Re: 2.3STABLE - stability issues

On Mon, 4 Sep 2000, Andres Kroonmaa wrote:

>
> I lately ran into problem. Squid swapdisk got nearly full,
> to the extent that Squid didn't have enough temporary space
> to write swap.state.new (or swap.state.clean) and it bailed
> out with fatal error. Squid is unable to recover on its own,
> although this isn't very fatal error.
> Manual recovery is quite dirty and implies removing contents
> of several L2 dirs by hand.
>
> Wouldn't it be appropriate to delay writeout of clean swaplog
> in case swapdir is too close to disk full?

I think what you are asking for implies that you delete the old
swap.log, then write the new one. Your assumption being that
removing the old log frees up enough disk space for writing the
new one?

I think its a bad idea if there is a point in time where NO
swap state file exists on the disk. If squid crashes before
writing the new log completes, you have a partial record, or
perhaps no record at all.

Maybe a better idea is for Squid to refuse to start unless
there is a certain amount of free disk space.

Or it disabled the cache_dir (for reading and writing) until some
space is freed up.

Duane W.
Received on Wed Sep 06 2000 - 21:55:24 MDT