[INFOD-WG] Dynamic consumer constraints
Fisher, SM (Steve)
S.M.Fisher at rl.ac.uk
Thu Jul 19 09:48:28 CDT 2007
> -----Original Message-----
> From: infod-wg-bounces at ogf.org
> [mailto:infod-wg-bounces at ogf.org] On Behalf Of Dieter Gawlick
> Sent: 12 July 2007 22:55
> To: INFOD
> Subject: [INFOD-WG] Dynamic consumer constraints
>
> All line references are from the published Base INFOD
> Specifications - see Current Documents folder
>
> Property Constraints (line 991) are used to derive Consumer
> Entry References (line 1788); Dynamic Consumer Constraints
> are past along to publisher in the Dynamic Consumer
> Constraints (line 1791).
>
> The specifications do allow that both types of constraints
> are specified concurrently, they do not say what the
> combination means. So, one possibility is to say: If both
> constraints are specified all messages go to consumers listed
> in the Consumer Entry References and additionally to the
> consumers that fit the Dynamic Consumer Constraints. Both
> entries and the Registry notification to the publishers are
> used. If we would like the Registry to provide the set of all
> possible consumers (for the Dynamic Consumer Constraints) we
> would have to add an entry in the notification.
That would be a bizarre reading of the spec. It makes no sense for one
constraint to be able to weaken another. Each time a new constraint is
imposed the total messge flow can be reduced but not increased. This
should be spelled out a little more clearly in section 1.1.1.4
> Calculating this list, however, seems to create fundamental
> problems; it may be very big and more importantly the
> registry may not understand the constraints because these
> constraints are referencing data vocabularies, which do not
> need to be XML vocabularies. So far, we avoided the registry
> to get involved in evaluating non-XML constraints.
This is something I wanted to define in the spec but was persuaded not
to. I think we either have to specify a plugin mechanism for a vocaulary
or define a vocabulary as a service. I much prefer the latter approach.
Meanwhile our spec does not, in practice, support non-XML data
vocabularies. This should probably be stated.
> We could also say that property constraints limit the
> eligible dynamic consumers; but this would have to be made
> very explicit.
We have always said that constraints are cumulative - I see no reason
for this to be an exception
> I recommend not to calculate the list of potential consumers
> and clarify the specifications in this respect.
>
> If we go this direction we do not have a solution for the
> publisher to become aware of new consumers for an existing message.
>
> However, here is a proposal for future versions: We specify
> that the registry has to support CQs (Continuous Queries). If
> a publisher likes to be informed about new consumers with a
> set of properties the publisher has to write a CQ against the
> registry. Obviously, we could also say that the registry has
> to accept subscriptions - CQs and subscription seems to be
> very closely related. We already had this in the
> specifications and removed in the NY F2F since we felt that
> the implementation is too difficult.
>
> Dieter
>
> --
>
>
> Oracle Email Signature Logo
> Dieter Gawlick | Architect | 650.506.8706
> Oracle Server Technologies
> 500 Oracle Parkway | Redwood Shores, CA 94065
>
More information about the infod-wg
mailing list