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

From: Henrik Nordstrom <henrik_at_henriknordstrom.net>
Date: Tue, 05 Aug 2008 11:31:28 +0200

On mån, 2008-08-04 at 16:07 -0700, Ramon Moreno wrote:

> 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?

It runs one background probe at a time while there is traffic, until
connections start succeeding again.

> 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.

Then in worst case there will be a 10 seconds lag before Squid sees the
server up again after a networking failure.

As long as the networking works proper (no packet loss) recovery is
immediate, requiring at most two request (one request initiating the
connection probe but not yet considering the peer alive, and then as
soon as that probe succeeds the peer is declared alive, some
milliseconds for the TCP handshake)

> 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?)

one second probably is too short for reliable result and will quite
likely result in too many false negatives..

Regards
Henrik
Received on Tue Aug 05 2008 - 09:31:32 MDT

This archive was generated by hypermail 2.2.0 : Tue Aug 05 2008 - 12:00:02 MDT