Re: Bug #1893; why's it incomplete?

On ons, 2008-06-11 at 01:18 +0800, Adrian Chadd wrote:
> Henrik,
>
> Whats incomplete about the Variant invalidation support in PURGE/CLR in
> Squid-2.HEAD?

There are many common corner cases where the existing vary_id isn't
picked up properly. The worst being if an object goes from not having
Vary to having Vary (as seen in IIS with gzip compression enabled.. the
plain uncompressed variant often does not have Vary at all..). It then
inherited the non-existing vary_id {0,0}, which caused the cached
objects to never be hit..

Another case is requests with no-cache if the vary index object isn't in
memory.

And I beleive the case of a caching a Vary object when the vary index
object isn't in memory also failed in similar ways, but not sure (the
code is meant to deal with this case..)

Probably better to rethink the patch to use the "vary id" as a timestamp
instead, marking objects with an older update time as stale.

Regards
Henrik