A couple of hours ago I wrote a post about migrating web services to a Debian VM running Squeeze, from one which had been running Lenny.  I said I’d switched to the prefork MPM under Apache2.

Well, if you’re reading this on my site, you’re reading it via the worker MPM once again – only a couple of hours later.  It became obvious pretty quickly once the site had real web pages, and real users, that prefork was not going to cut it.  The VM’s are small, only 256MB of memory and so I can’t run many Apache2 processes.  Although I tested a lot of accesses against PHP based pages using Apache’s AB, I missed doing some testing of both PHP content and the large amount of static content that goes with it (such as style sheets, javascript, images, etc.) at the same time.

Under those conditions, the server needed either so many Apache processes that it filled memory, or it reached the limits I had set and page loads took 20+ seconds.

So, I quickly switched back to the worker MPM and PHP running under Fast CGI, and the page loads are back down to 2 seconds or so on average.

I still have some work to do, to make sure I don’t start too many PHP5 CGI processes, but at least the sites are useable again.