[Rack] noisebridge.net apache runaway

Andy Isaacson adi at hexapodia.org
Wed Oct 31 20:20:41 UTC 2012


The webserver on noisebridge.net was paging itself to death with a load
average of 153 and was unable to serve the website.  I killed all the
apache worker processes and things seem to be back to normal.  We didn't
have a metrics graphing system installed, so there's not much I can do
in the post-mortem.  I've installed munin so that we have somewhat more
data in the future.

The MaxClients parameter at 150 is probably too high for the 2GB of RAM
we have allocated.  Looks like minimum "PSS" according to smem(1) is
about 25 MB per apache worker, so I'm going to loewr MaxClients to 50 to
leave some room for mysqld.

-andy


More information about the Rack mailing list