sql - 如何在以下更新 sql 上进行 SQL 调优
问题描述
大家，最近我在做oracle的一个SQL调优任务，我觉得我遇到了一个非常困难的问题，我什至可以说我被这个问题吓到了，我从DBA那里得到了AWR报告，似乎是AMR的红线SQL需要做一些调整（我已经将这些SQL粘贴在下面，这个sql在SP中）。但是我不知道是什么导致了性能不佳，任何人都可以帮助提供一些关于调整SQL的解决方案或想法？
如果您需要 AWR 提供的更多证据，请告诉我。
提前致谢...
UPDATE tax_ratio tar
SET
( ecm,
esm,
epm,
ecam,
update_dt,
update_by ) = (
SELECT
nvl(src_t1.ecm,0) AS ecm,
nvl(src_t1.esm,0) AS esm,
nvl(src_t1.epm,0) AS epm,
nvl(src_t1.ecam,0) AS ecam,
SYSDATE,
'ffee_user'
FROM
(
SELECT
city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type,
SUM(new_tax_current_mth) /12 AS ecm,
SUM(new_tax_miss_current_mth) /12 AS esm,
SUM(new_tax_get_current_mth) /12 AS epm,
SUM(new_tax_special_current_mth) /12 AS ecam
FROM
tax_ratio
WHERE
city_code ='001'
AND company_type ='typ_01'
AND tax_mth <= add_months(TO_DATE('08-JUL-2015'),-3)
AND tax_mth >= add_months(TO_DATE('08-JUL-2015'),-14)
AND eff_date =TO_DATE('08-JUL-2015')
AND tax_type = '00'
GROUP BY
city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type
HAVING SUM(new_tax_current_mth) <> 0
OR SUM(new_tax_miss_current_mth) <> 0
OR SUM(new_tax_get_current_mth) <> 0
OR SUM(new_tax_special_current_mth) <> 0
) src_t1
WHERE
tar.city_code = src_t1.city_code
AND tar.tax_type = src_t1.tax_type
AND tar.company_type = src_t1.company_type
AND tar.taxpayer = src_t1.taxpayer
AND nvl(tar.company_group,'-99999') = nvl(src_t1.company_group,'-99999')
AND (
src_t1.ecm IS NOT NULL
OR src_t1.esm IS NOT NULL
OR src_t1.epm IS NOT NULL
OR src_t1.ecam IS NOT NULL
)
AND tar.tax_mth =TO_DATE('08-JUL-2015')
AND tar.company_tax_type = src_t1.company_tax_type
)
WHERE
tar.city_code ='001'
AND tar.company_type ='typ_01'
AND tar.tax_mth =TO_DATE('08-JUL-2015')
AND EXISTS (
SELECT
1
FROM
(
SELECT
city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type,
SUM(new_tax_current_mth) /12 AS ecm,
SUM(new_tax_miss_current_mth) /12 AS esm,
SUM(new_tax_get_current_mth) /12 AS epm,
SUM(new_tax_special_current_mth) /12 AS ecam
FROM
tax_ratio
WHERE
city_code ='001'
AND company_type ='typ_01'
AND tax_mth <= add_months(TO_DATE('08-JUL-2015'),-3)
AND tax_mth >= add_months(TO_DATE('08-JUL-2015'),-14)
AND eff_date =TO_DATE('08-Aug-2015')
AND tax_type = '00'
GROUP BY
city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type
HAVING SUM(new_tax_current_mth) <> 0
OR SUM(new_tax_miss_current_mth) <> 0
OR SUM(new_tax_get_current_mth) <> 0
OR SUM(new_tax_special_current_mth) <> 0
) src_t1
WHERE
tar.city_code = src_t1.city_code
AND tar.tax_type = src_t1.tax_type
AND tar.company_type = src_t1.company_type
AND tar.taxpayer = src_t1.taxpayer
AND nvl(tar.company_group,'-99999') = nvl(src_t1.company_group,'-99999')
AND (
src_t1.ecm IS NOT NULL
OR src_t1.esm IS NOT NULL
OR src_t1.epm IS NOT NULL
OR src_t1.ecam IS NOT NULL
)
AND tar.tax_mth =TO_DATE('08-JUL-2015')
AND tar.company_tax_type = src_t1.company_tax_type
)
add the EXPLAIN PLAN
PLAN HASH VALUE: 3650439649
----------------------------------------------------------------------------------------------------------------------
| ID | OPERATION | NAME | ROWS | BYTES |TEMPSPC| COST (%CPU)| TIME |
----------------------------------------------------------------------------------------------------------------------
| 0 | UPDATE STATEMENT | | 1698 | 179K| | 6169K (1)| 00:08:02 |
| 1 | UPDATE | TAX_RATIO | | | | | |
|* 2 | HASH JOIN RIGHT SEMI | | 1698 | 179K| | 732K (1)| 00:00:58 |
| 3 | VIEW | | 39251 | 1111K| | 371K (2)| 00:00:29 |
|* 4 | FILTER | | | | | | |
| 5 | SORT GROUP BY | | 39251 | 2414K| 100M| 371K (2)| 00:00:29 |
|* 6 | TABLE ACCESS FULL | TAX_RATIO | 1140K| 68M| | 365K (2)| 00:00:29 |
|* 7 | TABLE ACCESS FULL | TAX_RATIO | 207K| 15M| | 361K (1)| 00:00:29 |
| 8 | VIEW | | 1 | 81 | | 484 (1)| 00:00:01 |
|* 9 | FILTER | | | | | | |
| 10 | SORT GROUP BY | | 1 | 63 | | 484 (1)| 00:00:01 |
|* 11 | FILTER | | | | | | |
|* 12 | TABLE ACCESS BY INDEX ROWID BATCHED| TAX_RATIO | 1 | 63 | | 483 (0)| 00:00:01 |
|* 13 | INDEX RANGE SCAN | TAX_RATIO_TAXPAYER_IDX | 544 | | | 3 (0)| 00:00:01 |
----------------------------------------------------------------------------------------------------------------------
PREDICATE INFORMATION (IDENTIFIED BY OPERATION ID):
---------------------------------------------------
2 - ACCESS("TAR"."CITY_CODE"="SRC_T1"."CITY_CODE" AND "TAR"."TAX_TYPE"="SRC_T1"."TAX_TYPE" AND
"TAR"."COMPANY_TYPE"="SRC_T1"."COMPANY_TYPE" AND "TAR"."TAXPAYER"="SRC_T1"."TAXPAYER" AND
NVL("TAR"."COMPANY_GROUP",'-99999')=NVL("SRC_T1"."COMPANY_GROUP",'-99999') AND "TAR"."COMPANY_TAX_TYPE"="SRC_T1"."COMPANY_TAX_TYPE")
4 - FILTER((SUM("NEW_TAX_CURRENT_MTH")<>0 OR SUM("NEW_TAX_MISS_CURRENT_MTH")<>0 OR SUM("NEW_TAX_GET_CURRENT_MTH")<>0 OR SUM("NEW_TAX_SPECIAL_CURRENT_MTH")<>0) AND
(SUM("NEW_TAX_CURRENT_MTH")/12 IS NOT NULL OR SUM("NEW_TAX_MISS_CURRENT_MTH")/12 IS NOT NULL OR SUM("NEW_TAX_GET_CURRENT_MTH")/12 IS NOT NULL OR
SUM("NEW_TAX_SPECIAL_CURRENT_MTH")/12 IS NOT NULL))
6 - FILTER("COMPANY_TYPE"='LIMIT' AND "TAX_TYPE"='00' AND "TAX_MTH">=TO_DATE(' 2017-03-31 00:00:00',
'SYYYY-MM-DD HH24:MI:SS') AND "TAX_MTH"<=TO_DATE(' 2018-02-28 00:00:00', 'SYYYY-MM-DD HH24:MI:SS') AND
"CITY_CODE"='001' AND "NEW_TAX_MISS_CURRENT_MTH"=TO_DATE(' 2200-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS'))
7 - FILTER("TAR"."TAX_MTH"=TO_DATE(' 2018-05-31 00:00:00', 'SYYYY-MM-DD HH24:MI:SS') AND
"TAR"."COMPANY_TYPE"='LIMIT' AND "TAR"."CITY_CODE"='001')
9 - FILTER((SUM("NEW_TAX_CURRENT_MTH")<>0 OR SUM("NEW_TAX_MISS_CURRENT_MTH")<>0 OR SUM("NEW_TAX_GET_CURRENT_MTH")<>0 OR SUM("NEW_TAX_SPECIAL_CURRENT_MTH")<>0) AND
(SUM("NEW_TAX_CURRENT_MTH")/12 IS NOT NULL OR SUM("NEW_TAX_MISS_CURRENT_MTH")/12 IS NOT NULL OR SUM("NEW_TAX_GET_CURRENT_MTH")/12 IS NOT NULL OR
SUM("NEW_TAX_SPECIAL_CURRENT_MTH")/12 IS NOT NULL))
11 - FILTER(:B1=TO_DATE(' 2018-05-31 00:00:00', 'SYYYY-MM-DD HH24:MI:SS') AND :B2='LIMIT' AND :B3='00' AND
:B4='001')
12 - FILTER("COMPANY_TYPE"='LIMIT' AND "COMPANY_TAX_TYPE"=:B1 AND "TAX_TYPE"='00' AND "TAX_MTH">=TO_DATE('
2017-03-31 00:00:00', 'SYYYY-MM-DD HH24:MI:SS') AND "TAX_MTH"<=TO_DATE(' 2018-02-28 00:00:00', 'SYYYY-MM-DD
HH24:MI:SS') AND NVL("COMPANY_GROUP",'-99999')=NVL(:B2,'-99999') AND "CITY_CODE"='001' AND "NEW_TAX_MISS_CURRENT_MTH"=TO_DATE('
2200-01-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS'))
13 - ACCESS("TAXPAYER"=:B1)
解决方案
我要尝试的第一件事是使用 MERGE 语句而不是 UPDATE 语句 - 这让我想起了，因为您在 where 子句的 set 子句中有效地重复子查询。
我认为您的 UPDATE 可以重写为：
MERGE INTO tax_ratio tgt
USING (SELECT city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type,
SUM(new_tax_current_mth) / 12 AS ecm,
SUM(new_tax_miss_current_mth) / 12 AS esm,
SUM(new_tax_get_current_mth) / 12 AS epm,
SUM(new_tax_special_current_mth) / 12 AS ecam
FROM tax_ratio
WHERE city_code = '001'
AND company_type = 'typ_01'
AND tax_mth <= add_months(to_date('08-JUL-2015', 'dd-MON-yyyy', 'nls_date_language = english'), -3)
AND tax_mth >= add_months(to_date('08-JUL-2015', 'dd-MON-yyyy', 'nls_date_language = english'), -3)
AND eff_date = to_date('08-JUL-2015', 'dd-MON-yyyy', 'nls_date_language = english')
AND tax_type = '00'
GROUP BY city_code,
tax_type,
company_type,
taxpayer,
company_group,
company_tax_type
HAVING SUM(new_tax_current_mth) <> 0
OR SUM(new_tax_miss_current_mth) <> 0
OR SUM(new_tax_get_current_mth) <> 0
OR SUM(new_tax_special_current_mth) <> 0)) src
ON (tgt.city_code = src.city_code
AND tgt.tax_type = src.tax_type
AND tgt.company_type = src.company_type
AND tgt.taxpayer = src.taxpayer
AND NVL(tgt.company_group, '-99999') = NVL(src.company_group, '-99999')
--AND COALESCE(src.ecm, src.esm, src.epm, src.ecam) IS NOT NULL -- unnecessary, since your having clause excludes rows where all are null anyway
AND tgt.company_tax_type = src.company_tax_type)
WHEN MATCHED THEN
UPDATE SET tgt.ecm = NVL(src.ecm, 0),
tgt.esm = NVL(src.esm, 0),
tgt.epm = NVL(src.epm, 0),
tgt.ecam = NVL(src.ecam, 0),
tgt.update_dt = SYSDATE,
tgt.update_by = 'ffee_user')
WHERE tgt.tax_mth = to_date('08-JUL-2015', 'dd-MON-yyyy', 'nls_date_language = english');
笔记：
- 我将其更改
to_dates()为包含格式掩码（否则，如果您的 NLS_DATE_FORMAT nls 参数已更改，to_date()则会失败），并且由于您使用“JUL”作为月份，因此附加的第三个参数将日期作为字符串转换 nls 设置独立（例如，如果您的 NLS_DATE_FORMAT nls 参数设置JUL为不是有效的短月，to_date()如果未设置第三个参数，则会失败）。 您可以通过将日期作为数字传递来避免使用第三个参数，例如to_date('05/07/2015', 'dd/mm/yyyy'). - 我改为
and ecm is not null and esm is not null and ...使用COALESCE，因为它返回列表中的第一个非空值。这样你的支票就变成了and COALESCE(ecm, esm, ...) is not null。 - 实际上并不需要 COALESCE，因为您的 HAVING 子句有效地排除了所有值为 NULL 和 0 的行。
- 用于使子查询相关的谓词成为目标表和源子查询之间的连接条件，这意味着您不再需要之前在相关子查询中拥有的外部查询。
- 我将条件移动到仅将具有特定纳税月份的行从连接子句更新到更新部分的 WHERE 子句中。我很确定它可以留在 join 子句中，但我认为如果将其放在更新的 WHERE 子句中，意图会更清楚。
我会测试新的 MERGE 语句以确保它做正确的事情（或修复它直到它正确），然后看看它如何影响性能。
如果它仍然作为 AWR 中的问题陈述弹出，那么我将进一步研究如何调整它；可能需要更新/附加索引，可能需要物化视图等。
推荐阅读
- apache-kafka - Kafka Connect：用于有条件替换字段值的 SMT
- tomcat - 使用根 uri ("/") 部署 .war 文件
- jenkins - Kubernetes - 来自服务器的错误没有找到已知地址的首选地址 - AWS EKS
- html - 水平居中 div 的特定用例
- spring-boot - java.lang.ClassNotFoundException：org.springframework.boot.legacy.context.web.SpringBootContextLoaderListener
- c++ - 如何继续使用 macOS Catalina 10.15.4 处理 OpenGL？
- c - sscanf 与动态分配的字符串兼容吗？
- go - 如何在 Golang 中将 *uint16 指针传递给 windows.CreateFile()
- css - 从 css 规则中排除自定义类
- javascript - Materialise CSS 的自动完成功能不起作用（Javascript + Google Apps 脚本）