[squid-users] Re: lockup problem

Stuart Clark wrote:

> Celeron 2.4, 2 gig ram, redhat 9, squid-2.5.STABLE1-3.9

I would suggest updating to a newer 2.5 STABLE release, though it might not
solve this particular problem.

> My squid server is locking up and needs rebooting every couple of days

> I know why. Its because I ran out of ram for the size of the cache (proven
> with testing).

> I have 2 gig ram and 105gig of cache, 7.3M objects (kinda blows the 1
> gig/100gig ratio outa the water)

That ratio is a rule of thumb - the amount of cache metadata in memory is
actually determined by the number of objects in your cache.

http://www.squid-cache.org/Doc/FAQ/FAQ-8.html#ss8.1

Besides metadata, there's also memory used for caching hot objects
(cache_mem setting), as well as memory used by other programs and the
operating system itself.

It could also be bad memory causing the problem - you can check this using
memtest86+.

> Is their any way of telling that the cache size is getting near the limit
> of the physical memory rather than just believing the 1gigram/100gig cache
> rule (which dosen't work anyway)?

Monitor your memory usage as the cache fills and see at what level the
lockup occurs.

> When it does lockup because of ram limitation what percentage should I
> reduce the cache? I tried reducing the cache by 3 gig and it locked up
> again.

Probably 10 - 15 percent would be a good number.

> I have every squid mrtg graph known to man and I cannot see any
> indications to anticipate a lockup.

Then either your aren't monitoring the right data, or your assumption of the
cause of the lockups is incorrect.

Adam
Received on Tue Dec 21 2004 - 19:22:40 MST