Re: [squid-users] Memory Leaks after OS upgrade

From: Amos Jeffries <squid3_at_treenet.co.nz>
Date: Sat, 09 Aug 2014 19:12:29 +1200

On 9/08/2014 6:28 a.m., "Nils Hügelmann (anonymoX.net)" wrote:
> Hi,
>
> after upgrading to opensuse 13.1, squid3.4 (same with current 3.HEAD)
> leaks memory after a short time and stops responding to snmp queries(not
> sure if that's related).
>

From what version? If it was 3.1 or older they have significantly
smaller memory cache by default.

> In a test setup i can reproduce the issue as follows:
>
> Test with 100 rps: Everything good
> Test with 250 rps: After some minutes, only 100 to 200 requessts pass,
> memory increases with constant speed, some snmp queries fail (timeout)
> or take longer to process, nothing in log
> Test with 500 rps: Only around 100 queries pass, memory increases
> faster, most snmp queries fail, nothing in logs
>
> Same with 1 or multiple workers.
>
> In cachemgr, i see that Cumulative allocated volume continues to
> increase to values like 7 GB while actual VIRT mem is 300MB, RSS 160MB

Cumulated allocation volume is an incrementing count of how much memory
has been allocated, it does not account for how much has been freed.

If the VIRT mem and current size are constant then there is no leak.

That said, we do know that 3.4 will free memory cache allocations a bit
ater than it should. Just not where they are happening.

>
> Any ideas how to fix this or otherwise how to get more debug information?
>

Your snmp_* settings would be useful.
 Also are you using SMP support? (workers directive)

http://wiki.squid-cache.org/KnowledgeBase/DebugSections
http://wiki.squid-cache.org/SquidFaq/BugReporting
http://wiki.squid-cache.org/KnowledgeBase/OpenSUSE

Amos
Received on Sat Aug 09 2014 - 07:12:48 MDT

This archive was generated by hypermail 2.2.0 : Sat Aug 09 2014 - 12:00:05 MDT