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

Hello everybody,

we are currently migrating our squid 2.7 installation to 3.2 and
facing some SNMP related problems.

We are using squid 3.2.6 on FreeBSD 9.1-amd64 and our system is
fully IPv6 enabled. Using the following configuration:

workers 2
snmp_port 163
snmp_access allow all

Cache.log reports:

2013/01/18 01:38:21 kid2| Sending SNMP messages from [::]:163
2013/01/18 01:38:21 kid1| Sending SNMP messages from [::]:163
2013/01/18 01:38:21 kid1| Accepting SNMP messages on [::]:163
2013/01/18 01:38:25 kid2| Accepting SNMP messages on [::]:163

Every SNMP query fails:

% snmpwalk -v2c -c public 127.0.0.1:163 1.3.6.1.4.1.3495
Timeout: No Response from 127.0.0.1:163

And cache.log reports:

2013/01/18 01:38:46 kid2| snmpHandleUdp: FD 15 recvfrom: (35) Resource temporarily unavailable
2013/01/18 01:38:47 kid1| snmpHandleUdp: FD 15 recvfrom: (35) Resource temporarily unavailable

(note that trying to query 'UDP6:[::1]':163 would result squid worker
crashes and restarts, but let's leave that out for now)

Changing the configuration to:

workers 2
snmp_port 163
snmp_incoming_address 127.0.0.1
snmp_access allow all

Would result successful SNMP queries:

% snmpget -v2c -c public 127.0.0.1:163 1.3.6.1.4.1.3495.1.1.1.0
SNMPv2-SMI::enterprises.3495.1.1.1.0 = INTEGER: 460

And cache.log would report just one line per SNMP query:

2013/01/18 02:20:26 kid1| snmpHandleUdp: FD 15 recvfrom: (35) Resource temporarily unavailable

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.

Thanks,
Panagiotis

-- 
Panagiotis J. Christias    Network Management Center
P.Christias_at_noc.ntua.gr    National Technical Univ. of Athens, GREECE