Re: [squid-users] squid dies with signal 6

Ron Vachiyer wrote:
>> I had some problems with exits like this, it was due to bad
>> hardware - a SCSI disk which was dying. However, this may not be
>> the case for you.
>>
>> The reason you have had few replies is because it is genuinely difficult
>> to pin down what this might be. Have you looked at memory usage
>> of the cache before it dies, numbers of requests, cache.log,
>> access.log... the processes running... and of course, you haven't
>> told us what your setup is like: aufs, diskd, ufs? how much
>> cache_mem and disk space for cache_dir, how much physical memory
>> you have, etc?
>
>
> Hello,
>
> I was previously on RH6.2 Linux (with the kernel.org source tree,
> not the redhat one) with 1GB RAM, 3 18G 10K SCSI disks on 3 10GB
> cache partitions, aufs, running Squid2.4STABLE4, on a PIII-800. I
> am also running wccp, 80Mb cache_mem.
>
> I have since rebuilt the machine on a RH7.2 P4-1.8Ghz, 2GB RAM and
> it did not die on signal 6 for 6 days. The cache dirs filled up
> and about 48 hours after being at 97% usage, I got three signal 6
> aborts in about 5 minutes, still with no cache.log entries as to
> why. I was hoping my GLIBC was to blame, however I suppose it is
> not since these crashes have continued on the new distro.
>
> I am still using the ip_wccp.o module on kernel 2.4.9-31 (redhat
proprietary
> source tree); could this be a potential reason? Other than the
> fact it is a "unsigned module", the kernel does not complain about it
> and it compiles and inserts correctly.

ip_wccp can not cause Squid to crash (if by some strange quirk it did,
then that would be an indicator of a bug in Squid's request handling
and would be a potential DoS). The WCCP decapsulation module has no
direct connection with the Squid process--it merely converts the
incoming packet into a standard IP packet.

This is a rather strange crash. What kind of request rate and bandwidth
is this box serving? You might just be overloading it (though Squid
should be more forthcoming about why it is keeling over). Usually the
problem of overload becomes apparent some time before a crash, though,
and errors also usually result (could not allocate memory is a common
one for an overloaded Squid--even if plenty of memory is available, I
guess fragmentation of the available memory leads to an inability for
the OS to free up enough contiguous memory to answer Squids allocation).

I haven't seen any random sig 6 crashes in my boxes, but we've only
recently upgraded everyone to the Squid 2.4 series from
2.2STABLE5+hno...maybe I just haven't begun to hear about them yet.

-- 
Joe Cooper <joe@swelltech.com>
http://www.swelltech.com
Web Caching Appliances and Support