Re: Bad behavior when cache_dir is full

On Thu, 18 Feb 1999, Pål Baltzersen wrote:

> Disk fragmentation problem is a known issue on Solaris2.6
> but couldn't Squid handle this more robust than:
>
> 1999/02/18 10:08:14| diskHandleWrite: FD 30: disk write error: (28) No space left on device
> FATAL: Write failure -- check your disk space and cache.log
> Squid Cache (Version 2.1.PATCH2): Terminated abnormally.
>
> I have lowered the Mbytes part of the cache_dir config several times
> with better and better stability, but still this happens once a while.
> Couldn't Squid be smarter and lower this parameter dynamically
> and/or do some tougher removal of of cached objects and then continue?
>
>
> Pål

At least one version of Squid 2 does do this. It reduces the available
swap dynamically until there is none left, since the LRU algorithm seems
to take some time to catch up. :(

My preference would be for the cache to go into proxy-only mode when the
disc fills up, and issue a warning in the cache log. Auto-tuning doesn't
seem to be effective at present.

John
Received on Thu Feb 18 1999 - 06:48:29 MST