLE ACCESS BY INDEX ROWID | WWH | 1 | 45 | 602 (1)| 00:00:08 |
| 11 | SORT AGGREGATE | | 1 | 24 | | |
|* 12 | COUNT STOPKEY | | | | | |
|* 13 | FILTER | | | | | |
|* 14 | INDEX RANGE SCAN | UNQ_WWD | 1 | 24 | 3 (0)| 00:00:01 |
| 15 | SORT AGGREGATE | | 1 | 27 | | |
|* 16 | TABLE ACCESS BY INDEX ROWID | WWD | 1 | 27 | 4 (0)| 00:00:01 |
|* 17 | INDEX RANGE SCAN | UNQ_WWD | 1 | | 3 (0)| 00:00:01 |
——————————————————————————————————–
Predicate Information (identified by operation id):
—————————————————
此部分看不到 exists 和 filter 信息了，说明有展开
Statistics
———————————————————-
1 recursive calls
0 db block gets
3 consistent gets
2 physical reads
0 redo size
842 bytes sent via SQL*Net to client
3546 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
3 sorts (memory)
0 sorts (disk)
0 rows processed
发现上面已走索引了，且逻辑读降为 3。
以上是在测试环境中模拟测试，因数据与正式环境有差异，我们再看下正式环境中查询操作性能比较：
select count(*) from WWH
where exists (select 1
from WWD
where 1 = 1
and WWH.WH= WWD.WH
and WWH.WAVE= WWD.WAVE
and WWD.SO in (‘SO201612345’)
and rownum = 1
)
and WWH.STATUS <> ’11’
and WWH.STATUS <> ’22’
— 逻辑读 625043，执行时间为 1 秒
改成 in 子查询后：
SELECT COUNT (*)
FROM WWH
WHERE (wwh.WH, wwh.WAVE) IN
(SELECT WWD.WH, WWD.WAVE
FROM WWD
WHERE 1 = 1
AND WWD.SO_NO IN (‘SO201612345’)
AND ROWNUM = 1
)
AND WWH.STATUS <> ’11’
and WWH.STATUS <> ’22’
— 逻辑读变为 8，执行时间 58 豪秒。
也可直接去掉 rownum=1。
总结：当 exists 子查询中出现有 rownum 时，子查询无法展开，可以改成 in 子查询或表联接，或根据实际逻辑需求，去掉 rownum。
更多 Oracle 相关信息见Oracle 专题页面 http://www.linuxidc.com/topicnews.aspx?tid=12
本文永久更新链接地址：http://www.linuxidc.com/Linux/2016-12/138202.htm
