|
5.捕获相关SQL
这里用到了我的以下脚本getsqlbysid:
SELECT sql_text
FROM v$sqltext a
WHERE a.hash_value = (SELECT sql_hash_value
FROM v$session b
WHERE b.SID = ’&sid’)
ORDER BY piece ASC
/ |
该脚本根据用户sid,结合v$session和v$sqltext视图,获得用户sql语句的完整文本。用该脚本,通过从v$session_wait中获得的等待全表或索引扫描的进程SID，捕获问题sql:
SQL> @getsql
Enter value for sid: 18
old 5: where b.sid=’&sid’
new 5: where b.sid=’18’
SQL_TEXT
----------------------------------------------------------------
select i.vc2title,i.numinfoguid from hs_info i where i.intenab
ledflag = 1 and i.intpublishstate = 1 and i.datpublishdate <=
sysdate and i.numcatalogguid = 2047 order by i.datpublishdate d
esc, i.numorder desc
SQL> /
Enter value for sid: 54
old 5: where b.sid=’&sid’
new 5: where b.sid=’54’
SQL_TEXT
----------------------------------------------------------------
select i.vc2title,i.numinfoguid from hs_info i where i.intenab
ledflag = 1 and i.intpublishstate = 1 and i.datpublishdate <=
sysdate and i.numcatalogguid = 33 order by i.datpublishdate des
c, i.numorder desc
SQL> /
Enter value for sid: 49
old 5: where b.sid=’&sid’
new 5: where b.sid=’49’
SQL_TEXT
----------------------------------------------------------------
select i.vc2title,i.numinfoguid from hs_info i where i.intenab
ledflag = 1 and i.intpublishstate = 1 and i.datpublishdate <=
sysdate and i.numcatalogguid = 26 order by i.datpublishdate des
c, i.numorder desc |
对几个进程进行跟踪,分别得到以上SQL语句，这些SQL可能就是问题产生的根源。以上语句如果良好编码应该使用绑定变量.但是现在这个不是我们关心的。使用该应用用户连接，检查以上SQL的执行计划:
SQL> set autotrace trace explain
SQL> select i.vc2title,i.numinfoguid
2 from hs_info i where i.intenabledflag = 1
3 and i.intpublishstate = 1 and i.datpublishdate <=sysdate
4 and i.numcatalogguid = 3475
5 order by i.datpublishdate desc, i.numorder desc ;
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=228 Card=1 Bytes=106)
1 0 SORT (ORDER BY) (Cost=228 Card=1 Bytes=106)
2 1 TABLE ACCESS (FULL) OF ’HS_INFO’ (Cost=218 Card=1 Bytes=106)
SQL> select count(*) from hs_info;
COUNT(*)
----------
227404 |
以上查询使用了全表扫描,该表这里有22万记录，全表扫描已经不再适合。
检查该表，存在以下索引:
SQL> select index_name,index_type from user_indexes where table_name=’HS_INFO’;
INDEX_NAME INDEX_TYPE
------------------------------ ---------------------------
HSIDX_INFO1 FUNCTION-BASED NORMAL
HSIDX_INFO_SEARCHKEY DOMAIN
PK_HS_INFO NORMAL |
检查索引键值:
SQL> select index_name,column_name
2 from user_ind_columns where table_name =’HS_INFO’;
INDEX_NAME COLUMN_NAME
------------------------------ --------------------
HSIDX_INFO1 NUMORDER
HSIDX_INFO1 SYS_NC00024$
HSIDX_INFO_SEARCHKEY VC2INDEXWORDS
PK_HS_INFO NUMINFOGUID
SQL> desc hs_info
Name Null? Type
--------------------------------- -------- --------
NUMINFOGUID NOT NULL NUMBER(15)
NUMCATALOGGUID NOT NULL NUMBER(15)
INTTEXTTYPE NOT NULL NUMBER(38)
VC2TITLE NOT NULL VARCHAR2(60)
VC2AUTHOR VARCHAR2(100)
NUMPREVINFOGUID NUMBER(15)
NUMNEXTINFOGUID NUMBER(15)
NUMORDER NOT NULL NUMBER(15)
DATPUBLISHDATE NOT NULL DATE
INTPUBLISHSTATE NOT NULL NUMBER(38)
VC2PUBLISHERID VARCHAR2(30)
VC2INDEXWORDS VARCHAR2(200)
VC2WAPPREVPATH VARCHAR2(200)
VC2WEBPREVPATH VARCHAR2(200)
VC2WAP2PREVPATH VARCHAR2(200)
NUMVISITED NOT NULL NUMBER(15)
INTENABLEDFLAG NOT NULL NUMBER(38)
DATCREATETIME NOT NULL DATE
DATMODIFYTIME NOT NULL DATE
VC2NOTES VARCHAR2(1000)
INTINFOTYPE NOT NULL NUMBER(38)
VC2PRIZEFLAG VARCHAR2(1)
VC2DESC VARCHAR2(1000) |
6.决定创建新的索引以消除全表扫描
检查发现在numcatalogguid字段上并没有索引,该字段具有很好的区分度，考虑在该字段创建索引以消除全表扫描。
SQL> create index hs_info_NUMCATALOGGUID on hs_info(NUMCATALOGGUID);
Index created.
SQL> set autotrace trace explain
SQL> select i.vc2title,i.numinfoguid
2 from hs_info i where i.intenabledflag = 1
3 and i.intpublishstate = 1 and i.datpublishdate <=sysdate
4 and i.numcatalogguid = 3475
5 order by i.datpublishdate desc, i.numorder desc ;
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=12 Card=1 Bytes=106)
1 0 SORT (ORDER BY) (Cost=12 Card=1 Bytes=106)
2 1 TABLE ACCESS (BY INDEX ROWID) OF ’HS_INFO’ (Cost=2 Card=1
Bytes=106)
3 2 INDEX (RANGE SCAN) OF ’HS_INFO_NUMCATALOGGUID’
(NON-UNIQUE) (Cost=1 Card=1) |
7.观察系统状况
原大量等待消失
SQL> select sid,event,p1,p1text from v$session_wait where event not like ’SQL%’;
SID EVENT P1 P1TEXT
---------- ------------------------------ ----------
1 pmon timer 300 duration
2 rdbms ipc message 300 timeout
3 rdbms ipc message 300 timeout
6 rdbms ipc message 180000 timeout
59 rdbms ipc message 6000 timeout
118 rdbms ipc message 6000 timeout
275 rdbms ipc message 30000 timeout
147 rdbms ipc message 6000 timeout
62 rdbms ipc message 6000 timeout
11 rdbms ipc message 30000 timeout
4 rdbms ipc message 300 timeout
SID EVENT P1 P1TEXT
---------- ------------------------------ ----------
305 db file sequential read 17 file#
356 db file sequential read 17 file#
19 db file scattered read 17 file#
5 smon timer 300 sleep time
15 rows selected. |
持续观察的CPU使用情况
bash-2.03$ vmstat 3
procs memory page disk faults cpu
r b w swap free re mf pi po fr de sr s6 s9 s1 sd in sy cs us sy id
20 0 0 5421792 1503488 38 434 136 0 0 0 0 0 0 2 0 2931 7795 2622 91 9 0
23 1 0 5416080 1500632 95 734 56 0 0 0 0 0 0 0 0 2949 8057 2598 89 11 0
26 0 0 5412016 1498480 210 1170 21 5 5 0 0 0 2 1 0 3301 9647 3116 90 10 0
25 0 0 5394912 1490160 242 1606 56 0 0 0 0 0 0 1 0 3133 9318 2850 89 11 0
40 0 0 5390200 1488112 162 1393 66 0 0 0 0 0 0 0 0 2848 9080 2502 90 10 0
40 0 0 5377120 1481792 136 1180 120 2 2 0 0 0 1 1 0 2846 9099 2593 92 8 0
36 0 0 5363216 1475168 134 1169 53 0 0 0 0 0 3 2 0 2871 7989 2621 88 12 0
39 0 0 5348936 1469160 157 1448 210 0 0 0 0 0 0 0 0 3660 10062 3480 88 12 0
35 0 0 5344552 1466472 7 15 56 0 0 0 0 0 0 0 0 2885 7663 2635 92 8 0
34 0 0 5343016 1465416 44 386 77 0 0 0 0 0 0 0 0 3197 8486 2902 92 8 0
31 0 0 5331568 1459696 178 1491 122 0 0 0 0 0 0 3 0 3237 9461 3005 89 11 0
31 0 0 5317792 1453008 76 719 80 0 0 0 0 0 0 0 0 3292 8736 3025 93 7 0
31 2 0 5311144 1449552 235 1263 69 2 2 0 0 0 1 0 0 3473 9535 3357 88 12 0
25 0 0 5300240 1443920 108 757 18 2 2 0 0 0 1 1 0 2377 7876 2274 95 5 0
19 0 0 5295904 1441840 50 377 0 0 0 0 0 0 0 1 0 1915 6598 1599 98 1 0
----以上为创建索引之前部分
----以下为创建索引之后部分，CPU使用率恢复正常
procs memory page disk faults cpu
r b w swap free re mf pi po fr de sr s6 s9 s1 sd in sy cs us sy id
40 1 0 5290040 1439208 315 3894 8 2 2 0 0 0 1 6 0 3631 13414 5206 61 9 30
0 1 0 5237192 1414744 731 6749 45 0 0 0 0 0 2 7 0 3264 13558 4941 52 14 34
0 0 0 5163632 1380608 747 6585 10 0 0 0 0 0 0 1 0 2617 12291 3901 46 12 41
1 0 0 5090224 1348152 712 6079 29 0 0 0 0 0 0 6 0 2825 12416 4178 50 12 39
1 0 0 5023672 1317296 714 6183 24 0 0 0 0 0 0 5 0 3166 12424 4745 47 13 40
0 0 0 4955872 1287136 737 6258 16 0 0 0 0 0 0 3 0 2890 11777 4432 44 12 44
1 0 0 4887888 1256464 809 6234 8 2 2 0 0 0 0 2 0 2809 12066 4247 45 12 43
0 0 0 4828912 1228200 312 2364 13 5 5 0 0 0 2 1 0 2410 6816 3492 38 6 57
0 0 0 4856816 1240168 8 138 0 0 0 0 0 0 1 0 0 2314 4026 3232 34 4 62
0 0 0 4874176 1247712 0 86 0 0 0 0 0 0 0 0 0 2298 3930 3324 35 2 63
2 0 0 4926088 1270824 34 560 0 0 0 0 0 0 0 0 0 2192 4694 2612 29 16 55
0 0 0 5427320 1512952 53 694 0 0 0 0 0 0 3 2 0 2443 5085 3340 33 12 55
0 0 0 5509120 1553136 0 37 0 0 0 0 0 0 0 0 0 2309 3908 3321 35 1 64
0 0 0 5562048 1577000 16 234 0 0 0 0 0 0 0 0 0 2507 5187 3433 35 8 57
0 0 0 5665672 1623848 252 1896 8 2 2 0 0 0 1 0 0 2091 6548 2939 34 5 61
0 0 0 5654752 1618208 5 173 16 0 0 0 0 0 0 0 0 2226 4218 3051 35 4 60
0 0 0 5727024 1651120 28 254 0 0 0 0 0 0 0 0 0 2126 4224 2982 38 2 60
0 0 0 5723184 1648880 93 562 8 2 2 0 0 0 1 1 0 2371 5140 3432 38 3 59
0 0 0 5730744 1652512 7 177 26 2 2 0 0 0 1 0 0 2465 4442 3575 36 3 61 |
至此，此问题得以解决.
8.性能何以提高?
回答这个问题似乎是多余的,我只想重申一点:
有效的降低SQL的逻辑读是SQL优化的基本原则之一，我们来比较一下前后两种执行方式的读取及性能差异。
全表扫描的性能：
SQL> select i.vc2title,i.numinfoguid
2 from hs_info i where i.intenabledflag = 1
3 and i.intpublishstate = 1 and i.datpublishdate <=sysdate
4 and i.numcatalogguid = 3475
5 order by i.datpublishdate desc, i.numorder desc ;
352 rows selected.
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=541 Card=1 Bytes=106)
1 0 SORT (ORDER BY) (Cost=541 Card=1 Bytes=106)
2 1 TABLE ACCESS (FULL) OF ’HS_INFO’ (Cost=531 Card=1 Bytes=106)
Statistics
----------------------------------------------------------
0 recursive calls
25 db block gets
3499 consistent gets
258 physical reads
0 redo size
14279 bytes sent via SQL*Net to client
2222 bytes received via SQL*Net from client
25 SQL*Net roundtrips to/from client
2 sorts (memory)
0 sorts (disk)
352 rows processed |
使用索引的性能：
SQL> select i.vc2title,i.numinfoguid
2 from hs_info i where i.intenabledflag = 1
3 and i.intpublishstate = 1 and i.datpublishdate <=sysdate
4 and i.numcatalogguid = 3475
5 order by i.datpublishdate desc, i.numorder desc;
352 rows selected.
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=12 Card=1 Bytes=106)
1 0 SORT (ORDER BY) (Cost=12 Card=1 Bytes=106)
2 1 TABLE ACCESS (BY INDEX ROWID) OF ’HS_INFO’ (Cost=2 Card=1
Bytes=106)
3 2 INDEX (RANGE SCAN) OF ’HS_INFO_NUMCATALOGGUID’
(NON-UNIQUE) (Cost=1 Card=1)
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
89 consistent gets
0 physical reads
0 redo size
14279 bytes sent via SQL*Net to client
2222 bytes received via SQL*Net from client
25 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
352 rows processed |
consistent gets从3499 到89,我们看到性能得到了巨大的提高。
结束语:
通常,开发人员很少注意SQL代码的效率,他们更着眼于功能的实现. 至于性能问题通常被认为是次要的,而且在应用系统开发初期,由于数据库数据量较少,对于查询SQL语句等,不容易体会出各种SQL句法的性能差异.
但是一旦这些应用作为生产系统上线运行,随着数据库中数据量的增加,大量并发访问,系统的响应速度可能就会成为系统需要解决的最主要的问题之一.
在少量用户下性能可以接受的SQL,可能在大量用户并发的条件下就会成为性能瓶颈。在我这个案例中,开发人员很难相信仅只一条SQL语句就导致了整个数据库的性能下降。
然而事实就是如此,一条低效的SQL语句就可能毁掉你的数据库,所以在系统设计及开发过程中,你必须考虑到诸多细节,严格的测试也是提早发现问题的有效方法。
如果不幸以上环节都被忽略,那么,DBA(也许就是你)就是最后的一环,你必须能够快速的诊断并解决各种复杂问题。 |
