Re: aufs 'mistake'

On Thursday 12 December 2002 03.25, Robert Collins wrote:
> Another of my too-busy-to-fix-now notes.
>
> Aufs starts <n>*cache_dir threads. It should be <n>*<aufs
> cache_dir> threads. This
>
> I've 48 threads here, for a single aufs dir :}.

Trivial to fix in Squid-2.5, but not as easy in Squid-2 or Squid-3 as
there is no count of how many aufs dirs there is.

How to best fix this in your unify_io branch I do not know, but I
assume there is some kind of per store-dir registration function
giving the store-dir access to the i/o layer? Keeping a counter there
of how many instanses of the async-io store layer and use this for
thread and load schredding should do the trick I think.

Regards
Henrik
Received on Thu Dec 12 2002 - 01:42:11 MST