Re: Squid going unresponsive

mjw@nielsenmedia.com wrote:

> I've been running Squid 2.2stable4 for a while on a Sun Ultra 5 with Solaris
> 2.6 for a while without any major problems. I had been running with both
> the cache and log files on the same physical disk. 2 days ago, I installed
> some new disks and pointed the log files there. The next day I had to
> restart Squid 5 times. Everything would work fine for a while and then
> response time would go through the roof. cachemgr would show dns lookups,
> user authentication, and download times increasing exponentially.
> Eventually cachemgr would become unresponsive, and then I would get warnings
> in cache.log that Squid was out of file descriptors. I am assuming that
> adding the new disk for log files eliminated disk contention as the
> bottleneck and that I am now hitting the wall elsewhere. The question I
> have is where? It doesn't look to me like I am having memory problems, but
> I might add more just to be sure. What happens when Squid is CPU bound?
> Does this behavior sound familiar to anyone?
>
> Thanks
> Mike

This sounds like a cycle-starvation problem of the sort that I had recently.
Very common if: you have authenticators and/or redirectors that are primarily
cpu-bound, and the incoming request-rate exceeds the handling capacity for a
time.

Of course, similar symptoms can be achieved with an up- or down-stream network
fault (20% packet-loss being a good way to simulate it)

D
Received on Tue Nov 02 1999 - 21:28:53 MST