[squid-users] Squid 3.5.7, cache_swap_high, bug or not bug ?

[Amos Jeffries]

On 21/08/2015 8:46 p.m., Stakres wrote:
> Amos,
> 
> in the meantime, I was thinking another point:
> We know there are at least 2 limit with cache_dirs, the max size and the 16+
> million entries (Filemap bits).
> the cache_swap_high should take care both.
> example: if the used cache is 95% of the space or if the Filemap bits is 95%
> of the 16+ million entries.
> 
> Because if the used cache is 100%, the squid could crash a few later, but if
> the squid is 16777216 entries it does crash immetiatly and will be looping
> in crashes.
> 
> See what I mean ?
> 

Yes. Good point.

Though actually reaching that sfileno limit of 2^25 objects is not the
problem in the crash issue. Just like memory and rock stores happily sit
at 100% full in capacity with no need to purge. So can any other cache
happily sit with 2^25 objects stored.

The crash part is in whatever specific bit of code is allowing stores to
go out-of-bounds on the sfilno value range. Anything we do here would be
a workaround at most, and at the cost of cache HITS for anyone who runs
with full caches but not crashing with that bug.

Amos