I have a dedicated server that runs a vbulletin forum. Right now we're opting to lighttpd as it is much reliable then apache 1.x. However, the lighttpd crashes alot with following message but i'm not sure if its the real cause to the crash
2006-10-06 13:41:22: (mod_fastcgi.c.2388) unexpected end-of-file (perhaps the fastcgi process died): pid: 20014 socket: unix:/tmp/php-fastcgi.socket-2
2006-10-06 13:41:22: (mod_fastcgi.c.3173) response not received, request sent: 26565 on socket: unix:/tmp/php-fastcgi.socket-2 for /newforums/newattachment.php , closing connection
2006-10-06 13:42:38: (mod_fastcgi.c.2388) unexpected end-of-file (perhaps the fastcgi process died): pid: 20027 socket: unix:/tmp/php-fastcgi.socket-3
2006-10-06 13:42:38: (mod_fastcgi.c.3173) response not received, request sent: 20037 on socket: unix:/tmp/php-fastcgi.socket-3 for /newforums/newattachment.php , closing connection
Well this is just a guess here and may be totally wrong, but it it always /newforums/newattachment.php? I'm thinking that maybe people are sending large attachments too big or connected for too long so the fcgi process dies?
Also I found this topic on the Lighttpd forums, it might help.
Updates : we revert the lighttpd version to 1.4.11 and its been running for 17hours straight with out problem now.
I've enabled the statistic, and it shows that all backend's died, disable, load and overloaded is 0 and of course connected variable varies from each another. Range from 2k to 366k. The last backend has the highest i think.
Here's my status
Uptime17 hours 23 min 56 s
Started at 2006-10-07 16:55:55
i am glade to see now you dont have any problem
PHP Accelerators ( apc / ea / xcache ) usually making vb forum 3-4 times faster ( mean 3-4 times lower cpu usage ) and i think main reason that you dont have dead or overloaded fastcgi proccess is becouse now php is spending lower time to proccess php scripts .
If you leave min- and max-procs at 4, you'll end up with four separate eAccelerator caches. However, if a php parent segfaults, you'll still have three groups available to respond to queries. For this reason, Jan recommends setting min- and max-procs to 2.