Re: [squid-users] cache_peer and tcp connection failed

Ramon Moreno wrote:
> Hi Squid Gurus,
>
> Had a quick question on tcp failures with cache_peers and squid.
>
> 2008/08/04 13:37:48| TCP connection to 10.1.1.2/80 failed
> 2008/08/04 13:37:48| Detected REVIVED Parent: 10.1.1.2
>
> Here is my scenario:
>
> I am using squid as a reverse proxy. If my origin misbehaves (response
> time starts to slow), squid creates a backlog, and cpu spikes to 100%.
> I get the errors above. This every once in a great while, and I
> usually have to restart squid to recover.
>
> Questions I have are:
>
> When tcp connections fail to cache peer does squid mark it down for x
> amount of time before it retries. If there is a default retry
> interval, what is that, and can that be adjusted?
>
> What I am trying to avoid is cache_peer being marked down by squid for
> 5 seconds, and then resulting in a greater backlog of connections.
>
> My timeout for server side connections is 10 seconds.
>
> I also use the following options as I heard this will help mark things
> up a little faster.
>
> monitorinterval=60 monitorurl=http://mysite/hi.jpg monitortimeout=180
> (does changing monitorinterval in 1 second in this scenario help?)
>
> The main goal is to try to get squid to behave as normal as possible,
> and recover as fast as possible in the event of a connection backlog.
> Any help or suggestions are much appreciated.
>
> Thanks in advance.

This is a known bug in the round-robin peer handling.
It's been fixed in the most recent releases of Squid 3.0.STABLE8+ and
2.7.STABLE3+

Amos

-- 
Please use Squid 2.7.STABLE3 or 3.0.STABLE8