RE: [ws-tx] Issue 016: WS-C: ReplaceParticipant (and partly 007 too)

["Peter Furniss" <peter.furniss@choreology.com>]

There is a general problem with assuming a recovered ws-tx entity can't
get back to the endpoint it started on, so they have to tell the other
side what the new EPR is. What happens if both side suffer such forced
moves at the same time (not at all improbable, especially with
long-running cases).  If both sides fail, recover and come up on
different endpoint addresses and there is no forwarding/fallback how
will they ever find each other to complete the coordination protocol
exchange ?  Timeouts won't help if the relationship has already reached
committing/prepared (in AT) or its equivalent.


I think we must assume (and perhaps state explicitly) that we are
assuming the EPR's exchanged on Register/RegisterResponse will continue
to be valid for the lifetime of the Participant/Coordinator. It would
still make sense to provide replacement (redirected) EPR's on by one
means or another, but it needs to be ensured that a message sent to the
original EPR will get there (though perhaps less efficiently than for
the new EPR).

In general, for EPRs this shouldn't be that onerous (certainly compared
to middle-period CORBA, say) given the flexibility of dns, the
possibility of local interpretation of the latter part of URI and the
use of returning ReferenceParameters. 



There are other possibilities, but perhaps a seasonal analogy would be
helpful.

In sending our Christmas cards, there were some cases where we aren't
sure whether the address we have is correct. If it isn't and the friends
from the past have lost our address too, we are probably never going to
be in contact again. But if the uncertainty is over my cousins' address,
I know I can ask my aunt, and can ensure communication is maintained
(assuming I want to, of course). Of course, for the long-lost friend, I
could always try FriendsReunited.

But I'm not sure of the feasibility of ParticipantsReunited.

Peter