agentzh
Apache stops serving
We have implemented this mod_libmemcached on a cluster of 4 apache 2.2 web servers using a separate memcached server, and initially had good results. But eventually each of the apache services would fail and stop serving pages (sometimes after a day and sometimes after and hour). Before we found mod_libmemcached we had mod_cache set-up with mod_disk_cache and had no problems.
Here are some of errors from our logs:
THIS ERROR IS IN OUR APACHE ERROR LOG: libmemcached_cache: content overflew.
THIS ERROR IS IN OUR SYSLOG [Wed Jun 16 14:44:40 2010] errorSoftware caused connection abort: cache: error returned while trying to return libmemcached cached data
We also see some MaxConnections errors in the apache error log at the same time, but we think those were caused by the above issues.
We can tell that when we first fire up the servers, they run very well, but hour by hour they start to degrade until eventually they fail completely . They each fail at different times, but eventually they all stop serving.
We would really like to continue with this solution, since it fits perfectly with our set-up. If any one can suggest some possible reasons or solutions, it would be greatly appreciated.
Thanks for the report. I'll look into this this weekend :)
A temporary work-around is to decrease the "MaxRequestsPerChild" setting in your httpd.conf to force the worker processes refresh themselves early if you're using the prefork mpm.
Thanks so much for checking this out. We have taken your advice on the reduction of MaxRequestsPerChild and will see how that goes.
Also, I was wondering if you had any advice on the issue with using gzip and mod_cache. It seems that if you use mod_deflate, mod_cache then uses the agents Accept request headers for cache key creation causing a very low cache hit rate (practically 0 for some browsers) because of the Vary header in the response. If we could crack that problem we would be home free with your mod! Any help would be greatly appreciated. Thanks again.