apache的MaxRequestsPerChild指令设置不合理导致apache崩潰

apache的MaxRequestsPerChild指令设置不合理导致并发连接过高

 

服务器负荷增大，尝试优化和调整服务器，首先是从 Apache 的优化和调整开始的。之前考虑到我们的生产机负荷并不大，因此可以让apache的每一个派生子进程都尽可能处理更多的请求再释放，这样有利于促进子进程的复用，从而提高处理效率。于是我将apache的MaxRequestsPerChild参数调整为4000，一直大半年相安无事。
但是今天却出现了一个问题，网站突然打不开了，症状是浏览器始终处于正在打开网页的状态。检查后排除了DDoS攻击的可能。再查看apache进程，居然没有！赶紧查看apache错误日志/var/log/httpd/error_log，发现了以下错误日志：

[Tue May 08 12:21:22 2012] [alert] (11)Resource temporarily unavailable: setuid: unable to change to uid: 48
[Tue May 08 12:21:23 2012] [alert] Child 8762 returned a Fatal error... Apache is exiting!
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (43)Identifier removed: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 12:21:24 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 12:21:25 2012] [emerg] (22)Invalid argument: couldn't release the accept mutex
[Tue May 08 15:39:47 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 15:39:47 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 15:39:47 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
[Tue May 08 15:39:47 2012] [emerg] (22)Invalid argument: couldn't grab the accept mutex
.

.

.

 

从时间上看，正是这次故障发生的时间点。重新启动apache后恢复正常。最后发现是 MaxRequestsPerChild 参数设置不合理。再次仔细阅读 Apache 手册关于 MaxRequestsPerChild 的说明后，认为在 prefork 下，KeepAlive 开启的情况下，每一个新连接都会导致一个 Apache 子进程开启， MaxRequestsPerChild 设置过大，导致新连接产生新的子进程后，长期空闲 ，这样并不一定合理。

根据我们网站的访问情况，其实远远不需要那么大的 MaxRequestsPerChild 值，每个用户点击网站的频率也不会很高，所以 MaxRequestsPerChild 设置得太高了反而适得其反，相反应该尽快释放连接，尽快回收系统资源，以尽可能快速的满足新的请求的连接需要。于是将 MaxRequestsPerChild 重新调整为64，估计不会再有这个问题产生了，具体情况是否如此还需要长期的观察。