Re: [squid-users] Squid observes cache_mem for days then goes crazy

Further to this you can see the MRTG graphs of Storage Mem at:

http://atlas.salford.ac.uk/stats/proxy/dione/cachesysvmsize.html

The other cache started last night at 9pm, a period of low usage. You can
see that here:

http://atlas.salford.ac.uk/stats/proxy/helene/cachesysvmsize.html

Any ideas what would cause this strange behaviour?
  Cheers.

On Mon, 10 Dec 2001, Mark Powell wrote:
> Hi,
> I upgraded our two site caches from FreeBSD 3.5 and squid 2.3-STABLE4 to
> FreeBSD 4.4-STABLE and squid 2.4-STABLE3 using diskd, about 2 weeks ago.
> They'd been running fine for about 2 years with the FreeBSD and the squid
> being periodically brought up to date.
> Config file options seem okay, but squid seems to run happily for a
> while then suddenly start consuming lots of memory. Both machines have 1GB
> RAM and the non-default config options for storage are:
>
> cache_mem 256 MB
> maximum_object_size 256 MB
> cache_dir diskd /cache1 7300 16 256
> cache_dir diskd /cache2 7300 16 256
> cache_dir diskd /cache3 7300 16 256
> cache_dir diskd /cache4 7300 16 256
>
> One machine ran for 10 days seemingly without a hitch. In that
> time the process gradually got larger and larger as it's cache filled up.
> It slowly crept up to the following sizes:
>
> /dev/da1s1a 8822066 6827371 1288930 84% /cache1
> /dev/da2s1a 8822066 6828831 1287470 84% /cache2
> /dev/da3s1a 8822066 6826014 1290287 84% /cache3
> /dev/da4s1a 8822066 6828091 1288210 84% /cache4
>
> It's been at this amount of disk usage for a while now so I think it's
> filled it's cache.
> Squid used 256MB of memory as I told it and hung around that mark:
>
> $ client mgr:info | grep size
> Storage Swap size: 25577703 KB
> Storage Mem size: 262136 KB
>
> However, now after running for a day or so at the 256MB mark it will
> suddenly start to grow without limit. i.e. it started again about 30 mins
> ago and is now at:
>
> $ client mgr:info | grep size
> Storage Swap size: 26910658 KB
> Storage Mem size: 579148 KB
>
> then a few minutes later:
>
> $ client mgr:info | grep size
> Storage Swap size: 26906399 KB
> Storage Mem size: 600240 KB
>
> It does this until it hits the resource limit and dies. During this
> accelerated growth there are many, many client timeouts and cache response
> is very poor.
> The other cache was upgraded second and has still to reach the above
> disk usage. Maybe that will exhibit the same problem when it's cache has
> been filled for a while.
> I just shutdown the cache before it consumed all memory and died. I saw
> the error:
>
> assertion failed: stmem.c:85: "current_offset == target_offset"
>
> Instead of the usual 30 secs wait it took 5 mins before it started writing
> out swap.state entries. Until it finally finished with:
>
> CPU Usage: 13914.695 seconds = 9469.761 user + 4444.934 sys
> Maximum Resident Size: 865888 KB
> Page faults with physical i/o: 26774
> 2001/12/10 15:28:23| Open FD 17 squid -> diskd
> 2001/12/10 15:28:23| Open FD 20 squid -> diskd
> 2001/12/10 15:28:23| Open FD 23 squid -> diskd
> 2001/12/10 15:28:23| Open FD 26 squid -> diskd
> 2001/12/10 15:28:23| Squid Cache (Version 2.4.STABLE3): Exiting normally.
>
> I've checked the FAQ, etc. and my config, but can't see what's wrong.
> It's beginning to seem like a bug to me.
> Any ideas what is causing this?
> Cheers.
>
> Mark Powell - UNIX System Administrator - The University of Salford
> Academic Information Services, Clifford Whitworth Building,
> Salford University, Manchester, M5 4WT, UK.
> Tel: +44 161 295 5936 Fax: +44 161 295 5888 www.pgp.com for PGP key
>
>
>

Mark Powell - UNIX System Administrator - The University of Salford
Academic Information Services, Clifford Whitworth Building,
Salford University, Manchester, M5 4WT, UK.
Tel: +44 161 295 5936 Fax: +44 161 295 5888 www.pgp.com for PGP key
Received on Tue Dec 11 2001 - 03:43:25 MST