Recently I found that one of the servers I look after that runs a high profile site was generating semi-high load at traffic peaks. You could generally say that this would be understandable but the server was shooting up to a load of around 10 for a few seconds and with that load jump I was able to graph an increase of Apache processes on top of it. Again though, this would generally be considered normal, but knowing how well the server performs and having nginx sitting on top handling all static content I knew something wasn’t quite right.
Looking through the logs I found quite a lot of requests from a badly written spider which was generating a lot of server load when it hit the server, but after IP banning the culprit I also found several instances of Apache waking it’s child processes.
127.0.0.1 - - [23/Mar/2010:17:02:38 +0000] "OPTIONS * HTTP/1.0" 200 - "-" "Apache (internal dummy connection)"
This in itself is normal and nothing to worry about, generally, it is known and also mentioned on the Apache wiki (explanation here) that servers running 2.2.6 or lower can suffer performance problems due to how Apache wakes it’s children because it used GET / to wake them, but I run 2.2.8 which doesn’t actually used that same method.
I applied the well known “work around” for this to test it and see if it made any difference.
RewriteEngine on
RewriteCond %{HTTP_USER_AGENT} ^.*internal dummy connection.*$ [NC]
RewriteRule ^/$ /blank.html [L]
Adding the virtual hosts with dynamic content will simple rewrite Apache’s internal dummy connection requests to blank.html, thus stopping it from trying to generate all dynamic page content just to wake the child.
All seems normal but time will tell as to whether this makes a difference with versions > 2.2.6.
