案例描述:
这是一个大型生产系统
问题出现时系统累计大量用户进程
用户请求得不到及时响应,新的进程不断尝试建立连接
连接数很快被用完
数据库版本:9.2.0.3
操作系统:Solaris8
1.检查alert文件
日志中记录如下错误信息,说明磁盘异步IO出现问题:
WARNING: aiowait timed out 2 times
Tue Aug 26 15:33:32 2003
WARNING: aiowait timed out 2 times
Tue Aug 26 15:33:34 2003
WARNING: aiowait timed out 2 times
Tue Aug 26 15:33:36 2003
WARNING: aiowait timed out 2 times
Tue Aug 26 15:33:38 2003
WARNING: aiowait timed out 2 times
Tue Aug 26 15:33:43 2003
WARNING: aiowait timed out 1 times
Tue Aug 26 15:33:46 2003
WARNING: aiowait timed out 1 times
Tue Aug 26 15:33:49 2003
WARNING: aiowait timed out 1 times
Tue Aug 26 15:33:51 2003
WARNING: aiowait timed out 1 times
Tue Aug 26 15:33:52 2003
WARNING: aiowait timed out 1 times
Tue Aug 26 15:33:53 2003
WARNING: aiowait timed out 1 times
.............
我们知道在SUN的某些版本上异步IO存在问题,而异步IO缺省是打开的
代码:
SQL show parameter disk_a
NAME TYPEVALUE
------------------------------------ ----------- ------------------------------
disk_asynch_io boolean 'TRUE'
针对此问题,我们停用了数据库的异步IO写入。
2.共享内存问题
alert文件中还记录了以下错误信息:
Tue Aug 26 21:37:40 2003
WARNING: EINVAL creating segment of size 0x0000000190400000
fix shm parameters in /etc/system or equivalent
该信息说明内核参数设置过小或者和SGA不匹配
我们检查system配置文件
$ cat /etc/system
.......................
set shmsys:shminfo_shmmax=4096000000
set shmsys:shminfo_shmmin=1
set shmsys:shminfo_shmmni=200
set shmsys:shminfo_shmseg=200
set semsys:seminfo_semmap=1024
set semsys:seminfo_semmni=2048
set semsys:seminfo_semmns=2048
set semsys:seminfo_semmnu=2048
set semsys:seminfo_semume=200
set semsys:seminfo_semmsl=2048
我们发现最大共享内存设置仅有4G
3.检查SGA设置
SQL*Plus: Release 9.2.0.3.0 - ProdUCtion on 星期二 8月 26 21:46:35 2003
Copyright (c) 1982, 2002, Oracle Corporation. All rights reserved.
Connected to:
Oracle9i Enterprise Edition Release 9.2.0.3.0 - 64bit Production
With the Partitioning, OLAP and Oracle Data Mining options
JServer Release 9.2.0.3.0 - Production
SQL show sga
Total System Global Area 6695660272 bytes
Fixed Size 740080 bytes
Variable Size 2399141888 bytes
Database Buffers 4294967296 bytes
Redo Buffers 811008 bytes
我们发现SGA设置接近7G,这也就是步骤2中错误提示出现的原因
4.交换区问题
我们用top工具检查系统运行状况
代码:
# /usr/local/bin/top
last pid: 16899;load averages:0.82,0.81,0.83 21:49:05
1230 processes:1228 sleeping, 1 running, 1 on cpu
CPU states: 50.1% idle,7.4% user,8.6% kernel, 33.9% iowait,0.0% swap
Memory: 8192M real, 118M free, 12G swap in use, 11G swap free
PID USERNAME THR PRI NICESIZE RES STATETIMECPU COMMAND
15751 oracle11440 6456M 6408M sleep0:020.49% oracle
15725 oracle11580 6458M 6410M sleep0:020.46% oracle
251 root12480 7096K 1944K sleep126:000.45% picld
16540 oracle11580 6458M 6411M sleep0:010.45% oracle
16766 root 1430 3744K 2248K cpu/10:010.41% top
16408 oracle11580 6457M 6410M sleep0:010.34% oracle
15989 oracle11580 6458M 6409M sleep0:010.34% oracle
15919 oracle11580 6457M 6409M sleep0:020.30% oracle
16404 oracle11580 6457M 6409M sleep0:000.28% oracle
16327 oracle11550 6457M 6410M sleep0:000.27% oracle
14870 oracle11580 6457M 6412M sleep0:050.24% oracle
16851 oracle11350 6457M 6411M sleep0:000.22% oracle
16467 oracle11580 6457M 6409M sleep0:000.21% oracle
16163 oracle11580 6457M 6408M sleep0:030.21% oracle
' 15159 oracle11580 6457M 6408M sleep0:050.21% oracle'
Memory: 8192M real, 118M free, 12G swap in use, 11G swap free
我们发现系统仅有8G RAM,物理内存仅有118M可用,现在SWAP区使用了12G
我们初步作出以下判定:
SGA设置过大(将近7G)导致运行时产生大量交换
大量SWAP交换进而引发磁盘问题,这也就应该是我们第一步看到
WARNING: aiowait timed out 1 times的原因
大量交换导致数据库性能急剧下降,进而导致用户请求得不到快速响应,堵塞、累积,直至数据库失去响应
5.解决方案
此问题主要是由于SGA设置不当引起,我们马上缩小了SGA设置:
SQL show sga
Total System Global Area 3591870848 bytes
Fixed Size 735616 bytes
Variable Size 1442840576 bytes
Database Buffers 2147483648 bytes
Redo Buffers 811008 bytes
此时,数据库减少了交换,达到了稳定运行,用户请求可以得到快速响应。
问题解决完成.
6.系统状态
调整后系统运行状况:
代码:
$ top
last pid: 12745;load averages:0.46,0.79,0.65 22:22:49
228 processes: 227 sleeping, 1 on cpu
CPU states: 92.3% idle,5.0% user,1.6% kernel,1.1% iowait,0.0% swap
Memory: 8192M real, 3817M free, 4015M swap in use, 15G swap free
PID USERNAME THR PRI NICESIZE RES STATETIMECPU COMMAND
12610 oracle 1510 3511M 22M sleep0:041.96% oracle
12595 oracle 1480 3511M 22M sleep0:030.92% oracle
12630 oracle 1380 3511M 21M sleep0:010.84% oracle
12614 oracle 1460 3511M 22M sleep0:010.64% oracle
12620 oracle 1580 3511M 22M sleep0:010.53% oracle
12709 oracle 1480 3511M 21M sleep0:000.45% oracle
265 root11380 7032K 1920K sleep3:160.42% picld
12729 oracle 1 00 3511M 20M sleep0:000.26% oracle
12741 oracle 1580 2768K 1760K cpu/30:000.19% top
12745 oracle 1440 3506M 16M sleep0:000.17% oracle
12711 oracle 1480 3506M 16M sleep0:000.11% oracle
12738 oracle 1430 3506M 16M sleep0:000.06% oracle
7606 oracle 1450 17M 6928K sleep0:070.05% tnslsnr
12721 oracle 1340 3506M 16M sleep0:000.05% oracle
'12723 oracle 1530 3506M 16M sleep0:000.05% oracle'
该系统调整完以后,一直稳定运行至今.
一点总结:
这个案例和前面我提到的另外一个极其相似,同样都是SGA设置不当引起的数据库问题
本身并不复杂
这一类问题应该在数据库规划和建设阶段就避免掉.
其时,该问题对我更像是个心理测试,当所有老板都站在你背后的时候,你能否冷静快速的找到并解决问题.
关于SUN上的aiowait timed out 有很多总情况及诱因