Doh! Looks like I'd already started to address this issue in the draft
revision I was working on...
--On Wednesday, February 27, 2002 21:02 -0600 Mike Dahlin
<dahlin@cs.utexas.edu> wrote:
> It is far from clear that "invalidation sets" and "registration sets"
> should be the same and/or have the same requirements. (For example, in
> most invalidation systems, "registration sets" are all objects at a
> server and "invalidation sets" are individual objects).
>
> I believe it will clarify the document to separate these concepts so
> that the document can, if necessary, specify different requirements
> for the different sets. Such a separate specification will (a) make
> the specific requirements more clear and (b) allow either an
> implementation that uses separate mechanisms or an implementation that
> uses the same mechanism for both.
It's probably useful for me to quote what I've got under the heading
"resource grouping" at present (with apologies for lack of context since
you can't see the rest of the draft):
<quote>
It is common for multiple resources that are related to be updated or
removed at the same time. By grouping such resources into RESOUCE GROUPS
(by examining content management systems or server logs, for example) that
are addressed in their own right it is possible to invalidate many
resourced with a single INVALIDATION.
RUP protocols MUST provide a mechanism to enable RUP CLIENTs to update
multiple entities in their corresponding caches as a result of receiving an
INVALIDATION for a RESOURCE GROUP from a RUP SERVER.
RUP SERVERs SHOULD send INVALIDATIONS for only those RESOURCE GROUPs to
which a given RUP CLIENT has subscribed.
</quote>
The notion of "invalidation sets" seems fairly OK there.
However, that RFC2119 "SHOULD" in what we're calling the "subscription set"
is a problem. We could change it to lowercase and lose the strict RFC2119
definition but I'm not sure that would help. Changing to "MAY" seems
meaningless.
Why the problem with SHOULD? There are going to be situations where it's
beneficial for the RUP SERVER to send INVALIDATIONS for RESOURCE GROUPS to
which the RUP CLIENT has not subscribed. Use of "SHOULD" implies that's a
Bad Idea(tm) that needs to be really thought about and I'm not convinced
that's the case.
... maybe it's just better for me to submit the draft as-is so folks can
see the changes? You'll probably want to make sure you have a copy of -02
downloaded already for comparison.
This archive was generated by hypermail 2b29 : Thu Nov 18 2004 - 11:23:01 MST