Re: MOO Meeting 970305

James R Grinter wrote:

> that would imply that we really need some way of invalidating
> content for a cluster of caches. A new ICP feature/requirement?

Maybe it is enought (that is works almost always) if the "end" cache
(the one handling the PUT request) marks that the object needs to be
revalidated with the source server (it should NOT accept validation
against a parent/neighbor cache).

Here I assume that the object is cached (or didn't exist before) in the
cache handling the PUT request, and that it doesn't matter that much
that other people (using other caches) still get the "old" object.

But the problem is still there, and a cache hierarchy can't fully follow
the spec unless a "invalidating" ICP query is implemented. Even if the
PUT request is passed along the hierarchy, old objects may exists in a
neighbor which does't notice the PUT request since it isn't a parent...

Duane, is it possible to add a new ICP operation to invalidate objects,
without breaking existing ICP implementations? And will even this be
enought? What about large hierarchies?

---
Henrik Nordstr=F6m