Re: [wsrm] Singleton group with SequenceNumber?

[Tony Graham <Tony.Graham@Sun.COM>]

Sunil Kunisetty <sunil.kunisetty@oracle.com> wrote at Sat, 21 Feb 2004 10:58:00 -0800:
> Tom Rutt wrote:
> > Tony Graham wrote:
> > >>From off-list mail:
> > >
> > >Sunil Kunisetty <sunil.kunisetty@oracle.com> wrote at Fri, 20 Feb 2004 17:49:54 -0800:
...
> > >A group of 1 is not valid since the 'status' attribute can't indicate
> > >both the start and end of a group, and the 'status' value has to be
> > >'Start' for the first message in a group (line 989):
> > >
> > >   The sender node MUST send a very first message, to guarantee the
> > >   message order, with "Start" for this attribute.
> 
>  Here is the glitch.  The 'end' message is NOT A MUST for a group to
>  be complete. As such the group can be terminated (and hence completed)
>  by other criteria such as group expiry time, max idle duration etc...

groupMaxIdleDuration="0" would terminate the group sometime in the
next second, yes.

Just because it would work doesn't make me think that purposely
creating groups of 1 is to be recommended.

>  So 'end' message is just a convenience. We cannot enforce that every
>  group MUST only be completed by an 'end' message.

Understood.

> > >...
> > >
> > >> Currently there is no way to distinguish a Group with only one message
> > >> (that uses SequenceNum)  as a Singleton unless Request Header is
> > >> also changed to reflect something similar.
> > >
> > >Then Section 2.3 should be changed or the 'status' attribute should
> > >allow a 'Single' value (or similar).  At this point I think that it
> > >would be simpler to change Section 2.3.
> 
>  I think this is not necessary as I explained above. It may help a
>  a group of 1 message case, but then it will be very error prone
>  for other cases.

I think I see your point.

If a sending RMP dies after successfully sending the first message in
a group and picks a new Group Id when it restarts, that first message
just became a group of 1.  So groups of 1 can happen even if we
legislate against them.

> > I agree it would be easier to have a singleton group never use the
> > sequence no in its request.
> 
>  We really cannot enforce this and atmost can suggest in bold letters
>  that it is efficient for singleton messages to be without no seq.o.

I think that the spec should make it seem like less of an option.

> > If this was the case, how would your proposed schema for the
> > messageReplies look like?

I need to rethink.

Regards,


Tony Graham
------------------------------------------------------------------------
Web Products, Technologies and Standards           Phone: +353 1 8199708
Sun Microsystems                                              x(70)19708
East Point Business Park, Dublin 3, Ireland