Catching Apache segfaults due to eAccelerator

Posted by Jeff Beeman on Thu, 05/15/2008 - 07:59

Last night we migrated our Linode for Gamers With Jobs to a new Xen VPS and we've noticed a significant performance boost. We did, however, start encountering a random issue with segmentation faults in Apache. If you haven't seen this happen before, it tends to begin innocently with one Apache process dying, and therefore giving errors (usually WSOD), but quickly balloons into dozens of dead processes. It essentially hoses Apache.

Apparently the issue is due to eAccelerator, so I reinstalled it and cleared its caches in the hope that it might limit its occurrence. Just in case, though, 2bits has a great fix for it, using the logwatcher script by Firebright, Inc. I was able to quickly get it going, and the only difference is that I used the Debian init.d script provided by Derek Laventure to run it.