隐蔽的笛卡尔集

今天接到一个工单，是一个抽取数据工单，语句如下，初看语法没有问题，并且两张表的数据量差不多都是100多万，

就在数据库上跑了起来，半天还没运行完，当时就查看了这个会话，显示该会话的等待事件 gr cr request。 

select  count(*)
from t_cmp_type_bookitem bk,t_cmp_type_content ctc
where
bk.objectid=ctc.objectid
and ctc.status = 13
and bk.itemtype = '1' or bk.itemtype = '5';

当时也觉得很正常，毕竟这是一个4 节点的race， 当时想，看下有关这两张表的sql 主要在哪个节点，然后

就把该sql 放在哪个节点上跑，听取了意见，一查发现4个节点上都有，这下犯难了

select inst_id,SQL_TEXT  from gv$sql where executions > 1
and PARSING_SCHEMA_NAME='MREAD'
--AND LAST_LOAD_TIME >SYSDATE - 1
AND LAST_ACTIVE_TIME > SYSDATE -1
AND ( SQL_TEXT LIKE '%t_cmp_type_bookitem%' OR SQL_TEXT LIKE '%t_cmp_type_content%')

想想，使用CTAS 方法创建 两张临时表，然后用临时表代替原表，这样就不会出现 gr cr request 等待事件了

如是就按这种方法做了，可以登了很久还没跑完，看了一下执行计划，执行计划评估的结果有 179G，不至于

这么大啊，当时请教了开发DBA。

SQL> select * from table(dbms_xplan.display_cursor('8rkakfj35d2jr', null, 'advanced'));
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
SQL_ID  8rkakfj35d2jr, child number 0
-------------------------------------
select count(*)            from t_cmp_type_bookitemhlb
bk,t_cmp_type_contenthlb ctc          where
bk.objectid=ctc.objectid           and ctc.status = :"SYS_B_0"
and bk.itemtype = :"SYS_B_1" or bk.itemtype = :"SYS_B_2"
Plan hash value: 2230377177
--------------------------------------------------------------------------------
| Id  | Operation                | Name                         | Rows  | Bytes
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT         |                              |       |
|   1 |  SORT AGGREGATE          |                              |     1 |    21
|   2 |   CONCATENATION          |                              |       |
|   3 |    MERGE JOIN CARTESIAN  |                              |    10G|   197G
|*  4 |     TABLE ACCESS FULL    | T_CMP_TYPE_BOOKITEMHLB       | 10096 |   108K
|   5 |     BUFFER SORT          |                              |   998K|  9753K
|   6 |      INDEX FAST FULL SCAN| IDX_T_CMP_TYPE_CONTENT_OSHLB |   998K|  9753K
|*  7 |    HASH JOIN             |                              |  9995 |   204K
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
|*  8 |     TABLE ACCESS FULL    | T_CMP_TYPE_BOOKITEMHLB       |  9995 |   107K
|*  9 |     INDEX FAST FULL SCAN | IDX_T_CMP_TYPE_CONTENT_OSHLB | 90793 |   886K
--------------------------------------------------------------------------------

开发DBA看了会，果断在最后一个and 条件后加了括号，问题立马解决。看下执行计划，这次执行

计划评估的结果不到1 M ，仔细看是消除了 笛卡尔集。

select  count(*)
from t_cmp_type_bookitemhlb bk,t_cmp_type_contenthlb ctc
where
bk.objectid=ctc.objectid
and ctc.status = 13
and（ bk.itemtype = '1' or bk.itemtype = '5'）;

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 601123799
--------------------------------------------------------------------------------
| Id  | Operation              | Name                         | Rows  | Bytes |
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT       |                              |     1 |    21 |
|   1 |  SORT AGGREGATE        |                              |     1 |    21 |
|*  2 |   HASH JOIN            |                              | 10096 |   207K|
|*  3 |    TABLE ACCESS FULL   | T_CMP_TYPE_BOOKITEMHLB       | 10096 |   108K|
|*  4 |    INDEX FAST FULL SCAN| IDX_T_CMP_TYPE_CONTENT_OSHLB |   440K|  4302K|
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("BK"."OBJECTID"="CTC"."OBJECTID")
3 - filter("BK"."ITEMTYPE"='1' OR "BK"."ITEMTYPE"='5')
4 - filter("CTC"."STATUS"=13)