[Rack] www.noisebridge wedge, me and SuperQ restarted it
ben at nerp.net
Tue Sep 11 18:13:33 UTC 2012
Mostly the system was in a swap use death spiral thanks to munin and
apache. I purged munin and stopped apache. That dropped the wap use from
1.9G to 0.2G.
I shutdown the instance and added a slice to it. This bumped the ram from
1G to 2G. That should help with traffic spikes in the future.
superq at m2:~$ free -m
total used free shared buffers cached
Mem: 2003 850 1153 0 13 196
-/+ buffers/cache: 640 1363
Swap: 2047 0 2047
I'll keep an eye on memory use and see if there is any need to tune
apache to reduce the memory footprint.
We could also try swapping to apache2-mpm-itk instead of
apache2-mpm-prefork. That should reduce the memory footprint and also let
us tweak the UIDs for the various vhosts.
On Tue, 11 Sep 2012, Danny O'Brien wrote:
> It wasn't clear what took it down -- the CPU load was 150 by the time I
> magnaged to get on, and a munin process was taking up 98% of CPU, but
> Apache was also pretty wedged.
> I'm not necessarily giving munin the stink-eye here, but after trying to
> shut it down various ways, I ended up apt-get remove'ing it. The
> configuration files will still be there, I didn't apt-get purge, but
> there was no documentation on rack about putting a munin process on www,
> nor did I see anything in the motd.
> I feel that it's important to keep www running with a low number of
> critical systems -- essentially www, ssh, the wiki, email, dns, mailman
> and its own backup systems. It's supposed to stay up all the time, and a
> lot of things break if it doesn't.
> Let's walk through what else is running on there now, how it got there,
> whether we need it, and how to maintain it if we do need it. And we
> should restart the MOTD changelog too.
> Rack mailing list
> Rack at lists.noisebridge.net
More information about the Rack