共计 9491 个字符，预计需要花费 24 分钟才能阅读完成。
以下内容适用于 Oracle 10.2.0.5 及其以上版本
一个查询很慢，原始 SQL 如下：
select
a.*
from (select
ssi.ID,
‘small_station_info’ TB,
(select sbi.name
from scene_base_info sbi
where sbi.id = ssi.antenna_selection) as antenna_selection,
ssi.antenna_height,
ssi.down_angle,
ssi.azimuth_angle,
ssi.ITI_ID,
sa.longitude,
sa.latitude,
sa.attach_id
from consolidation_demand cd
left join demand_test_info dti
on cd.id = dti.cd_id
left join demand_plan_info dpi
on dti.id = dpi.tdl_id
left join building_plan_info bpi
on dpi.id = bpi.dpi_id
left join NEAR_FAR_PLACE_INFO nfpi
on bpi.id = nfpi.bpi_id
left join SMALL_STATION_INFO ssi
on nfpi.id = ssi.nfpi_id
left join site_attachment sa
on TO_NUMBER(sa.longitude) is not null
AND TO_NUMBER(sa.latitude) > 26.074423
AND TO_NUMBER(sa.latitude) < 26.077573
AND TO_NUMBER(sa.longitude) > 119.191148
AND TO_NUMBER(sa.longitude) < 119.197649
AND sa.attach_name =
substr(ssi.AZIMUTH_ANGLE_PHOTO,
instr(ssi.AZIMUTH_ANGLE_PHOTO, ‘/’, -1) + 1,
length(ssi.AZIMUTH_ANGLE_PHOTO))) a
where a.longitude is not null
表都不大，执行计划如下：
已选择 12 行。
执行计划
———————————————————-
Plan hash value: 1917963167
—————————————————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
—————————————————————————————————————
| 0 | SELECT STATEMENT | | 1 | 253 | 519 (2)| 00:00:07 |
| 1 | TABLE ACCESS BY INDEX ROWID | SCENE_BASE_INFO | 1 | 14 | 1 (0)| 00:00:01 |
|* 2 | INDEX UNIQUE SCAN | SCENE_BASE_INFO_PK | 1 | | 0 (0)| 00:00:01 |
| 3 | VIEW | | 1 | 253 | 519 (2)| 00:00:07 |
|* 4 | FILTER | | | | | |
|* 5 | HASH JOIN OUTER | | 1 | 251 | 519 (2)| 00:00:07 |
|* 6 | HASH JOIN OUTER | | 83 | 8134 | 505 (1)| 00:00:07 |
|* 7 | HASH JOIN OUTER | | 83 | 7304 | 501 (1)| 00:00:07 |
|* 8 | HASH JOIN OUTER | | 83 | 6391 | 493 (1)| 00:00:06 |
|* 9 | HASH JOIN OUTER | | 83 | 5478 | 271 (1)| 00:00:04 |
| 10 | MERGE JOIN CARTESIAN | | 36 | 2052 | 21 (0)| 00:00:01 |
|* 11 | TABLE ACCESS BY INDEX ROWID| SITE_ATTACHMENT | 1 | 53 | 16 (0)| 00:00:01 |
|* 12 | INDEX RANGE SCAN | IDX_SITE_ATTACHMENT_JWD | 1 | | 15 (0)| 00:00:01 |
| 13 | BUFFER SORT | | 6725 | 26900 | 5 (0)| 00:00:01 |
| 14 | INDEX FAST FULL SCAN | PK_CONSOLIDATION_DEMAND | 6725 | 26900 | 5 (0)| 00:00:01 |
| 15 | TABLE ACCESS FULL | DEMAND_TEST_INFO | 15459 | 135K| 249 (1)| 00:00:03 |
| 16 | TABLE ACCESS FULL | DEMAND_PLAN_INFO | 8787 | 96657 | 221 (1)| 00:00:03 |
| 17 | TABLE ACCESS FULL | BUILDING_PLAN_INFO | 3244 | 35684 | 8 (0)| 00:00:01 |
| 18 | TABLE ACCESS FULL | NEAR_FAR_PLACE_INFO | 389 | 3890 | 3 (0)| 00:00:01 |
| 19 | TABLE ACCESS FULL | SMALL_STATION_INFO | 594 | 90882 | 13 (0)| 00:00:01 |
—————————————————————————————————————
Predicate Information (identified by operation id):
—————————————————
– access(“SBI”.”ID”=:B1)
– filter(“SA”.”ATTACH_NAME”=SUBSTR(“SSI”.”AZIMUTH_ANGLE_PHOTO”,INSTR(“SSI”.”AZIMUTH_ANGLE_PHOTO”,’
/’,-1)+1,LENGTH(“SSI”.”AZIMUTH_ANGLE_PHOTO”)))
– access(“NFPI”.”ID”=”SSI”.”NFPI_ID”(+))
– access(“BPI”.”ID”=”NFPI”.”BPI_ID”(+))
– access(“DPI”.”ID”=”BPI”.”DPI_ID”(+))
– access(“DTI”.”ID”=”DPI”.”TDL_ID”(+))
– access(“CD”.”ID”=”DTI”.”CD_ID”(+))
– filter(“SA”.”LONGITUDE” IS NOT NULL)
– access(TO_NUMBER(“LONGITUDE”)>119.191148 AND TO_NUMBER(“LATITUDE”)>26.074423 AND
TO_NUMBER(“LONGITUDE”)<119.197649 AND TO_NUMBER(“LATITUDE”)<26.077573)
filter(TO_NUMBER(“LONGITUDE”) IS NOT NULL AND TO_NUMBER(“LATITUDE”)<26.077573 AND
TO_NUMBER(“LATITUDE”)>26.074423)
这个执行计划，看起来无比正常，应该要 left join 的都有。
但问题的关键在于 10 步骤 – MERGE JOIN CARTESIAN。笛卡尔乘积的排序合并连接，这个需要耗费很长时间。
等待这个结果要耗费几十秒，甚至要更久！
如何解决这样的问题，有以下几个方法：
- 重新收集每个表的统计数据 – 这个没有实验过，但即使那么做，可能也无效。不过从本例看，很有可能是这个导致的。
- 启用 leading 提示，结合其它提示
- 使用 materialize 提示
使用 leading 提示
select /*+ no_merge(a) no_push_pred(a) */
a.*
from (select
/*+ leading(cd dti dpi bpi ssi) */
…) a
where a.longitude is not null
/
执行计划
已选择 12 行。
执行计划
———————————————————-
Plan hash value: 1844304918
———————————————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
———————————————————————————————————
| 0 | SELECT STATEMENT | | 1 | 253 | 520 (2)| 00:00:07 |
| 1 | TABLE ACCESS BY INDEX ROWID | SCENE_BASE_INFO | 1 | 14 | 1 (0)| 00:00:01 |
|* 2 | INDEX UNIQUE SCAN | SCENE_BASE_INFO_PK | 1 | | 0 (0)| 00:00:01 |
| 3 | VIEW | | 1 | 253 | 520 (2)| 00:00:07 |
|* 4 | HASH JOIN | | 1 | 251 | 520 (2)| 00:00:07 |
|* 5 | TABLE ACCESS BY INDEX ROWID| SITE_ATTACHMENT | 1 | 53 | 16 (0)| 00:00:01 |
|* 6 | INDEX RANGE SCAN | IDX_SITE_ATTACHMENT_JWD | 1 | | 15 (0)| 00:00:01 |
|* 7 | HASH JOIN RIGHT OUTER | | 23606 | 4564K| 503 (2)| 00:00:07 |
| 8 | TABLE ACCESS FULL | SMALL_STATION_INFO | 594 | 90882 | 13 (0)| 00:00:01 |
|* 9 | HASH JOIN RIGHT OUTER | | 15459 | 679K| 490 (2)| 00:00:06 |
| 10 | TABLE ACCESS FULL | NEAR_FAR_PLACE_INFO | 389 | 3890 | 3 (0)| 00:00:01 |
|* 11 | HASH JOIN RIGHT OUTER | | 15459 | 528K| 486 (2)| 00:00:06 |
| 12 | TABLE ACCESS FULL | BUILDING_PLAN_INFO | 3244 | 35684 | 8 (0)| 00:00:01 |
|* 13 | HASH JOIN RIGHT OUTER | | 15459 | 362K| 477 (1)| 00:00:06 |
| 14 | TABLE ACCESS FULL | DEMAND_PLAN_INFO | 8787 | 96657 | 221 (1)| 00:00:03 |
|* 15 | HASH JOIN OUTER | | 15459 | 196K| 255 (1)| 00:00:04 |
| 16 | INDEX FAST FULL SCAN | PK_CONSOLIDATION_DEMAND | 6725 | 26900 | 5 (0)| 00:00:01 |
| 17 | TABLE ACCESS FULL | DEMAND_TEST_INFO | 15459 | 135K| 249 (1)| 00:00:03 |
———————————————————————————————————
Predicate Information (identified by operation id):
—————————————————
– access(“SBI”.”ID”=:B1)
– access(“SA”.”ATTACH_NAME”=SUBSTR(“SSI”.”AZIMUTH_ANGLE_PHOTO”,INSTR(“SSI”.”AZIMUTH_ANGLE_PH
OTO”,’/’,-1)+1,LENGTH(“SSI”.”AZIMUTH_ANGLE_PHOTO”)))
– filter(“SA”.”LONGITUDE” IS NOT NULL)
– access(TO_NUMBER(“LONGITUDE”)>119.191148 AND TO_NUMBER(“LATITUDE”)>26.074423 AND
TO_NUMBER(“LONGITUDE”)<119.197649 AND TO_NUMBER(“LATITUDE”)<26.077573)
filter(TO_NUMBER(“LONGITUDE”) IS NOT NULL AND TO_NUMBER(“LATITUDE”)<26.077573 AND
TO_NUMBER(“LATITUDE”)>26.074423)
– access(“NFPI”.”ID”=”SSI”.”NFPI_ID”(+))
– access(“BPI”.”ID”=”NFPI”.”BPI_ID”(+))
– access(“DPI”.”ID”=”BPI”.”DPI_ID”(+))
– access(“DTI”.”ID”=”DPI”.”TDL_ID”(+))
– access(“CD”.”ID”=”DTI”.”CD_ID”(+))
没有笛卡尔的 merge join . 步骤 4 还是一个 hash join。
执行很快，大概可以 0.17 秒
使用 materialize 提示
1 WITH A AS
2 (select /*+MATERIALIZE */
3 .....)
4 select a.* from A WHERE a.longitude is not null执行计划
已选择 12 行。执行计划
----------------------------------------------------------
Plan hash value: 3536941173
--------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 23606 | 5832K| 700 (2)| 00:00:09 |
| 1 | TEMP TABLE TRANSFORMATION | | | | | |
| 2 | LOAD AS SELECT | | | | | |
|* 3 | HASH JOIN RIGHT OUTER | | 23606 | 5786K| 520 (2)| 00:00:07 |
| 4 | TABLE ACCESS BY INDEX ROWID| SITE_ATTACHMENT | 1 | 53 | 16 (0)| 00:00:01 |
|* 5 | INDEX RANGE SCAN | IDX_SITE_ATTACHMENT_JWD | 1 | | 15 (0)| 00:00:01 |
|* 6 | HASH JOIN RIGHT OUTER | | 23606 | 4564K| 503 (2)| 00:00:07 |
| 7 | TABLE ACCESS FULL | SMALL_STATION_INFO | 594 | 90882 | 13 (0)| 00:00:01 |
|* 8 | HASH JOIN RIGHT OUTER | | 15459 | 679K| 490 (2)| 00:00:06 |
| 9 | TABLE ACCESS FULL | NEAR_FAR_PLACE_INFO | 389 | 3890 | 3 (0)| 00:00:01 |
|* 10 | HASH JOIN RIGHT OUTER | | 15459 | 528K| 486 (2)| 00:00:06 |
| 11 | TABLE ACCESS FULL | BUILDING_PLAN_INFO | 3244 | 35684 | 8 (0)| 00:00:01 |
|* 12 | HASH JOIN RIGHT OUTER | | 15459 | 362K| 477 (1)| 00:00:06 |
| 13 | TABLE ACCESS FULL | DEMAND_PLAN_INFO | 8787 | 96657 | 221 (1)| 00:00:03 |
|* 14 | HASH JOIN OUTER | | 15459 | 196K| 255 (1)| 00:00:04 |
| 15 | INDEX FAST FULL SCAN | PK_CONSOLIDATION_DEMAND | 6725 | 26900 | 5 (0)| 00:00:01 |
| 16 | TABLE ACCESS FULL | DEMAND_TEST_INFO | 15459 | 135K| 249 (1)| 00:00:03 |
|* 17 | VIEW | | 23606 | 5832K| 180 (2)| 00:00:03 |
| 18 | TABLE ACCESS FULL | SYS_TEMP_0FD9D68A2_721EF047 | 23606 | 4103K| 180 (2)| 00:00:03 |
--------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("SA"."ATTACH_NAME"(+)=SUBSTR("SSI"."AZIMUTH_ANGLE_PHOTO",INSTR("SSI"."AZIMUTH_ANGLE_PHOT
O",'/',-1)+1,LENGTH("SSI"."AZIMUTH_ANGLE_PHOTO")))
5 - access(TO_NUMBER("LONGITUDE"(+))>119.191148 AND TO_NUMBER("LATITUDE"(+))>26.074423 AND
TO_NUMBER("LONGITUDE"(+))<119.197649 AND TO_NUMBER("LATITUDE"(+))<26.077573)
filter(TO_NUMBER("LONGITUDE"(+)) IS NOT NULL AND TO_NUMBER("LATITUDE"(+))<26.077573 AND
TO_NUMBER("LATITUDE"(+))>26.074423)
6 - access("NFPI"."ID"="SSI"."NFPI_ID"(+))
8 - access("BPI"."ID"="NFPI"."BPI_ID"(+))
10 - access("DPI"."ID"="BPI"."DPI_ID"(+))
12 - access("DTI"."ID"="DPI"."TDL_ID"(+))
14 - access("CD"."ID"="DTI"."CD_ID"(+))
17 - filter("A"."LONGITUDE" IS NOT NULL) 也很快，大约 0.19~0.2 左右。
之所以慢，主要是因为要先生成 gt 表 SYS_TEMP_0FD9D68A2_721EF047。
总结
1. 最好先收集统计数据
2. 在收集统计数据无效的情况下，考虑使用 leading 提示，其次 materialize 提示也会破坏 oracle 优化器一些自以为明智的计划（优化器的不足，oracle 已经提到了，这就是 hint 的由来 )
3.dba 要优化一个库，不是一个很容易的事情，需要做很多工作。
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/2017-08/146048.htm
