ORA-01652，temp表空间不足的相关问题及处理

来源：互联网发布：js 异步方法同步执行编辑：程序博客网时间：2024/06/08 09:33

十一长假期间也不得轻松，某日接到业务保障，数据库报错，导致某关键业务不能正常执行，需要立即处理

原因分析

1，登录数据库，查看主机日志，报错内容为ORA-01652，temp表空间不足
ORA-01652: unable to extend temp segment by 128 in tablespace TEMP01

2，让业务部门重新执行相关脚本，发现占用temp表空间的具体语句如下，目前temp表空间96GB，大约1个小时会被该sql使用满，sql异常退出

Sql具体如下INSERT INTO www.WWW_BILL_DTL_TEMP_0101(ACCT_ID,SERV_ID,FEE,BRAND,PHONE_ID,USER_TYPE)  SELECT ACCT_ID, SERV_ID, sum(FEE) FEE,BRAND,PHONE_ID,USER_TYPE  FROM (  SELECT ACCT_ID, SERV_ID, SUM(FEE) FEE,BRAND,PHONE_ID,USER_TYPE  FROM  (select a.acct_id,e.serv_id,sum(b.unpay_fee) FEE,a.brand,a.phone_id,a.user_type from www.ACC_BILL_010120121010 A , www.WWW_BILL_DTL_010120121010 B ,  www.OWE_MONITOR_QUEUE_ACTION E  where a.bill_id=b.bill_id and A.ACCT_ID=E.ACCT_ID and a.brand in (:"SYS_B_00",:"SYS_B_01",:"SYS_B_02",:"SYS_B_03",:"SYS_B_04",:"SYS_B_05")  and b.fee_item_id>:"SYS_B_06"group by a.acct_id,e.serv_id,a.brand,a.phone_id,a.user_type )  group by ACCT_ID, SERV_ID,BRAND,PHONE_ID,USER_TYPE  UNION ALL  SELECT ACCT_ID, SERV_ID, -:"SYS_B_07"*SUM(FEE) FEE,BRAND,PHONE_ID,USER_TYPE  FROM  (select a.acct_id,e.serv_id,sum(b.unpay_fee) FEE,a.brand,a.phone_id,a.user_type from www.WWW_BILL_010120121005 A , www.WWW_BILL_DTL_010120121005 B ,  www.WWW_MONITOR_QUEUE_ACTION E  where a.bill_id=b.bill_id and A.ACCT_ID=E.ACCT_ID and a.brand in (:"SYS_B_08",:"SYS_B_09",:"SYS_B_10",:"SYS_B_11",:"SYS_B_12",:"SYS_B_13")  and b.fee_item_id>:"SYS_B_14" group by a.acct_id,e.serv_id,a.brand,a.phone_id,a.user_type )  group by ACCT_ID, SERV_ID,BRAND,PHONE_ID,USER_TYPE ) GROUP BY ACCT_ID, SERV_ID,BRAND,PHONE_ID,USER_TYPE

执行计划如下

Plan hash value: 3236377944--------------------------------------------------------------------------------------------------------------------| Id  | Operation                            | Name                        | Rows  | Bytes | Cost (%CPU)| Time     |--------------------------------------------------------------------------------------------------------------------|   0 | INSERT STATEMENT                     |                             |       |       | 19281 (100)|          ||   1 |  HASH GROUP BY                       |                             |     2 |   184 | 19281   (2)| 00:03:52 ||   2 |   VIEW                               |                             |     2 |   184 | 19280   (2)| 00:03:52 ||   3 |    UNION-ALL                         |                             |       |       |            |          ||   4 |     SORT GROUP BY                    |                             |     1 |    92 | 19271   (2)| 00:03:52 ||   5 |      VIEW                            |                             |     1 |    92 | 19271   (2)| 00:03:52 ||   6 |       SORT GROUP BY                  |                             |     1 |   144 | 19271   (2)| 00:03:52 ||*  7 |        HASH JOIN                     |                             |     1 |   144 | 19270   (2)| 00:03:52 ||   8 |         MERGE JOIN CARTESIAN         |                             |     1 |    65 |  8717   (2)| 00:01:45 ||   9 |          TABLE ACCESS FULL           | WWW_MONITOR_QUEUE_ACTION    |     1 |    26 |     2   (0)| 00:00:01 ||  10 |          BUFFER SORT                 |                             |   257K|  9810K|  8715   (2)| 00:01:45 ||* 11 |           TABLE ACCESS FULL          | WWW_BILL_DTL_010120121010   |   257K|  9810K|  8715   (2)| 00:01:45 ||* 12 |         TABLE ACCESS FULL            | WWW_BILL_010120121010       | 16755 |  1292K| 10552   (1)| 00:02:07 ||  13 |     SORT GROUP BY                    |                             |     1 |    53 |     9  (12)| 00:00:01 ||  14 |      VIEW                            |                             |     1 |    53 |     9  (12)| 00:00:01 ||  15 |       SORT GROUP BY                  |                             |     1 |    79 |     9  (12)| 00:00:01 ||  16 |        TABLE ACCESS BY INDEX ROWID   | WWW_BILL_DTL_010120121005   |     1 |    18 |     3   (0)| 00:00:01 ||  17 |         NESTED LOOPS                 |                             |     1 |    79 |     8   (0)| 00:00:01 ||  18 |          NESTED LOOPS                |                             |     1 |    61 |     5   (0)| 00:00:01 ||  19 |           TABLE ACCESS FULL          | WWW_MONITOR_QUEUE_ACTION    |     1 |    26 |     2   (0)| 00:00:01 ||* 20 |           TABLE ACCESS BY INDEX ROWID| WWW_BILL_010120121005       |     1 |    35 |     3   (0)| 00:00:01 ||* 21 |            INDEX RANGE SCAN          | ITDX_ACCT_ID_10120121005    |     1 |       |     2   (0)| 00:00:01 ||* 22 |          INDEX RANGE SCAN            | TPK_BILL_DTL_ID_10120121005 |     1 |       |     2   (0)| 00:00:01 |--------------------------------------------------------------------------------------------------------------------

3，继续分析原因。仔细分析该sql语句，实际上可以拆分为两个并列子sql。后一部分的sql执行计划是好的，不优的是上半部分，因此解决方案就很简单了，把上半部分的统计信息和索引做成和下半部分同样的结构。经查确认，上部分的sql缺少相关索引，且由于表今天创建，且插入了大量数据，且没有统计信息，选择列没有设置索引，导致执行计划不优。

SQL>  select * from dba_objects where object_name ='WWW_BILL_DTL_010120121010';OWNER                          OBJECT_NAME                                                                                                                      SUBOBJECT_NAME------------------------------ -------------------------------------------------------------------------------------------------------------------------------- ------------------------------ OBJECT_ID DATA_OBJECT_ID OBJECT_TYPE         CREATED      LAST_DDL_TIM TIMESTAMP           STATUS  T G S---------- -------------- ------------------- ------------ ------------ ------------------- ------- - - -ZC                             WWW_BILL_DTL_010120121010   1209877        1209877 TABLE               06-OCT-12    06-OCT-12    2012-10-06:11:19:25 VALID   N N N

解决方法

1，表WWW_BILL_DTL选择列ACCT_ID，在表WWW_BILL上bill_id，FEE_ITEM_ID添加组合索引，

2，然后重新收集统计信息，sql即走合适的执行路径了

exec dbms_stats.gather_table_stats('ZC',WWW_BILL_DTL_010120121010,method_opt => 'FOR ALL COLUMNS SIZE 1',cascade =>TRUE,estimate_percent => 40,granularity=>'all',degree => 4,no_invalidate => false);exec dbms_stats.gather_table_stats('ZC',WWW_BILL_010120121010,method_opt => 'FOR ALL COLUMNS SIZE 1',cascade =>TRUE,estimate_percent => 40,granularity=>'all',degree => 4,no_invalidate => false)

简单总结：

1，本次问题出在新建的临时表，没有同时新建相关索引，且在数据创建和大量数据插入后，没有立即收集统计信息，导致oracle无法选择合适的执行计划，进而占用大量临时表空间，sql无法正常执行完成。
2，merge join cartesian（笛卡儿算法）：是每个集合的任务一个成员都要与其他集合的每个成员进行匹配。因此原有执行计划需要执行对两个大表的N×N次查询和排序，占用大量temp表空间