--On May 4, 2006 1:29:00 PM -0300 tpeixoto@xxxxxxxxxxxxxxx wrote:
Hello! Squid is dying with no apparent reason from time to time... I only see in cache.log: FATAL: Received Segment Violation...dying.
Segfaults are usually one of about three things (most common to least common) -- bad programming, bad memory, or bad CPU fan. Are you using async I/O? In the past I've had lots of issues with that. Outside of that what follows are more or less general recommendations...
I don't run 2.5.13 on FreeBSD 5.4 nor AMD64 so I can't vouch for it's stability there however on Linux and FreeBSD 4.x it's been very stable. If you can take the machine down for a few hours download memtest86 from Memtest86.com burn the ISO to a disc, and boot off of it. Let it run a pass or two. If it comes up with any errors, there's your problem, bad memory most likely. If it locks up, reboots, or shuts down, you've got a cooling issue.
Assuming that's all clear then my next step would be to turn on core dumps and after it dies use gdb to gether the backtrace information. If you haven't built with debugging symbols and automatic backtraces by adding --enable-stacktraces to your existing ./configure...while you're at that can you post what your existing ./configure is.
Note that I know atleast int eh past AIO/aufs (i think is what it's called now) has been the cause of much grief for me so I use diskd exclusively. I'm not a huge fan of big monolithic programs anyway.
... and them it restarts automatically. I'm running squid-2.5.STABLE13 in FreeBSD 5.4-RELEASE (amd64). CPU usage is low. It's difficult to see Squid getting more than 30% of CPU usage. I have 2GB of RAM, Squid eats 1.2GB and the system always has at least 200MB free, except swap. Cache dir is about 100GB. Anyone has a similar issue? Is there any way I can debug this problem? Thank you in advance.
-- "Genius might be described as a supreme capacity for getting its possessors into trouble of all kinds." -- Samuel Butler
