Re: [squid-users] Resource temporarily unavailable tests

Brian wrote:
> $ egrep -i "storeAufsOpenDone" cache.log | wc -l
> 408
If only 12 people are using this, perhaps 408 is a lot - I don't get that
particular error so wouldn't know.

> Squid Cache: Version 2.5.STABLE1
> configure
options: --prefix=/usr/local --disable-dependency-tracking --enable-async-i
o=32 --with-aio --enable-snmp --enable-ssl --with-openssl=/usr/local/ssl --e
nable-poll

Well I assume you know the chorus of: upgrade to a newer version - yours is
not so out of date but still a lot has been fixed. Check what was fixed
http://www.squid-cache.org/Versions/v2/2.5/squid-2.5.STABLE2-RELEASENOTES.ht
ml and here for bug info:
http://www.squid-cache.org/Versions/v2/2.5/bugs/#STABLE1 - or just do the
easy/lazy thing and upgrade to 2.5STABLE3 before further testing.

Also: why did you set async-io to 32? I just used "--enable-async-io" and
let squid set it's default. Maybe your way is better but just so you know
why you did that and that it's not at all related to your aufs problem....
Also I thought you had to include aufs as one of the compile options? I did
(--enable-storeio=aufs,ufs). Maybe I didn't have to since I assume your
squid would squawk if it weren't compiled in (does "squid -k parse" run
cleanly?). Unrelated but I'd add --enable-removal-policies=heap,lru so you
can later try the heap algorithms (we use GDSF). Lastly, why are you
enabling ssl? If you think you need to enable ssl to pass https/443 traffic
that is incorrect. From reading Henrik and other's posts, it looks like
lots of people compile it in when that wasn't what they really wanted (see
the archives for better explanations). If you don't need it, I'd suggest
leaving it out.

> cache_dir aufs /squid/s00 1781188 16 256
> cache_dir aufs /squid/s01 1593568 16 256

I'm pretty sure Henrik confirmed for me that the cache_dir sizing parameter
is in MB's. It looks like earlier you posted the output from a "df -k" and
if so, you've got 2GB partitions. Above you've defined cache_dirs of
1781GB (gigabytes). I just checked my conf default and in fact the value is
MB's. So drop that down to: cache_dir aufs /squid/s00 1781 16 256

For such small cache_dir's, your RAM vis-a-vis sizing is fine but as you
make a larger cache, be sure to use the FAQ to calculate the memory
correctly so you don't make your cache_dir's larger than RAM+squid's other
RAM needs will require.

Since you original problem "resource temporarily unavailable" sounds like a
resource contention issue, I would try to get squid it's OWN disk, not just
partition. aufs is great (I use it as well) but my uninformed opinion would
be to give it all the I/O you can - check the archives I am pretty sure they
would confirm that. The fact that you don't see this error often in the
archives would imply that it is probably your configuration. The fact that
you've done two things squid recommends you not do (have a huge/incorrecly
defined cach_dir allotment and put them on non-dedicated disks) my guess
would be that the error stems from one or both of those and once you've
fixed both you should be ok.

hth

Adam
Received on Thu Aug 07 2003 - 11:07:34 MDT