[SQU] no space left... but there is!

Ok, now, i put two 9 GB HDDs, and started squid with these new caches.
At some moment, it started to die repeatedly, complaining there is no
space:

FATAL: msgget failed
Squid Cache (Version 2.4.PRE-STABLE): Terminated abnormally.
CPU Usage: 0.061 seconds = 0.019 user + 0.042 sys
Maximum Resident Size: 3712 KB
Page faults with physical i/o: 0
2001/02/21 10:46:00| Starting Squid Cache version 2.4.PRE-STABLE for
mips-sgi-irix6.5...
2001/02/21 10:46:00| Process ID 1864363
2001/02/21 10:46:00| With 2500 file descriptors available
2001/02/21 10:46:00| Performing DNS Tests...
2001/02/21 10:46:00| Successful DNS name lookup tests...
2001/02/21 10:46:00| DNS Socket created on FD 4
2001/02/21 10:46:00| Adding nameserver 192.26.51.29 from squid.conf
2001/02/21 10:46:00| Adding nameserver 198.29.75.111 from squid.conf
2001/02/21 10:46:00| User-Agent logging is disabled.
2001/02/21 10:46:00| Referer logging is disabled.
2001/02/21 10:46:00| Unlinkd pipe opened on FD 9
2001/02/21 10:46:00| Swap maxSize 16384000 KB, estimated 1260307 objects
2001/02/21 10:46:00| Target number of buckets: 63015
2001/02/21 10:46:00| Using 65536 Store buckets
2001/02/21 10:46:00| Max Mem size: 8192 KB
2001/02/21 10:46:00| Max Swap size: 16384000 KB
2001/02/21 10:46:00| storeDiskdInit: msgget: (28) No space left on
device
FATAL: msgget failed
Squid Cache (Version 2.4.PRE-STABLE): Terminated abnormally.
CPU Usage: 0.068 seconds = 0.019 user + 0.048 sys
Maximum Resident Size: 3712 KB
Page faults with physical i/o: 0

But there is space available! See:

# df
Filesystem Type blocks use avail %use Mounted on
/dev/root xfs 15670928 2341272 13329656 15 /
/dev/dsk/dks0d3s7 xfs 17768080 9460176 8307904 54 /usr/people
/dev/dsk/dks0d4s7 xfs 17768080 2984 17765096 1
/usr/local/squid/cache1
/dev/dsk/dks0d5s7 xfs 17768080 2984 17765096 1
/usr/local/squid/cache2
#

Only 1% use!

It seems that squid doesn't respond to -k shutdown. I can only terminate
it with "killall squid diskd". It seems that when i try to kill it with
-k, some new diskd processes are respawned, the old diskd processes
doesn't die, and the squid main process is just restarted.

It also looks like all these problems (because of them i'm flooding you
guys with messages) are showing up more often when squid can use a lot
of bandwidth, when the accessed site has a lot of bandwidth, and when
the accessed pages contains lots of objects (images, frames...). I'm not
sure on this, though, but i used 2.4-pre a lot on a SLOW link, and it
was fine.
I didn't tried to put more users on this proxy (and see what happens
with lots of requests), because of its weird behaviour.

squid-2.4-200102200000 on Irix-6.5 (but i'll try the newest CVS build
soon, and maybe with some other compile options, like less optimisations
and so on...)

P.S.: Am i distressing someone with these multiple e-mails? Should i
continue to feed you with my further observations on squid?
I'm sending you all these things because i have to start the "official"
test phase with the new proxy on Friday, and hopefully put it in
production sometimes next week. If during the next few days i'll be
unable to eliminate these bugs, i'll have to revert to the old 2.3
series (but i'll try to avoid that if i can).
Thank you for your patience and support.

-- 
Florin Andrei
--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html