cxf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Glynn, Eoghan" <eoghan.gl...@iona.com>
Subject RE: Identification of Partial Responses
Date Fri, 12 Jan 2007 10:47:22 GMT

> -----Original Message-----
> From: Dan Diephouse [mailto:dan@envoisolutions.com] 
> Sent: 11 January 2007 17:44
> To: cxf-dev@incubator.apache.org
> Subject: Re: Identification of Partial Responses
> 
> A quick question about the two way case - if its already 
> decoupled, why can't the server sent out a 
> <SequenceAcknowledgement> at some point during the two way 
> operation.

The wsa:replyTo and wsrm:acksTo are specified *independently*, and may
well be different.

The twoway case in which we absolutely *need* partial responses, is
where the wsa:replyTo is non-anonymous but the wsrm:acksTo is anonymous.

So even though the MEP is decoupled for application-level message, ACKs
cannot be sent as you describe in this scenario.

> For instance we have this scenario
> 
> - A sends message to B
> - B takes an hour before it sends its response back to A
> - During this time B sends a one way SequenceAcknowledgement 
> to A so as to avoid resends

Only if *both* the wsa:replyTo and wsrm:acksTo are non-anonymous, a
SequenceAcknowledgement could be sent as you describe. In fact we
already do this.

But its still useful (if not absolutely necessary) to send partial
responses even in this case. Remember that RM ACKs acknowledge the
*entire* set of messages received thus far for a particular sequence,
not just the individual current message. So we may well have pending
ACKs backed up from the previous incoming messages, when the current
message is received. Rather than going to the trouble of immediately
opening a separate server->client connection to send these backed up
ACKs, its much more sensible to piggy-back them on the back-channel of
the incoming connection on which we received the current request.

However, this is just for convenience, so lets not argue the point. The
main issue that makes your suggestion unworkable is pointed out above,
i.e. the decoupled replyTo but anonymous acksTo case.

At this stage we really need to bring this thread to a conclusion.
Challenging the groupthink is a very useful exercise, but its beginning
to burn a lot of time, and I feel that we're inching inexorably to a
consensus that partial responses (with some minor mods) are actually
required. So can we agree that the partial response mechanism has
survived the (sustained ;) challenge?

Cheers,
Eoghan


Mime
View raw message