Re: 2.2.STABLE5 Possible Memory Leak?

From: Martin Robbins <tgdroma5@dont-contact.us>
Date: Wed, 10 Nov 1999 11:21:40 +0100

Hi Jim,

I'm running Squid 2.2STABLE5 on Solaris 2.6 patched to 1st Oct 1999,
compiled with -mv8 -O3 and using egcs-1.1.2

I have had similar trouble in the past, a very effective fix for
my problem was enabling priority paging (add "set priority_paging=1"
to /etc/system) (as outlined in the FAQ
http://www.squid-cache.org/Doc/FAQ/FAQ-14.html#ss14.1 and in the Sun
FaqDoc 2833) also requires a patch on 2.6 to work which is in the Rec
patch cluster.

The problem I had the memory was eaten and would very slowly get
returned, but very slowly. Solaris tries to cache the world in RAM
by all accounts. Plus I put the DNS cache on another machine very
close by.

Also I had to set the cache_mem to 128 Mb on a system with 1024Mb
of RAM and it sits happily with some free and runs well.

Not sure this will help, and if anyone else can figure the mysteries
of Solaris, Squid and RAM please let me know.

Excuse me if you have already done this or it doesn't help.

regards,
Martin

> On Tue, Nov 09, 1999 at 02:43:24PM -0500, Jim Richey wrote:
> Running on Solaris 2.6 with latest patches. Compiled with gcc 2.8.1 with
> --enable-dlmalloc, --enable-cache-digests, and --disable-ident-lookups.
> When the system is first booted there is about 500MB of free memory and
> squid has 22MB according to top. After six days the amount of free
> memory is 375MB and top reports that squid has 44MB. If I stop squid
> with squid -k shutdown the system gets back the 44MB squid was using,
> but the other 81MB never gets freed. There are no other applications
> running on this system, just squid and the Solaris OS programs. If I
> reboot the system, it comes back up with the 500MB free.

> Are you running named or some similar caching domain name server on
> Solaris? At this point, named eats up about 60MB on our BSD/OS server,
> and you might not have been looking there for where your memory went.
>
> Otherwise, I'd say you've got a Solaris kernel bug. It should not be
> possible for an application to allocate RAM from the kernel in a way
> that does not get returned when the application terminates.

> HTH
> -- Clifton

---
Martin Robbins
CIT-SEO-SEC-IPE Internet Services
Swisscom AG, Bern, CH
Received on Wed Nov 10 1999 - 03:33:54 MST

This archive was generated by hypermail pre-2.1.9 : Wed Apr 09 2008 - 11:57:32 MDT