AIX6.1/11.2.0.3数据库上关于SWAP的一个BUG

昨天去南京某客户那里调优新上线的业务数据库，在查看alert.log日志时发现在过去的一段时间里，每过几个小时或间隔一段时间，就会报类以下的内容： Thu Aug 21 09:01:26 2014WARNING: Heavy swapping observed on system in last 5 mins.pct of memory swappe

昨天去南京某客户那里调优新上线的业务数据库，在查看alert.log日志时发现在过去的一段时间里，每过几个小时或间隔一段时间，就会报类似以下的内容：
Thu Aug 21 09:01:26 2014 WARNING: Heavy swapping observed on system in last 5 mins. pct of memory swapped in [8.42%] pct of memory swapped out [2.16%]. Please make sure there is no memory pressure and the SGA and PGA are configured correctly. Look at DBRM trace file for more details.
Thu Aug 21 14:56:27 2014 WARNING: Heavy swapping observed on system in last 5 mins. pct of memory swapped in [5.40%] pct of memory swapped out [8.63%]. Please make sure there is no memory pressure and the SGA and PGA are configured correctly. Look at DBRM trace file for more details.
......
Sat Oct 18 22:13:48 2014 WARNING: Heavy swapping observed on system in last 5 mins. pct of memory swapped in [7.76%] pct of memory swapped out [0.33%]. Please make sure there is no memory pressure and the SGA and PGA are configured correctly. Look at DBRM trace file for more details.
客户的环境是IBM P570，AIX 6.1，安装了Oracle 11.2.0.3单实例数据库，物理内存64G，仅仅分配了20G给SGA，采用memory自动管理
查阅了一下MOS，发现是AIX平台上的一个bug，相关文档为：[1508575.1]
对应的数据库和平台：
Oracle Database - Enterprise Edition - Version 11.2.0.3 to 11.2.0.3 [Release 11.2] IBM AIX on POWER Systems (64-bit)
症状：

There is new warning message in alert.log in 11.2.0.3 similar to

WARNING: Heavy swapping observed on system in last 5 mins.
pct of memory swapped in [2.08%] pct of memory swapped out [0.12%].
Please make sure there is no memory pressure and the SGA and PGA
are configured correctly. Look at DBRM trace file for more details.

On AIX platform this message can be seen even when there is no virtual memory swapping at all. --物理内存足够，而且根本没有使用swap交换空间

You may compare the vmstat from AIX level with DBRM trace file entries to see the differences.

原因：

The issue is caused by unpublished Bug:14731911.

Swap usage messages are based on statistics that do not reflect the actual usage.

The v$osstat does not reflect proper stats for the swap space paging.

解决方法：

Apply Patch:11801934 on top of your IBM AIX on POWER Systems (64-bit) platform.

P.S: Bug is port-specific. --这个bug是针对端口指定的平台的

The issue is fixed in patchset 11.2.0.4 and release 12.1. --说是在12.1的patch中修复了，但实际上12.1还是会有这个问题，会有ora-700错误，详见文档：[ID 1919850.1]

来看一下BUG：14731911的描述：

Bug 属性

---

类型	B - Defect	已在产品版本中修复
严重性	2 - Severe Loss of Service	产品版本	11.2.0.3
状态	96 - Closed, Duplicate Bug	平台	212 - IBM AIX on POWER Systems (64-bit)
创建时间	2012-10-8	平台版本	6.1
更新时间	2014-10-11	基本 Bug	11801934
数据库版本	11.2.0.3	影响平台	Port-Specific
产品源	Oracle	与此 Bug 相关的知识, 补丁程序和 Bug

相关产品

---

产品线	Oracle Database Products	系列	Oracle Database Suite
区域	Oracle Database	产品	5 - Oracle Database - Enterprise Edition

Hdr: 14731911 11.2.0.3 RDBMS 11.2.0.3 VOS PRODID-5 PORTID-212 11801934
Abstract: FALSE SWAP WARNING MESSAGES PRINTED TO ALERT.LOG ON AIX
 
*** 10/08/12 04:52 am ***
  
  
  BUG TYPE CHOSEN
  ===============
  Code
  
  SubComponent: Virtual Operating System
  ======================================
  DETAILED PROBLEM DESCRIPTION
  ============================
  Oracle process seems to check wrong OS local statistic (which include also
  FILESYSTEM caching etc.)
  
  Alert log shows WARNING: Heavy swapping observed on system in last 5 mins.
  pct of memory swapped in [2.08%] pct of memory swapped out [0.12%].
  Please make sure there is no memory pressure and the SGA and PGA 
  are configured correctly. Look at DBRM trace file for more details.
  
  but this is not reflected at OS level.
  
  DIAGNOSTIC ANALYSIS
  ===================
  1. nmon shows virtual memory swapping does not occur at all - see attached file --nmon根本没有监控到swap动作
  
  2. Oracle Database Server is 11.2.0.3 and contains fix for 10220118
  
  3. Server configuration
  real mem: 144GB
  lowest value of fre memory : 87,65 GB --剩余内存充足
  
  4. DBRM seems to use a wrong OS statistics - trace file is attached
  
  WORKAROUND?
  ===========
  No
  
  TECHNICAL IMPACT
  ================
  Wrong diagnostic analyze.
  Message is bothering customer's DBA when in fact the warning message is
  misleading
  
  RELATED ISSUES (bugs, forums, RFAs)
  ===================================
  http://myforums.oracle.com/jive3/thread.jspa?threadID=1104581 
  10220118
  
  HOW OFTEN DOES THE ISSUE REPRODUCE AT CUSTOMER SITE?
  ====================================================
  Always
  
  DOES THE ISSUE REPRODUCE INTERNALLY?
  ====================================
  No
  
  EXPLAIN WHY THE ISSUE WAS NOT TESTED INTERNALLY.
  ================================================
  Unavailable Data Volume
  
  IS A TESTCASE AVAILABLE?
  ========================
  No
  
  Link to IPS Package:
  ====================
  not available

DBRM(Database Resource Manager)是11gR2中新特性中出现的后台进程，会在alert.log告警日志中反映OS操作系统最近5分钟是否有剧烈的swap活动，而在AIX平台上，由于BUG：14731911的存在，oracle的这个进程谎报了内存进行了swapin和swapout动作。我们知道，只有当物理内存真的不够用的情况下，才会去用swap（通常会配置成物理内存的2倍），而swap是非常耗费性能的（从物理磁盘读写）。但是个人认为这个bug的危害性并不大，仅仅只是在alert.log日志中报了一个WARNING，并没有因为这个影响导致对数据库更加负面的影响，因此是否打补丁到11.2.0.4就见仁见智了，如果想让alert.log平安无事，那么就可以升级一下patch。当然了，如果真的是因为OS内存吃紧造成的swap动作，就要区别对待了，因为此时的确会对数据库造成严重影响。要区分是否真的内存不足而非系统误报，那么主要还是通过nmon，topas，vmstat等监控工具来进行分析（linux下还可以用free监控）

对于AIX平台，其实还有另一个bug，只不过是unpublished base bug，而不是port-specific bug

AIX Platform

If your Platform is IBM-AIX then this is not the only possible reason for this alert log message.
For IBM AIX on POWER Systems (64-bit), there is also next known port-specific bug:
Bug 14731911 - FALSE SWAP WARNING MESSAGES PRINTED TO ALERT.LOG ON AIX
with unpublished base bug:
Bug 11801934 : WRONG PAGE-IN AND PAGE-OUT OS VM STATS IN AIX.

在vmware平台中的这个WARNING信息，如果不是bug引起，则很有可能和ora-04031/ora-04030相关，这个就严重多了

VMWare

Under VMWare, the messages may perhaps indicate a more serious issue, even when no memory related ORA-4031/ORA-4030 errors are reported.
Under circumstances, an instance in a virtual machine may be simply terminated by PMON due to error 471 without further errors in the alert log.
The OS logs may in such case report an out of memory condition like below:

[root@vmh ~]# grep Kill /var/log/messages*
/var/log/messages-20140629:Jun 27 18:29:06 vmh-msfc-dodp02 kernel: [1895074.304941] Out of memory: Kill process 42094 (oracle) score 391 or sacrifice child
/var/log/messages-20140629:Jun 27 18:29:06 vmh-msfc-dodp02 kernel: [1895074.305203] Killed process 42094, UID 303, (oracle) total-vm:189081588kB, anon-rss:27412kB, file-rss:109612

通常解决OS内存swap问题有以下几种方案：

1. 诊断是否存在内存泄露的进程，解决内存泄露 2. 调优SGA/PGA，减少oracle对内存的占用 3. 利用/proc/sys/vm/drop_caches，暂时释放一些cache的内存（Linux） 4. 调整系统VM内存管理参数， 例如Linux上sysctl.conf中的以下几个参数:

vm.min_free_kbytes:Raising the value in /proc/sys/vm/min_free_kbytes will cause the system to start reclaiming memory at an earlier time than it would have before.

vm.vfs_cache_pressure:At the default value of vfs_cache_pressure = 100 the kernel will attempt to reclaim dentries and inodes at a “fair” rate with respect to pagecache and swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer to retain dentry and inode caches. Increasing vfs_cache_pressure beyond 100 causes the kernel to prefer to reclaim dentries and inodes.

vm.swappiness:default 60,Apparently /proc/sys/vm/swappiness on Red Hat Linux allows the admin to tune how aggressively the kernel swaps out processes'memory. Decreasing the swappiness setting may result in improved Directory performance as the kernel holds more of the server process in memory longer before swapping it out.

设置以下值，以减少OOM(Out Of Memory)的可能性:

# Oracle-Validated setting for vm.min_free_kbytes is 51200 to avoid OOM killer
vm.min_free_kbytes = 51200
vm.swappiness = 40
vm.vfs_cache_pressure = 200