Re: [squid-users] Squid/Network timeout troubleshooting

On Monday 02 September 2002 16.05, Gino LV.Ledesma wrote:

> There's a shell script from another box that monitors other servers
> through a variety of tools, one of them being ping. ping responds
> with reasonable response times (well within the "normal" range),
> but attempting to connect to 3128 results in either timeouts or,
> worse, rejections. There are times when ping packets are dropped,
> though.

Ok. Then the hardware seems to work decently.

But just to be sure, verify that the box and your switch agrees on
half/full duplex. A mismatch can cause quite strange effects some
times on applications trying to use networking..

> Which is why its surprising. I checked the logs recently and squid
> was complaining about running out of file descriptors. This is
> surprising (to me, at least) considering its a none-caching,
> proxy-only box. The logs also complain about the parent proxies
> being "dead" (although they really aren't). I'll check the file
> descriptor problems again tomorrow in the parent proxy boxes, if
> there ever are some.

Hmm.. if there is communication problems between Squid and your parent
(for example caused by a half/full duplex mismatc) and you have
"half_closed_clients on" (the default) then quite easily a lot of
connections to your parents can get piled up until the point where
you are out of filedescriptors.

If you see the filedescriptor usage screen in cachemgr then you should
be able to tell what all those filedescriptors are beeing used for.

> Which is the right way of setting FD limits at runtime? I added
> ulimit -HSn 8192 to /etc/rc.local, /etc/init.d/squid, and
> /etc/profile, but squid "still complains" about running out of FDs.

For only a couple of hundred requests/minute the default of 1024
should be way more than sufficient.

See my "Linux-2.2" page quoted from the FAQ if you want to increase
the limit. This page applies to Linux-2.2 and later.

Regards
Henrik
Received on Tue Sep 03 2002 - 05:28:37 MDT