又是latch: cache buffers chains惹得祸
前言
一大早，客户给我打电话说:
xx，应用很慢，查询数据总是超时，让我看看。。。
根据多年DBA经验，首当其冲的肯定是去查询数据库在这段时间都在干嘛。
分析
导出awr报告分析
1）. 数据库在此时间段非常繁忙。
2）. 查看Top 5 Timed Events，出现了Concurrency等待事件latch: library cache**
3）. 查看SQL ordered by Gets,不看不知道，一看吓一跳
4）. *50pwxa3bzp7gkSQL语句
select *
from (select d.*, rownum as num
from (SELECT A.BILLNO,
A.BILLCODE,
A.GETDATE,
A.GETUNITCODE,
A.GETCODE,
A.GETORGANCODE,
A.USEORGANCODE,
A.USEDATE,
A.USEUNITCODE,
A.USERCODE,
A.CURRENCYCODE,
A.AMOUNT,
A.NAME,
A.NOTES,
A.STATUSCODE,
A.IFPAGEONHOLE,
A.OPCODE,
A.OPUNITCODE,
A.OPDATE,
A.LOCKTIME,
A.GETAGENTCODE,
(SELECT D.AGENTNAME
FROM SYN_MM_AGENTCODE_TC D
WHERE D.AGENTCODE = A.GETAGENTCODE) AS GETAGENTNAME,
A.USEAGENTCODE,
A.OUTSTATUS,
CASE A.BILLCODE
WHEN 'B2010005' THEN
A.FACTBILLCODE
ELSE
''
END FACTBILLCODE,
A.SALES,
A.FROMDATE,
A.TODATE,
(SELECT BILLNAME
FROM BD_BILLCODE
WHERE BILLCODE = A.BILLCODE) BILLNAME,
(SELECT HANDLERNAME
FROM BD_HANDLER
WHERE HANDLERCODE = A.USERCODE) USERNAME,
(SELECT HANDLERNAME
FROM BD_HANDLER
WHERE HANDLERCODE = A.GETCODE) GETERNAME,
(SELECT NO3
FROM B_BILLDETAIL
WHERE BILLNO = A.BILLNO
AND BILLCODE = A.BILLCODE
AND FACTBILLCODE = A.FACTBILLCODE) ONLINEINVOICENO,
(SELECT NO4
FROM B_BILLDETAIL
WHERE BILLNO = A.BILLNO
AND BILLCODE = A.BILLCODE
AND FACTBILLCODE = A.FACTBILLCODE) ONLINEINVOICECODE
FROM B_BILL A
WHERE 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND TRIM(BILLNO) >= :B1
AND TRIM(BILLNO) <= :B2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
ORDER BY A.BILLNO) d
where rownum <= 1)
where num > 0
执行计划：
Plan hash value: 4085294641
------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 4632 | 9 (0)| 00:00:01 |
|* 1 | TABLE ACCESS FULL | CHAGENTBASE | 1 | 56 | 6 (0)| 00:00:01 |
| 2 | TABLE ACCESS BY INDEX ROWID | BD_BILLCODE | 1 | 31 | 1 (0)| 00:00:01 |
|* 3 | INDEX UNIQUE SCAN | PK_BD_BILLCODE | 1 | | 0 (0)| 00:00:01 |
| 4 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 5 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
| 6 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 7 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
| 8 | TABLE ACCESS BY INDEX ROWID | B_BILLDETAIL | 1 | 50 | 4 (0)| 00:00:01 |
|* 9 | INDEX RANGE SCAN | PK_B_BILLDETAIL_02 | 1 | | 3 (0)| 00:00:01 |
| 10 | TABLE ACCESS BY INDEX ROWID | B_BILLDETAIL | 1 | 50 | 4 (0)| 00:00:01 |
|* 11 | INDEX RANGE SCAN | PK_B_BILLDETAIL_02 | 1 | | 3 (0)| 00:00:01 |
|* 12 | VIEW | | 1 | 4632 | 9 (0)| 00:00:01 |
|* 13 | COUNT STOPKEY | | | | | |
| 14 | VIEW | | 2 | 9238 | 9 (0)| 00:00:01 |
| 15 | TABLE ACCESS BY INDEX ROWID| B_BILL | 17395 | 3822K| 9 (0)| 00:00:01 |
|* 16 | INDEX FULL SCAN | PK_B_BILL_01 | 2 | | 8 (0)| 00:00:01 |
------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("D"."AGENTCODE"=:B1)
3 - access("BILLCODE"=:B1)
5 - access("A"."CODE"=:B1)
7 - access("A"."CODE"=:B1)
9 - access("BILLNO"=:B1 AND "BILLCODE"=:B2 AND "FACTBILLCODE"=:B3)
11 - access("BILLNO"=:B1 AND "BILLCODE"=:B2 AND "FACTBILLCODE"=:B3)
12 - filter("NUM">0)
13 - filter(ROWNUM<=1)
16 - filter(TO_NUMBER(TRIM("BILLNO"))>=601710100010 AND
TO_NUMBER(TRIM("BILLNO"))<=601710100010)5）. 对sql语句进行分析
1、SQL语句中有很多标量子查询，我们可以利用left join 对其改写。
2、id = 1 为TABLE ACCESS FULL，表示CHAGENTBASE走的是全表扫描，在标量子查询中，主表返回多少
行，子表也跟着被扫描多少次，所以需要对CHAGENTBASE建索引。
3、SQL语句中出现TRIM(BILLNO) >= :B1 AND TRIM(BILLNO) <= :B2，导致ID = 16 为
INDEX FULL SCAN。对主键进行索引全扫描，这种访问方式是最垃圾的。
优化
1）创建索引
create indexIDX_CHAGENTBASE_TEST on CHAGENTBASE (AGENTCODE);
2）标量改成left join
select *
from (select G.*, rownum as num
from (SELECT A.BILLNO,
A.BILLCODE,
A.GETDATE,
A.GETUNITCODE,
A.GETCODE,
A.GETORGANCODE,
A.USEORGANCODE,
A.USEDATE,
A.USEUNITCODE,
A.USERCODE,
A.CURRENCYCODE,
A.AMOUNT,
A.NAME,
A.NOTES,
A.STATUSCODE,
A.IFPAGEONHOLE,
A.OPCODE,
A.OPUNITCODE,
A.OPDATE,
A.LOCKTIME,
A.GETAGENTCODE,
/* (SELECT D.AGENTNAME
FROM SYN_MM_AGENTCODE_TC D
WHERE D.AGENTCODE = A.GETAGENTCODE) AS GETAGENTNAME*/
D.AGENTNAME AS GETAGENTNAME,
A.USEAGENTCODE,
A.OUTSTATUS,
CASE A.BILLCODE
WHEN 'B2010005' THEN
A.FACTBILLCODE
ELSE
''
END FACTBILLCODE,
A.SALES,
A.FROMDATE,
A.TODATE,
/* (SELECT BILLNAME
FROM BD_BILLCODE
WHERE BILLCODE = A.BILLCODE) BILLNAME,*/
B.BILLNAME,
/* (SELECT HANDLERNAME
FROM BD_HANDLER
WHERE HANDLERCODE = A.USERCODE) USERNAME,*/
C.HANDLERNAME USERNAME,
/* (SELECT HANDLERNAME
FROM BD_HANDLER
WHERE HANDLERCODE = A.GETCODE) GETERNAME,*/
E.HANDLERNAME GETERNAME,
F.no3 ONLINEINVOICENO,
F.no4 ONLINEINVOICECODE
FROM B_BILL A
LEFT JOIN SYN_MM_AGENTCODE_TC D
ON D.AGENTCODE = A.GETAGENTCODE
LEFT JOIN BD_BILLCODE B
ON B.BILLCODE = A.BILLCODE
LEFT JOIN BD_HANDLER C
ON C.HANDLERCODE = A.USERCODE
LEFT JOIN BD_HANDLER E
ON E.HANDLERCODE = A.GETCODE
LEFT JOIN B_BILLDETAIL F
ON F.BILLNO = A.BILLNO
AND F.BILLCODE = A.BILLCODE
AND F.FACTBILLCODE = A.FACTBILLCODE
WHERE 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND trim(A.BILLNO) >= '601710100010'
AND trim(A.BILLNO) <= '601710100010'
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
AND 3 > 2
ORDER BY A.BILLNO)G
where rownum <= 1)
where num > 0;
Plan hash value: 1528527901
------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 4632 | 21 (0)| 00:00:01 |
|* 1 | VIEW | | 1 | 4632 | 21 (0)| 00:00:01 |
|* 2 | COUNT STOPKEY | | | | | |
| 3 | VIEW | | 2 | 9238 | 21 (0)| 00:00:01 |
| 4 | NESTED LOOPS OUTER | | 2 | 832 | 21 (0)| 00:00:01 |
| 5 | NESTED LOOPS OUTER | | 2 | 770 | 19 (0)| 00:00:01 |
| 6 | NESTED LOOPS OUTER | | 2 | 718 | 17 (0)| 00:00:01 |
| 7 | NESTED LOOPS OUTER | | 2 | 614 | 12 (0)| 00:00:01 |
| 8 | NESTED LOOPS OUTER | | 2 | 562 | 10 (0)| 00:00:01 |
| 9 | TABLE ACCESS BY INDEX ROWID| B_BILL | 17395 | 3822K| 9 (0)| 00:00:01 |
|* 10 | INDEX FULL SCAN | PK_B_BILL_01 | 2 | | 8 (0)| 00:00:01 |
| 11 | TABLE ACCESS BY INDEX ROWID| CHAGENTBASE | 1 | 56 | 1 (0)| 00:00:01 |
|* 12 | INDEX RANGE SCAN | IDX_CHAGENTBASE_TEST | 1 | | 0 (0)| 00:00:01 |
| 13 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 14 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
| 15 | TABLE ACCESS BY INDEX ROWID | B_BILLDETAIL | 1 | 52 | 3 (0)| 00:00:01 |
|* 16 | INDEX RANGE SCAN | PK_B_BILLDETAIL_02 | 1 | | 2 (0)| 00:00:01 |
| 17 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 18 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
| 19 | TABLE ACCESS BY INDEX ROWID | BD_BILLCODE | 1 | 31 | 1 (0)| 00:00:01 |
|* 20 | INDEX UNIQUE SCAN | PK_BD_BILLCODE | 1 | | 0 (0)| 00:00:01 |
------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("NUM">0)
2 - filter(ROWNUM<=1)
10 - filter(TRIM("A"."BILLNO")>='601710100010' AND TRIM("A"."BILLNO")<='601710100010')
12 - access("D"."AGENTCODE"(+)="A"."GETAGENTCODE")
14 - access("A"."CODE"(+)="A"."GETCODE")
16 - access("F"."BILLNO"(+)="A"."BILLNO" AND "F"."BILLCODE"(+)="A"."BILLCODE" AND
"F"."FACTBILLCODE"(+)="A"."FACTBILLCODE")
18 - access("A"."CODE"(+)="A"."USERCODE")
20 - access("B"."BILLCODE"(+)="A"."BILLCODE")
3） 把Trim去掉
Execution Plan
----------------------------------------------------------
Plan hash value: 1229065410
------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 4632 | 13 (0)| 00:00:01 |
|* 1 | VIEW | | 1 | 4632 | 13 (0)| 00:00:01 |
|* 2 | COUNT STOPKEY | | | | | |
| 3 | VIEW | | 1 | 4619 | 13 (0)| 00:00:01 |
| 4 | NESTED LOOPS OUTER | | 1 | 416 | 13 (0)| 00:00:01 |
| 5 | NESTED LOOPS OUTER | | 1 | 390 | 11 (0)| 00:00:01 |
| 6 | NESTED LOOPS OUTER | | 1 | 364 | 9 (0)| 00:00:01 |
| 7 | NESTED LOOPS OUTER | | 1 | 308 | 8 (0)| 00:00:01 |
| 8 | NESTED LOOPS OUTER | | 1 | 277 | 7 (0)| 00:00:01 |
| 9 | TABLE ACCESS BY INDEX ROWID| B_BILL | 1 | 225 | 4 (0)| 00:00:01 |
|* 10 | INDEX RANGE SCAN | PK_B_BILL | 1 | | 3 (0)| 00:00:01 |
| 11 | TABLE ACCESS BY INDEX ROWID| B_BILLDETAIL | 1 | 52 | 3 (0)| 00:00:01 |
|* 12 | INDEX RANGE SCAN | PK_B_BILLDETAIL_02 | 1 | | 2 (0)| 00:00:01 |
| 13 | TABLE ACCESS BY INDEX ROWID | BD_BILLCODE | 1 | 31 | 1 (0)| 00:00:01 |
|* 14 | INDEX UNIQUE SCAN | PK_BD_BILLCODE | 1 | | 0 (0)| 00:00:01 |
| 15 | TABLE ACCESS BY INDEX ROWID | CHAGENTBASE | 1 | 56 | 1 (0)| 00:00:01 |
|* 16 | INDEX RANGE SCAN | IDX_CHAGENTBASE_TEST | 1 | | 0 (0)| 00:00:01 |
| 17 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 18 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
| 19 | TABLE ACCESS BY INDEX ROWID | T_EMPLOYEE_VIEW | 1 | 26 | 2 (0)| 00:00:01 |
|* 20 | INDEX RANGE SCAN | IND_T_EMPLOYEE_VIEW | 1 | | 1 (0)| 00:00:01 |
------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("NUM">0)
2 - filter(ROWNUM<=1)
10 - access("A"."BILLNO"='601710100010')
12 - access("F"."BILLNO"(+)='601710100010' AND "F"."BILLCODE"(+)="A"."BILLCODE" AND
"F"."FACTBILLCODE"(+)="A"."FACTBILLCODE")
filter("F"."BILLNO"(+)="A"."BILLNO")
14 - access("B"."BILLCODE"(+)="A"."BILLCODE")
16 - access("D"."AGENTCODE"(+)="A"."GETAGENTCODE")
18 - access("A"."CODE"(+)="A"."GETCODE")
20 - access("A"."CODE"(+)="A"."USERCODE")
Statistics
----------------------------------------------------------
621 recursive calls
0 db block gets
229 consistent gets
17 physical reads
0 redo size
2937 bytes sent via SQL*Net to client
2086 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
16 sorts (memory)
0 sorts (disk)
1 rows processed优化效果
可以看出消耗的 buffer cache 从之前的
882,856,212.00降到了229，效率提升了N倍.
