Dear Chris,
This sounds like a good one.
As part of the logging, I need to log the current number of file
descriptors in use and if after a crash I notice the last values hitting
the peaks, I can almost accurately say that REDIRECT of iptables is the
cause. Any pointers to monitor the file descriptors?
From what your last sentence says, you are not using redirect but ask
customers to setup the proxy settings, probably using .pac files.
Thanks,
Mitesh
Chris Robertson wrote:
Mitesh P Choksi wrote:
Dear Tim/Neil,
I will look up the details and send it over to the group. The
messages and the cache.log looks normal. The access.log is not set,
i.e. /dev/null. I also realised that as soon as number of users are
high, i.e. when I start redirecting (-j REDIRECT) more users to
Squid, it starts crashing, but if I keep 2-3 IPs then it's not a
problem.
It does look like an OS related issue from the details from Neil,
however, if I don't use squid i.e. no -J REDIRECT then the server
keeps running for months, but as soon as I turn on the REDIRECT, it
starts crashing almost every week.
Maybe I keep a strace on squid and dump the data and also keep the
vmstat 5 dumping data so that I can co-relate and come up to some
conclusion.
I wanted ideas on what to trace so that the root cause is
identifieid. Maybe it's a hardware problem but due to the fact that
it only shows up after I turn on squid, its is difficult to justify
the need for different hardware.
Thanks in advance.
Regards,
Mitesh
Squid is not hooked into the kernel. IPTables is. I would posit that
to be the source of your trouble. You stated that you are on a
satellite link, which means your are going to have a higher number of
open file descriptors. Perhaps IPTables is "running out" and causing
the kernel to hang (complete conjecture).
For what it's worth, I have a number of sites on satellite links, but
I don't use interception.
Chris
