We have just upgraded pound from a custom-hacked version based on
1.8.2-1sarge1 to a custom-hacked version based on 2.0-1.2 etch (yes,
we're running debian on our servers). After the upgrade, we've
experienced crashes. We've had to "kill -9" and then restart it. It
starts with the error message:
HTTP accept: Too many open files
and thanks to syslog being a bit smart ...
last message repeated 1298067 times
It could of course be that our local hacks have an impact, and my
colleague found that we have forgotten to set the timeout for the
backends ... but I would believe there is a sane default for the
timeout? In any case, pound should not crash? Shouldn't the situation
with too many open file descriptors be handled a bit more gracefully?
We've been running the old version of pound for several years without
any significant problems, never had crashes like this. Now we have
upgraded two servers with low traffic, and both those have been crashing
on us, several times already ... I don't think this is a (D)DoS-attack
or anything like that.
We will investigate further, just wondered if anyone has experienced
similar problems?
|
