Re: [wsrm] New Issue

["iwasa" <kiwasa@jp.fujitsu.com>]

Sunil,

I am neutral whether we allow to send back Poll
reply in a different HTTP connection or not.

However I don't think the proposed changes are appropriate,
because it creates inconsistency. (E.g. Figure 3)
The essential concern I have with your proposal is
inconsistency of the relationship among Response, Callback
and Poll reply pattern.
Because an appearance of replyTo attribute is used to
differentiate bindings of sync Poll Reply(which reply is
on the same HTTP connection with PollRequest) and
async Poll Reply(which reply is on the different HTTP
connection with PollRequest).
In that case, why don't we do the same thing
for Response reply pattern and Callback reply
pattern?

In other words, we could have two messaging models(or
reply pattern or whatever) but not three reply pattern
as follows:
    Model1:
        1. Sender sends Reliable Message to receiver.
        2. Receiver sends Ack or Fault to the sender.
    Model2:
        1. Sender sends Reliable Message to receiver.
        2. Sender sends Poll message to receiver.
        3. Receiver sends Ack or Fault to the sender.

And model1 will be either:
    - The current Response reply pattern
      (Ack or Fault is in the HTTP Response) or
    - The current Callback reply pattern
      (Ack or Fault is in the HTTP Request)
And model2 will be either:
    - The current Poll reply pattern
      (Ack or Fault is in the HTTP Response) or
    - New poll reply pattern
      (Ack or Fault is in the HTTP Request)
And the appearance of replyTo differentiate
those two binding patterns for each of above models.

Actually this is what I have insisted in the F2F meeting
last year. Despite of that, we decided on the current three
reply pattern with your request (which is mixing HTTP binding
with messaging models, IMHO). I would say it was not
good decision and it made the spec depend on the HTTP like
(Req/Res type) underlying protocol. But anyway we chose it and
the many of the spec depend on the three reply pattern now.
So if we want to change that, it might require to update many more
portion of the spec.

In conclusion, what I would propose are:
    1. We keep it as is.  (It means that we do not allow
        Poll reply in the HTTP request.), or
    2. Allow to send Poll reply in HTTP request. And
        update other portion of the spec consistent as
        described above.
        (Replacing three reply pattern with the above
          two messaging models.)

Since we have time limited, 1. above might be
better for now unless TC member agrees with
this change immediately, and accept only three days review
period (from Monday to Wednesday next week)
for this change. In that case, I will do this change by
the end of this week.

Anyway I don't think this is a critical issue
we have to resolve immediately.
We can discuss this for the next version if TC exists,
and if we want. Timing to finalize the current spec
is more critical, I believe.

Thanks,

Iwasa




----- Original Message ----- 
From: "Sunil Kunisetty" <Sunil.Kunisetty@oracle.com>
To: <wsrm@lists.oasis-open.org>
Sent: Wednesday, March 03, 2004 6:47 AM
Subject: Re: [wsrm] New Issue


>
>  If we do accept this, here are the  changes that we need to do.
>
>   1. Add an optional replyTo attribute to PollRequest
>   2. pg 6/line 177: Remove the replyTo attribute for Poll pattern in
Request  (this may contradict to what I said in the editorial comments...)
>   3. pg 7/line 212:Title of Example 3 would read "Acknowledgment Message
embedded in HTTP Request"
>   4. pg 7/line 213:HTTP Headers will have to change (should use POST)
>   5. pg 8/line 239:Title of Example 4 would read "Fault  Message embedded
in HTTP Request"
>   6. pg 8/line 240:HTTP Headers will have to change (should use POST)
>   7. pg 11/lines 334-339 should be reworded as follows:
>
>           We say that Polling RM-Reply pattern is used if a second
underlying request is issued to the
>           receiver of a previous message, in order to obtain a RM-Reply.
The RM-Reply can be either
>           contained in the underlying response to this poll request or in
a separate underlying request
>           from the receiver to the sender. This poling pattern is
generally expected to be used in
>           situations where it is inappropriate for the sender of reliable
messages to receive underlying
>           protocol requests (behind the firewall cases) or to avoid
resending bulk messages often.
>
>   1. pg 15/Figure 3. The 3rd line should be titled "Underlying protocol
Response/Request".
>   2. pg 28/section 4.3
>        1. Table 9: Add an optional attribute call replyTo of type anyURI
>        2. We need to mention that RM-Reply MUST be contained in the
underlying response of the Poll request if this attribute doesn't exist and
should be sent  in an underlying request to the endpoint identified by this
attribute if exists.
>   3. And finally the schema has to reflect this  by adding an optional
attribute to the PollRequest element.
>
>  -Sunil
>
> Sunil Kunisetty wrote:
>
> >  The current definition of the Poll RM-Reply pattern says that the Poll
response
> >  should be the same underlying transport connection as that of the Poll
request.
> >
> >  This is good and useful for Senders behind the firewall case and if
this is the
> >  only usecase we were supporting. This was indeed the case onetime.
> >
> >  However, since then, we have relaxed the usage of Poll and made it a
general
> >  purpose status query kind of thingy which is usable even with Request
and
> >  Callback.
> >
> >  However, we cannot use it with Callback if the underlying transport is
truly one-way
> >  such as JMS transport.
> >
> >  Take the following example:
> >
> >  Sender sends a one-way message with RM-GD feature. However, it hasn't
received
> >  the Ack or Fault for a long time. Assume he is using a pure one-way
transport.
> >
> >  Instead of retrying. He wants to poll it again. Since his transport is
one-way, he
> >  can't get the response on the same connection as that of the request.
Instead,
> >  he wants the Receiver to send him the Poll Ack/fault to its listener
just as the
> >  callback.
> >
> >  Note that this use case assumes the Sender can listen to acks and
faults.
> >
> >  However, the current specification doesn't have this provision. And
hence
> >  I propose that we remove the restriction on the response.
> >
> >  I propose that PollRequest takes an optional ReplyTo element and if
> >  present, the Receiver sends the Poll response to this endpoint.
> >
> >  If this element doesn't exist, then the Receiver sends (or rather
attempts)
> >  the response in the same connection.
> >
> >  Comments???
> >
> >  -Sunil
> >
> > To unsubscribe from this mailing list (and be removed from the roster of
the OASIS TC), go to
http://www.oasis-open.org/apps/org/workgroup/wsrm/members/leave_workgroup.php.
>