Re: bogus Directory Store Stats?

David Luyer wrote:

> I think Henrik's point is something like, when you get some types
> of SCSI errors, that disk from that time on will still work for
> reads but will return ENOSPC for all writes (even when the error
> was a simple timeout with failed retry on some systems!). So
> every time squid decides to try and write a file on the failed
> device, it reduces the cache size and decides to unlink various
> files, and eventually removes all files including those on the
> working drives.

You got the point 100%.

Or that the OS remounts the disk read-only since it thins read-only is
better than writing on a disk that behaves strangely.

Another similar situation is when the disk got filled by a huge core
dump, or something else external to Squid, which has a similar effect to
Squid.

A third situation is when one misconfigures the size of one of the cache
directories much larger than it really is.

When Squid fails to write to one cache_dir then the total store size is
trimmed down by approximately "lost space" * "number of cache dirs"
after some activity. If a whole cache dir is lost (or a large percentage
of one) then it has a servere impact on the total cache size, even if
the cache dir is only a fraction of the total cache.

/Henrik
Received on Tue Jul 29 2003 - 13:15:52 MDT