Re: [squid-users] Squid Core Dumping

On Wed, Jan 17, 2007, Everett, Vernon wrote:

> The one used for authentication, is misbehaving.
> >From time to time - and it appears random - the squid application will
> crash and burn, leaving only a core dump and log message to prove it
> ever existed.
> RunCache respawns squid, and everything gets back up and running, but my
> users have to re-authenticate, and they find this annoying.
> Has anybody seen this before, and what (if anything) did you do to
> correct it?

I'd first look at upgrading it to the latest Squid-2.6. It'll be easier to debug
and trace problems than 2.5.

> Jan 15 15:09:18 prowler squid[15485]: [ID 702911 daemon.alert] Too many
> queued ntlmauthenticator requests (1001 on 200)
> Jan 15 15:09:21 prowler genunix: [ID 603404 kern.notice] NOTICE:
> core_log: squid[15485] core dumped:
> /var/core/core_prowler_squid_60001_60001_1168841358_15485

That queue depth is pretty deep. It looks like its spiralling out of control.
Whats the system load at this point?

> cache_dir ufs /u01/app/squid-front/cache 100 16 256

You really should be using aufs instead of ufs under Solaris. Solaris has a sane
enough threading model for this to work well. It'll drop system load and improve
performance.

> coredump_dir /u01/app/squid-front/var/cache

Could you fire up gdb on that coredump and email me a back trace (using 'bt') ?

Adrian
Received on Tue Jan 16 2007 - 18:14:17 MST