Memory problems .... ?

Config: SPARCStation 5, 192Mb Ram, 5Gb cache disk, Solaris 2.6 running
        Squid-1.NOVM.22 with Arjan's proxy auth patch and dl-malloc.

I noticed something very strange and extremely alarming today, which caused
squid to stop responding to requests and needed a dirty restart (kill only
produced the "Waiting 10 seconds for active connections to finish" message
and didn't proceed any further).

The process was at around 70Mb resident, according to dl-malloc the actual
amount of that in use by squid was around 60Mb. I wanted to check a stat
from the IP cache so I issued the request to squid via the cachemgr.

At this point, squid stopped talking (not refusing, just not responding to
requests). After a few seconds (I was starting to worry!) I traced the squid
process and it was calling brk() like mad.

It did this for a total of nearly 4 minutes, at which point entries in the
IP cache started getting marked as "bad" and it was still not responding to
requests.

Lots of ERR_CONNECT_FAIL messages appeared in the cache log for sites I
briefly tested and verified as being accessible. Very alarming was the fact
that from ~70Mb the process size had risen to ~289Mb, of which around 150Mb
was marked as resident (no idea on what mallinfo() thought it was doing as I
couldn't get any cachemgr output from the proccess).

I've got nothing else to work with -- couldn't get a core dump out of it.
Has anyone else ever seen anything like this happening?

I poll squid with the cache manager every five minutes. The prospect that
this can cause problems like this is scary, to put it mildly ...

Cheers,

Chris

ps:

ipcache_size 10240
ipcache_low 90
ipcache_high 95

-- 
Chris Tilbury, UNIX Systems Administrator, IT Services, University of Warwick
EMAIL: cudch+s@csv.warwick.ac.uk PHONE: +44 1203 523365(V)/+44 1203 523267(F)
                            URL: http://www.warwick.ac.uk/staff/Chris.Tilbury