Re: Strange crashes on Tru-64 Unix

Meseemeth to me that this is a filesystem fragmentation issue. Sun's UFS
behaves like this, IIRC. You get to about 75 or so percent full and
there's no fragment left large enough to store your file, so you get a
write-fail.

Squid _should_ continue operating after this, though. It should treat
the amount of diskspace at which an error occurs to be the maximum, and
then throw away files to make room for the new object. If this is indeed
a fragmentation issue then it may take a few rounds of this before the
filesystem ultimately believes that it has sufficient space again.

Change filesystem types? Or reduce your configured disk-space to around
60% or 70% of the total?

D

Mark Borrie wrote:
>
> We recently upgraded our Compaq server to Tru-64 Unix 4.0f (This was
> formerly Digital Unix). We had upgraded to squid 2.2 Stable 4 prior to this
> and it seemed to be working OK on version 4.0b.
>
> After the upgrade we struck 2 problems. One was with the default route off
> campus being reconfigured by malformed packets. We have applied a patch from
> Compaq for this.
>
> The second problem we still do not have a solution. After running for a week
> or so squid dies claiming that it is out of disk space even though the disk
> is only 75% full. I have even tried reducing the cache size significantly
> but we have started getting these crashes again.
>
> After a crash we have to usually delete the swap.state file, or use a
> completely new cache directory to get going again. Deleting the swap.state
> file doesn't seem like a good thing as I did see the cache size get far too
> big once after doing that. There seemed to be files in the cache that
> weren't getting removed.
>
> My feeling at the moment is that we have a problem with advfs rather than
> squid. Has anyone else experienced anything similar? Any ideas?
>
> My set up is
> Squid 2.2 Stable 4
> Tru-64 Unix 4.0f
> Cache size 6Gb on a 8Gb advfs volume
>
> I have recompiled squid on the new OS (and we rebuilt gcc as well).
>
> Error messages from the cache.log file....
> Squid Cache (Version 2.2.STABLE4): Terminated abnormally.
> 1999/08/25 12:22:58| Starting Squid Cache version 2.2.STABLE4 for
> alpha-dec-osf4 .0...
> 1999/08/25 12:22:58| Process ID 2133
> 1999/08/25 12:22:58| With 4096 file descriptors available
> 1999/08/25 12:22:58| Performing DNS Tests...
> 1999/08/25 12:22:58| Successful DNS name lookup tests...
> 1999/08/25 12:22:58| helperOpenServers: Starting 10 'dnsserver'
> processes
> 1999/08/25 12:22:59| Unlinkd pipe opened on FD 13
> 1999/08/25 12:22:59| Swap maxSize 5120000 KB, estimated 393846 objects
> 1999/08/25 12:22:59| Target number of buckets: 7876
> 1999/08/25 12:22:59| Using 8192 Store buckets, replacement runs every 10
> seconds
> 1999/08/25 12:22:59| Max Mem size: 69632 KB
> 1999/08/25 12:22:59| Max Swap size: 5120000 KB
> 1999/08/25 12:22:59| Store logging disabled
> 1999/08/25 12:22:59| Rebuilding storage in Cache Dir #0 (DIRTY)
> 1999/08/25 12:22:59| Set Current Directory to /www/cache
> 1999/08/25 12:22:59| Loaded Icons.
> 1999/08/25 12:22:59| Accepting HTTP connections on port 81, FD 36.
> 1999/08/25 12:22:59| Accepting ICP messages on port 3131, FD 37.
> 1999/08/25 12:22:59| Ready to serve requests.
> 1999/08/25 12:22:59| diskHandleWrite: FD 12: disk write error: (28) No space
> lef t on device
>
> 1999/08/25 12:22:59| Rebuilding storage in Cache Dir #0 (DIRTY)
> 1999/08/25 12:22:59| Set Current Directory to /www/cache
> 1999/08/25 12:22:59| Loaded Icons.
> 1999/08/25 12:22:59| Accepting HTTP connections on port 81, FD 36.
> 1999/08/25 12:22:59| Accepting ICP messages on port 3131, FD 37.
> 1999/08/25 12:22:59| Ready to serve requests.
> 1999/08/25 12:22:59| diskHandleWrite: FD 12: disk write error: (28) No
> space lef t on device FATAL: Write failure -- check your disk space and
> cache.log Squid Cache (Version 2.2.STABLE4): Terminated abnormally.
>
> Thanks, Mark.
Received on Thu Aug 26 1999 - 16:57:57 MDT