Updating B can invalidate a whole set of keys, and B cannot be aware
that A depends on it*. Because of memcached's unreliability of having
B at given time.

So if B drops out of memcached all keys dependant on it have to be
invalidated.

* Seen a few implementations where B contains a set of keys to
invalidate, but those implementations are wrong. If B drops, then
values tagged with it cannot be invalidated/deleted.

Jared

On Sep 12, 7:30 am, Henrik Schröder <[email protected]> wrote:
> Why don't you just invalidate A when you update B instead? That would save
> you a lot of operations having to check B every time you fetch A?
>
> /Henrik
>
>
>
> On Fri, Sep 11, 2009 at 18:10, Ren <[email protected]> wrote:
>
> > Hi,
>
> > If A depends B would it be safe to embedded B's CAS token into A's
> > value,
> > then when A is retrieved see if B's CAS differs to the one stored
> > within A, and therefore invalidating A?
>
> > Jared

Reply via email to