Re: Squid seems to "HANG UP"! (NEW FINDINGS)

On Thu, 16 Sep 1999, Sarah Rodriguez wrote:

> I made some discoveries a while ago. When I experienced the "hang up"
> around 30 minutes ago ... I tried to modify the proxy settings of one of
> the clients from 172.16.12.254:8085 to 202.8.254.3:8085 without touching the
> physical network. This client was able to connect to the proxy. Also I
> have observed that during these moments of "hang up" .. I can ping eth1

> (172.16.12.254) but could not telnet to it ... but i can telnet eth0 from
> any client. i have to wait sometime before any client can access the proxy
> again.
>
> Here is my question ... if I am running out of file descriptors .... why was
> I be able to access the same proxy using another interface? .. if my problem
> is TCP ... doing a netstat -a | grep 8085 lists only a few tcp items:

Is there any messages from the kernel in the syslog? (or anything untoward
in the output of a dmesg?) My guess is that the ethernet card is hanging
and is waiting to be reset by the kernel.

Look for messages that look similar to:

eth0: Transmit timed out: status 0050 0090 at 909508821/909508836 command 000c0000.
eth0: Trying to restart the transmitter...

Occasionally if a machine is heavily loaded, and using certain ethernet
cards, specifically those supporting early recieve/transmit, then packets
can start getting sent out from the ethernet card onto the PCI bus before
the whole packet's arrived. If the disk subsystem (or something else)
wakes up at this instant, and also attempts to send something out at that
point, the ethernet card no longer has the whole packet and can hang,
causing this symptom. (this is only a problem with a very few ethernet
drivers, and only when the whole IO system is under heavy load)

So it _could_ be something like this, but then again may not be. A linux
kernel hacker may be able to help you more with this - since it may be a
problem with the 2.2.6 kernel insteadl. Try looking in the
changelogs/release note for later kernel versions and looking for bug
fixes!

Michael.

--
National & Local Web Cache Support        R: G117
Manchester Computing                      T: 0161 275 7195
University of Manchester                  F: 0161 275 6040
Manchester UK M13 9PL                     M: Michael.Sparks@wwwcache.ja.net