尽管成本低、基数低，但 Oracle 中的选择查询需要很长时间
Posted
技术标签:
【中文标题】尽管成本低、基数低，但 Oracle 中的选择查询需要很长时间【英文标题】：Select query in Oracle taking a long time despite low cost and low cardinality 【发布时间】：2019-09-03 09:17:19 【问题描述】：尝试执行选择查询，该查询基本上使用正则表达式将列中的多个值转换为多行 -
select INCIDENTID, trim(regexp_substr(CASUALFACTORS,'[^,]+', 1, level) ) value
FROM PSA.HSERS_INCIDENT_PSA WHERE ACTIVE_FLAG='1'
AND ISDELETED = 'F'
connect by regexp_substr(CASUALFACTORS, '[^,]+', 1, level) is not null
order by INCIDENTID
在解释计划中查询的成本为 3206，基数为 1,08,849。
仍然需要很长时间才能执行（即使在 30 分钟后也会继续）。
请提出建议。
之前的想法是使用 XML 表来创建这个 SQL，但成本更高 -
SELECT INCIDENTID,
COLUMN_VALUE AS CASUALFACTORS
FROM (
SELECT INCIDENTID,
CASUALFACTORS AS STR
FROM PSA.HSERS_INCIDENT_PSA
WHERE ACTIVE_FLAG='1'
AND ISDELETED = 'F'
) T,
XMLTABLE ( ('"' || REPLACE (str, ',', '","') || '"'))
Sample data -
select incidentid, dbms_lob.substr(CASUALFACTORS,3000)
from PSA.HSERS_INCIDENT_PSA WHERE ACTIVE_FLAG='1'
AND ISDELETED = 'F' and incidentid = 526849
526849 8,7,26
There are no duplicates for incidentid column in the table
DDL for the table -
ALTER TABLE PSA.HSERS_INCIDENT_PSA
DROP PRIMARY KEY CASCADE;
DROP TABLE PSA.HSERS_INCIDENT_PSA CASCADE CONSTRAINTS;
CREATE TABLE PSA.HSERS_INCIDENT_PSA
(
INCIDENTID INTEGER,
INCIDENTTYPE INTEGER,
SUPPLEMENTALINCIDENTTYPE INTEGER,
PENDINGREVIEW VARCHAR2(1 BYTE),
NEARMISSTYPE INTEGER,
NEARMISSSUBCONTRACTOR INTEGER,
RELEASEDATE TIMESTAMP(6),
NEARMISSIDENTIFICATION INTEGER,
NEARMISSCATEGORY INTEGER,
PROFITCENTER VARCHAR2(10 BYTE),
INCIDENTNUMBER NUMBER(19),
INCIDENTCODE VARCHAR2(15 BYTE),
HIRNUMBER VARCHAR2(8 BYTE),
REDBORDERALERTSENT VARCHAR2(1 BYTE),
TASKORDER VARCHAR2(3 BYTE),
LOGCAPAREA INTEGER,
TEAMCONNECTNUMBER VARCHAR2(20 BYTE),
INCIDENTDATE TIMESTAMP(6),
INCIDENTTIME VARCHAR2(10 BYTE),
REPORTDATE TIMESTAMP(6),
REPORTTIME VARCHAR2(5 BYTE),
INSERTDATETIME TIMESTAMP(6),
REPORTEDBY VARCHAR2(40 BYTE),
SUPERVISOR1 VARCHAR2(40 BYTE),
SUPERVISOR2 VARCHAR2(40 BYTE),
CLIENT INTEGER,
PROJECTLOCATION INTEGER,
INCIDENTAREA INTEGER,
INCIDENTLOCATION INTEGER,
INCIDENTAREADESCRIPTION VARCHAR2(80 BYTE),
DRUGALCOHOLTEST INTEGER,
NODRUGTESTRESPONSE INTEGER,
NODRUGTESTCOMMENTS CLOB,
FACTS CLOB,
POTENTIALCONSEQUENCES INTEGER,
LIKELIHOODRATING INTEGER,
RISKASSESSMENTSEVERITY INTEGER,
COVEREDBYTSTI_JSA VARCHAR2(1 BYTE),
INJURINGEVENTDISCUSSEDTSTI_JSA VARCHAR2(1 BYTE),
EXTERNALASSESSMENTFINEASSESED NUMBER(15,2),
REGULATORYAGENCY INTEGER,
INSPECTIONNUMBER VARCHAR2(12 BYTE),
INSPECTIONRESULTS INTEGER,
INSPECTIONCLOSEDATE TIMESTAMP(6),
EXTERNALASSESSMENTSEVERITY INTEGER,
CORRECTIVEMEASURES CLOB,
WITNESSDETAILS CLOB,
EXTERNALINVESTIGATORCOMMENTS CLOB,
CREATEDBY VARCHAR2(50 BYTE),
CREATEDON TIMESTAMP(6),
UPDATEDBY VARCHAR2(50 BYTE),
UPDATEDON TIMESTAMP(6),
ISDELETED VARCHAR2(1 BYTE),
INCIDENTSTATUS VARCHAR2(50 BYTE),
PARENTINCIDENTID NUMBER(19),
KEYSTOLIFE CLOB,
OFFICEBEHAVIORS CLOB,
CASUALFACTORS CLOB,
STANDARDSVIOLATED CLOB,
SUBCONTRACTORADDRESS CLOB,
CONFIDENTIALFACTS CLOB,
LASTVISITEDTAB VARCHAR2(50 BYTE),
LAST_RQST_ID VARCHAR2(60 BYTE),
DATA_SYS_SK VARCHAR2(3 BYTE),
CREATED_BY VARCHAR2(50 BYTE),
CREATED_DDTM DATE,
LAST_MODIFIED_BY VARCHAR2(50 BYTE),
LAST_MODIFIED_DDTM DATE,
ACTIVE_FLAG VARCHAR2(1 BYTE)
)
LOB (NODRUGTESTCOMMENTS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (FACTS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (CORRECTIVEMEASURES) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (WITNESSDETAILS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (EXTERNALINVESTIGATORCOMMENTS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (KEYSTOLIFE) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (OFFICEBEHAVIORS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (CASUALFACTORS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (STANDARDSVIOLATED) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (SUBCONTRACTORADDRESS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
LOB (CONFIDENTIALFACTS) STORE AS (
TABLESPACE PSA_DATA
ENABLE STORAGE IN ROW
CHUNK 16384
RETENTION
NOCACHE
LOGGING
STORAGE (
INITIAL 80K
NEXT 1M
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
))
TABLESPACE PSA_DATA
RESULT_CACHE (MODE DEFAULT)
PCTUSED 0
PCTFREE 10
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 80K
NEXT 1M
MAXSIZE UNLIMITED
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
)
LOGGING
NOCOMPRESS
NOCACHE
NOPARALLEL
MONITORING;
-- There is no statement for index PSA.SYS_C002335882.
-- The object is created when the parent object is created.
CREATE OR REPLACE SYNONYM BI_REPORTS_USER.HSERS_INCIDENT_PSA FOR PSA.HSERS_INCIDENT_PSA;
ALTER TABLE PSA.HSERS_INCIDENT_PSA ADD (
PRIMARY KEY
(INCIDENTID, LAST_RQST_ID)
USING INDEX
TABLESPACE PSA_DATA
PCTFREE 10
INITRANS 2
MAXTRANS 255
STORAGE (
INITIAL 80K
NEXT 1M
MAXSIZE UNLIMITED
MINEXTENTS 1
MAXEXTENTS UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
)
ENABLE VALIDATE);
【问题讨论】：
请edit您的问题添加一些示例数据。您给出的第一个查询无法在层次结构的不同级别中的行之间建立关联，因此如果有多个输入行，随着层次结构级别的增加，它将创建成倍增加的重复行；这就是为什么重要的是要知道您输入的内容，因为单行会“工作”，但多行可能会创建数百万个您不期望的重复行。 Oracle 性能调优很棘手，因为有很多因素在起作用。请阅读this answer in another SO thread，其中解释了我们回答您的问题所需的信息。它还可能为您提供足够的关于调整的线索，您可以自己解决。 【参考方案1】：您可以使用递归子查询因式分解子句和简单的字符串函数（而不是慢速正则表达式）：
Oracle 设置：
CREATE TABLE HSERS_INCIDENT_PSA ( incidentid, casualfactors, active_flag, isdeleted ) AS
SELECT 1, 'a,b,c,d,e,f', '1', 'F' FROM DUAL UNION ALL
SELECT 2, 'g,h,i,j,k', '1', 'F' FROM DUAL UNION ALL
SELECT 3, 'l', '1', 'F' FROM DUAL;
查询：
WITH casualfactors_bounds ( incidentid, casualfactors, startidx, endidx ) AS (
SELECT incidentid,
casualfactors,
1,
INSTR( casualfactors, ',', 1 )
FROM HSERS_INCIDENT_PSA
WHERE ACTIVE_FLAG = '1'
AND ISDELETED = 'F'
UNION ALL
SELECT incidentid,
casualfactors,
endidx + 1,
INSTR( casualfactors, ',', endidx + 1 )
FROM casualfactors_bounds
WHERE endidx > 0
)
SELECT incidentid,
CASE
WHEN endidx = 0
THEN SUBSTR( casualfactors, startidx )
ELSE SUBSTR( casualfactors, startidx, endidx - startidx )
END AS casualfactor
FROM casualfactors_bounds
ORDER BY incidentid, startidx
输出：
事件ID |休闲因素 ---------: | :----------- 1 |一种 1 | b 1 | C 1 | d 1 | e 1 | F 2 | G 2 | H 2 |一世 2 | j 2 | ķ 3 | l
说明计划：
|计划表输出 | | :------------------------------------------------ -------------------------------------------------- ------------- | |计划哈希值：2740663158 | | | | -------------------------------------------------- -------------------------------------------------- ------------- | | |身份证 |操作 |姓名 |行 |字节 |成本 (%CPU)|时间 | | | -------------------------------------------------- -------------------------------------------------- ------------- | | | 0 |选择声明 | | 6 | 276 | 7 (15)| 00:00:01 | | | | 1 |排序方式 | | 6 | 276 | 7 (15)| 00:00:01 | | | | 2 |查看 | | 6 | 276 | 6 (0)| 00:00:01 | | | | 3 |联合所有（递归）广度优先| | | | | | | | |* 4 |表访问完全 | HSERS_INCIDENT_PSA | 3 | 78 | 3 (0)| 00:00:01 | | | |* 5 |带泵的递归 | | | | | | | | -------------------------------------------------- -------------------------------------------------- ------------- | | | |谓词信息（由操作 id 标识）：| | -------------------------------------------------- - | | | | 4 - 过滤器（“ACTIVE_FLAG”='1' AND “ISDELETED”='F'）| | 5 - 过滤器（“ENDIDX”>0）| | | |注意 | | ----- | | - 用于此语句的动态采样 (level=2) |
db小提琴here
【讨论】：
非常感谢您的帮助！查询在 10 秒内返回行！【参考方案2】：FROM PSA.HSERS_INCIDENT_PSA WHERE ACTIVE_FLAG='1' AND ISDELETED = 'F' 是否返回不止一行？如果是这样，CONNECT BY 子句将生成所有行的笛卡尔积。这会打击你的基数并导致极端的运行时间。
你可以通过一个简单的技巧来避免这种情况：
select INCIDENTID, trim(regexp_substr(hip.CASUALFACTORS,'[^,]+', 1, level) ) value
FROM PSA.HSERS_INCIDENT_PSA hip
WHERE hip.ACTIVE_FLAG='1'
AND hip.ISDELETED = 'F'
connect by regexp_substr(CASUALFACTORS, '[^,]+', 1, level) is not null
and hip.rowid = prior hip.rowid
and prior sys_guid() is not null
order by hip.INCIDENTID
【讨论】：
谢谢！查询仍然需要时间来获取结果（超过 15 分钟并且仍在执行），我想我做错了什么。【参考方案3】：如果数据库正在生成错误的计划，可能是因为该表上的统计信息已过时。您可以执行SELECT LAST_ANALYZED FROM ALL_TABLES WHERE OWNER = 'PSA' AND TABLE_NAME = 'HSERS_INCIDENT_PSA' 来查看上次在此表上收集统计信息的时间，但您也可以继续使用
BEGIN
DBMS_STATS.GATHER_TABLE_STATS(OWNNAME => 'PSA',
TABNAME => 'HSERS_INCIDENT_PSA',
CASCADE => TRUE);
END;
然后重新评估计划。
【讨论】：
以上是关于尽管成本低、基数低，但 Oracle 中的选择查询需要很长时间的主要内容，如果未能解决你的问题，请参考以下文章