[squid-users] Unbounded memory growth when running on a Virtuozzo VPS

From: Will Roberts <ironwill42_at_gmail.com>
Date: Sat, 02 Jul 2011 15:23:40 -0400

Hi,

I'm running about 50ish squids on different machines, and they work
great except for 4 of them. The 4 I'm having problems with are running
on Debian 6 in a Virtuozzo VPS.

Kernel: 2.6.18-028stab070.5
Squid3: 3.1.6-1.2

Non-access related config options:
cache_dir ufs /var/spool/squid3 100 16 256
cache_mem 10 MB
external_acl_type ipauth children=1 cache=100 %SRC /etc/squid3/ip_auth.sh

Basically what I see is when squid starts the main process uses about
16m (RES) then on the machines where I'm having trouble the process
grows at a rate of 100MB+ a day with no end in sight then once the
memory on the machine is exhausted squid gets killed. I'm running with
dlmalloc on one of the boxes, but that hasn't seemed to make any
difference. The squids are handling about 35GBish a day in traffic,
averaging around 100-200KB/s and spiking up to 1MB/s for short periods.

Top currently shows this squid at 261m (VIRT) and 191m (RES). Here's the
memory info from cachemgr:

Memory usage for squid via mallinfo():
        Total space in arena: 182212 KB
        Ordinary blocks: 182070 KB 34 blks
        Small blocks: 0 KB 0 blks
        Holding blocks: 34588 KB 9 blks
        Free Small blocks: 0 KB
        Free Ordinary blocks: 141 KB
        Total in use: 216658 KB 100%
        Total free: 141 KB 0%
        Total size: 216800 KB
Memory accounted for:
        Total accounted: 120116 KB 55%
        memPool accounted: 120116 KB 55%
        memPool unaccounted: 96683 KB 45%
        memPoolAlloc calls: 928095129
        memPoolFree calls: 932772909

The wiki[1] says that I should expect squid to use 14MB per GB +
cache_mem + buffers and some other stuff. For my config that looks like
it'd be 1.4MB + 10MB + ?MB => 11MB which is close to the 16MB I see when
it starts.

I guess my question is, how do I control how much RAM squid uses? I can
provide my full config and the full cachemgr output if needed, and any
other information that might help. I also have a 224MB core dump from
one of the machines if it would help track down where the memory is
being used.

Thanks,
--Will

[1] http://wiki.squid-cache.org/SquidFaq/SquidMemory
Received on Sat Jul 02 2011 - 19:23:48 MDT

This archive was generated by hypermail 2.2.0 : Sun Jul 03 2011 - 12:00:01 MDT