我们将使用以下SQL删除表中的大量行,如此处所述:
DELETE FROM MYTABLE
WHERE UPDT_TIMESTMP < v_Cut_Off_Date
AND ROWNUM <= C_MAX_DELETE;
我注意到UPDT_TIMESTMP可以为 NULL。此字段存储初始创建后上次更新记录时的 TIMESTAMP 值。因此,如果更新时间为 NULL,我希望修改我的 SQL 以考虑创建时间。
DELETE FROM MYTABLE
WHERE NVL(UPDT_TIMESTMP, CRET_TIMESTMP) < v_Cut_Off_Date
AND ROWNUM <= C_MAX_DELETE;
我的偏好是不允许 NULL 并将UPDT_TIMESTMP列的值更新为CRET_TIMESTMP值,但这不是一个选项。
由于该表很大,每月大约 20M 条记录,每个月我都会删除一个月的旧数据,因此我想确保我可以快速找到要删除的记录。
使用此原始 SQL,
DELETE FROM COMMRCL_CORE_CLM_DTL
WHERE UPDT_TIMESTMP < SYSDATE AND ROWNUM <= C_MAX_DELETE;
。没有索引,这是使用 Toad for Oracle 的查询计划:
Plan
DELETE STATEMENT ALL_ROWSCost: 2 Bytes: 41 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 TABLE ACCESS FULL TABLE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL Cost: 2 Bytes: 41 Cardinality: 1
添加了上次更新时间的索引:
CREATE INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_UPDTM ON FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
(UPDT_TIMESTMP)
LOGGING
TABLESPACE USERS
PCTFREE 10
INITRANS 2
MAXTRANS 255
STORAGE (
MAXSIZE UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
)
NOPARALLEL;
在上次更新时间添加索引后的查询计划(使用索引(
Plan
DELETE STATEMENT ALL_ROWSCost: 0 Bytes: 41 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 INDEX RANGE SCAN INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_UPDTM Cost: 0 Bytes: 41 Cardinality: 1
修改了查询以在更新时间为 NULL 时使用创建日期
DELETE FROM COMMRCL_CORE_CLM_DTL
WHERE NVL(UPDT_TIMESTMP, CRET_TIMESTMP) < SYSDATE AND ROWNUM <= C_MAX_DELETE;
在创建时间上添加了单独的索引
CREATE INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_CRET ON
FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
(CRET_TIMESTMP)
LOGGING
TABLESPACE USERS
PCTFREE 10
INITRANS 2
MAXTRANS 255
STORAGE (
MAXSIZE UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
)
NOPARALLEL;
添加了 2 个单独索引后检查了查询计划。
DELETE STATEMENT ALL_ROWSCost: 2 Bytes: 54 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 TABLE ACCESS FULL TABLE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL Cost: 2 Bytes: 54 Cardinality: 1
问题:为什么两个索引都没有使用?
添加了在同一索引中同时包含"上次更新"和"创建时间"列的新索引
CREATE INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_UPDCRT ON FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
(UPDT_TIMESTMP, CRET_TIMESTMP)
LOGGING
TABLESPACE USERS
PCTFREE 10
INITRANS 2
MAXTRANS 255
STORAGE (
MAXSIZE UNLIMITED
PCTINCREASE 0
BUFFER_POOL DEFAULT
FLASH_CACHE DEFAULT
CELL_FLASH_CACHE DEFAULT
)
NOPARALLEL;
尽管如此,没有使用索引。为什么?
Plan
DELETE STATEMENT ALL_ROWSCost: 2 Bytes: 54 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 TABLE ACCESS FULL TABLE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL Cost: 2 Bytes: 54 Cardinality: 1
我意识到表中没有太多数据会影响解释计划(我很少。我必须生成数百万行才能真正了解预期内容,还是可以在不这样做的情况下获得大致想法?
为什么上面的示例中没有使用索引,或者我误读了计划?
更新:
当我采用 Mat 的建议将 DELETE 分解为两个更新时,第一个按创建日期:
DELETE FROM COMMRCL_CORE_CLM_DTL
WHERE UPDT_TIMESTMP < SYSDATE AND ROWNUM <= variable;
。更新日期上的索引用于第一个
Plan
DELETE STATEMENT ALL_ROWSCost: 0 Bytes: 54 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 INDEX RANGE SCAN INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_UPDCRT Cost: 0 Bytes: 54 Cardinality: 1
对于第二个 SQL...
DELETE FROM COMMRCL_CORE_CLM_DTL
WHERE UPDT_TIMESTMP IS NULL AND CRET_TIMESTMP < SYSDATE AND ROWNUM <= Variable;
使用包含两列的索引:
Plan
DELETE STATEMENT ALL_ROWSCost: 0 Bytes: 54 Cardinality: 1
3 DELETE FIN_IT_RPT.COMMRCL_CORE_CLM_DTL
2 COUNT STOPKEY
1 INDEX RANGE SCAN INDEX FIN_IT_RPT.COMMRCL_CORE_CLM_DTL_UPDCRT Cost: 0 Bytes: 54 Cardinality: 1
在第两种情况下只使用单独的 DELETE 语句而不使用 NVL:
DELETE FROM MYTABLE
WHERE UPDT_TIMESTMP IS NULL AND CRET_TIMESTMP < v_Cut_Off_Date
AND ROWNUM <= C_MAX_DELETE;
您可以将这两个语句合并为一个语句,... WHERE UPDT_TIMESTMP < v_Cut_Off_Date OR (UPDT_TIMESTMP IS NULL AND CRET_TIMESTMP < v_Cut_Off_Date) ...
如果您只有几条记录带有 UPDT_TIMESTMP IS NULL
,则创建一个基于函数的MY_NVL(UPDT_TIMESTMP,CRET_TIMESTMP)
索引,其中MY_NVL
返回UPDT_TIMESTMP的函数为 NULL 的 CRET_TIMESTMP,而 NULL 的 NULL(UPDT_TIMESTMP 的 NULL 不为 NULL,则 where 条件如下所示... WHERE UPDT_TIMESTMP < v_Cut_Off_Date OR MY_NVL(UPDT_TIMESTMP,CRET_TIMESTMP) < v_Cut_Off_Date ...
您也可以尝试使用 NVL(UPDT_TIMESTMP, CRET_TIMESTMP)
使用基于函数的索引(正如大卫最初提议的那样 - 对不起大卫,我还没有阅读您的评论(