On an apache server on a big hardware setup, sometimes, for no long, simple request are served really slowly.
I've activated request duration logging, and for a very small number of request ( 0.05% served over 60s ) , apache could take over 300s to serve a small 5k js script.
Low bandwidth client connection could not explain as this happen all days, at same rate.
Server loadaverage is always low, no disk/io bottleneck.
Problem is that at some random times, mysql server comes with lot of Sleeping connexions, and could blow away, probably cause by apache childs "freezing".
I'm actually testing if that could be a KeepAlive problem but ...
Where to look first ?
thanks in advance ...
== Details == Run apache 2.2.22(mpm_prefork) with mod_php on gentoo/GNULinux
The box is quad Intel(R) Xeon(R) CPU X5690 @ 3.47GHz ( 24 cores) , with 48 Go Ram. Running Raid1 Hard Array on LSI Logic / Symbios Logic MegaRAID It's a Numa memory architecture box.