Re: Cascading invalidations (draft-ietf-httpbis-cache-groups)

On Mon, Apr 28, 2025, 9:16 PM Martin Thomson <mt@lowentropy.net> wrote:
>
> On Sat, Apr 26, 2025, at 13:05, Mark Nottingham wrote:
> > It surprised me too, but it's a reasonable reading of the current spec.
> > The problem is that -- by necessity -- we don't specify or constrain
> > what the source of invalidations might be, and that can be read to
> > include invalidations *caused* by group membership.
>
> The domino effect seems pretty obvious here, but I think that the best approach would be to set expectations: only those resources that share one of the groups is affected.  You don't then trigger invalidation logic again.  Otherwise, incautious setup of groups could wipe your entire cache, which is a DoS attack recipe ingredient, if nothing else.

If a resource is in multiple groups do all the groups trigger, or just
one when invalidation happens? Right now we have it as all. But let's
say we have a  splash page showing prices of pool noodles and
frisbees, and the price of pool noodles changes, so the origin knows
that the group 'has-pool-noodle-price' is now invalid.  The splash
page getting invalidated would invalidate all pages with both, but
really we want to be more selective.

I regret not noticing this earlier, and of course 'purge-by-group'
propertitary mechanisms solve this somewhat, so it isn't completely an
issue.
>

Received on Tuesday, 29 April 2025 04:31:12 UTC