[squid-users] startup problems following a crash

From: Robin Stevens <[email protected]>
Date: Wed, 20 Feb 2002 17:54:15 +0000

Squid crashed this morning on one of my servers. catchsegv trapped some
information which I will be reporting separately to the appropriate address.

However, on startup I notice several strange things, which I'm not
convinced are related to the crash on the previous run but on problems upon
restart. I'd like to know whether it's a problem in squid itself or my
particular setup.

The squid version is 2.4STABLE3 with SNMP patch applied last week, running
on a single-processor PIII under Linux (Redhat 7.1). squid is run through
catchsegv via RunCache with the options "-NsY -D -F".

Firstly the machine was taking requests while reading the swaplog entries,
which I thought the -F option prevented.

Secondly, the filedescriptor behaviour seems most unusual. It's compiled
with support for 32768 filedescriptors, far more than are generally needed,
but on this occasion it warned of running out of filedescriptors about 10
minutes after restart and again another 20 minutes later. Looking at the
MRTG logs I see that the "reserved filedescriptors" count has been at 24984
rather than the usual 100 - even so it was on those occasions using over
7000 FDs rather than the usual ~2000.

Thirdly, hit ratios are substantially down on normal. This is something
I've seen at least once before if a cache has been taking requests while
reading the swaplog, but in the event of a planned shutdown it can be
avoided by removing the machine from the pool at our L4 switch. On the
past occasion the system appeared to encounter many SWAPFAIL errors until I
restarted with an empty cache, but this time it's not logging any such
failures.

Fourthly, miss times are many times higher than normal - 1-4 seconds
instead of the usual 300-450ms.

Any advice would be appreciated. So far I've not yet given squid another
restart in case there's any useful information to be extracted from the
currently-running process (fortunately we've got spare capacity at
present).

Thanks,
        Robin

-- 
--------------- Robin Stevens  <robin.stevens@oucs.ox.ac.uk> -----------------
Oxford University Computing Services ----------- Web: http://www.cynic.org.uk/
------- (+44)(0)1865: 273212 (work) 273275 (fax)  Mobile: 07776 235326 -------
Received on Wed Feb 20 2002 - 10:54:17 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:06:28 MST