Re: [squid-users] Squid 3.2, multiple workers, SNMP (and a bit of IPv6)

From: Amos Jeffries <squid3_at_treenet.co.nz>
Date: Tue, 02 Apr 2013 21:30:46 +1300

On 2/04/2013 6:38 p.m., Eugene M. Zheganin wrote:
> Hi.
>
> On 18.01.2013 06:42, Panagiotis Christias wrote:
>
>> In both cases, every SNMP adds two or new stale entries (filedescriptors,
>> sockets or whatever) as reported by lsof:
>>
>> # lsof -i -nP | egrep 'PID|:163$'
>> COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
>> squid 5007 squid 15u IPv4 0xfffffe0006b14790 0t0 UDP 127.0.0.1:163
>> squid 5008 squid 15u IPv4 0xfffffe0006b14790 0t0 UDP 127.0.0.1:163
>> squid 5008 squid 19u IPv4 0xfffffe0006b14790 0t0 UDP 127.0.0.1:163
>> squid 5008 squid 41u IPv4 0xfffffe0006b14790 0t0 UDP 127.0.0.1:163
>> (and the list gets longer and longer as our monitoring systems keep
>> quering squid..).
>>
>> Everything (SNMP-related) works correctly when we use just one worker
>> but currently this is no option since a single squid 3.2 worker seems
>> to be unable to handle as many requests as squid 2.7 (in our case at
>> least).
>>
>> Any feedback would be greatly appreciated.
>>
> Yeah, recently I discovered that I have exactly same issue with 3.2.9.
> Furthermore, I can say that I even have more workers than intended:
>
> 22972 ?? S 0:02.92 (squid-2) -f /usr/local/etc/squid/squid.conf
> (squid)
> 23146 ?? S 0:02.44 (squid-1) -f /usr/local/etc/squid/squid.conf
> (squid)
> 23237 ?? S 0:00.74 (squid-coord-3) -f
> /usr/local/etc/squid/squid.conf (squid)
> 31139 ?? R 0:00.11 (squid-1) -f /usr/local/etc/squid/squid.conf
> (squid)
> 31192 ?? R 0:00.01 (squid-1) -f /usr/local/etc/squid/squid.conf
> (squid)
> 31193 ?? R 0:00.00 (squid-1) -f /usr/local/etc/squid/squid.conf
> (squid)
>
> Some of they don't die after squid -k kill, I have to kill them
> explicitely with -9.
> Is that normal ? Right now I decided to switch back to non-SMP.
> And yeah, I just don't like this situation with thousands of open FDs,
> wich doesn't happen with one worker.
>
> Is this resolved in 3.3.x ?
>
> Thanks.
> Eugene.

No, unfortuately nobody is working on that part yet.

As a workaround you should be able to retrieve SNMP information
per-worker by using ${process_number} in the snmp_port directive to
assign each worker a unique port for SNMP contact.

Amos
Received on Tue Apr 02 2013 - 08:30:51 MDT

This archive was generated by hypermail 2.2.0 : Tue Apr 02 2013 - 12:00:04 MDT