Re: [squid-users] Re: should squid crash when ulimit reached?

On Friday 12 August 2005 08:50, Henrik Nordstrom wrote:

> On Thu, 11 Aug 2005, joaoBR wrote:
> > Yes I know that FAQ
> > the question is that squid CRASHES when reaching the limit
>
> There is not much else it can do than to report the error and
> restart. And is why there is an FAQ entry on it.
>

IMO this is a strange behavior and a big minus for squid.

When an aplication grows to the available memory limit it should swap
out or reduce its memory use but not crash. In squid's case it could
probably done by a FIFO strategy to maintain objects in memory or
increase/decrease dynamically the cache_mem and/or
maximum_object_size_in_memory values depending on what is available?
or fork squid child processes? But this is only an idea I do not know
the code.

When squid crashes it can cause long service downtime depending on the
cache_dir size. Overall the cache_dirs are dirty when squid restarts
after and normally a fsck run is necessary, that at last on FreeBSD
with softupdates. Once crashed it often happens that it crashes in
shorter periods then after (before getting to the mem limit) what
probably is caused by fs-problems what makes it necessary unmount the
volume and fsck it and remount it to get it back on service. This is
often not a task a normal system admin of a small ISP knows how to do
and so he often stays hours with the service down.

So if the latter happens because all resources are out (mem + swap)
then ok, then it is the owner's problem but squid should not crash
before this happens.

Hans

> > BTW, the FreeBSD part of the FAQ could need some overwork
>
> Updates welcome to squid-faq@squid-cache.org
>
> Regards
> Henrik

--