Re: Re: Heuristics in BTP atoms

[Mark Little <mcl@arjuna.com>]

> Perhaps some of this issue comes down to "what does CANCELLED mean from a
subordinate?". If this occurs before READY (PREPARED) has been sent by the
subordinate or before CONFIRM has been received by the subordinate, then the
subrdinate is informing that it can no longer meet the guarantee implied by
the transaction. That does not necessarily mean that the subordinate
*cannot* perform the action requested. It may be able to, or it may not be
able to. (It only agreed to tie up resources for a specific period of time,
and that period has elapsed - whether those resources are still available is
another matter.)
>

There is definitely a distinction to be made about when CANCELLED occurs.
I'm assuming that Sazi is talking about it being sent after PREPARED,
because other participants have replied with CONFIRMED. Continuing with that
assumption, I'd say we are now into the domain of a workflow system -
definitely in the domain of the application, because only it knows what it
means to have a heuristic and how to (attempt to) resolve it. Individual
participants don't have that knowledge, since BTP does not tie them to
services, and does not require them to even reside on the same node as a
service. They are the 2PC actor, and not the "book taxi/hotel" actor.

> If the subordinate *is* able to perform the requested action after all,
perhaps under these circumstances we should give it the opportunity to do
so.
>

This is an interesting point, but one I'd prefer to discuss in some revision
task force.

> We could allow this by permitting the superior to send a confirm for the
atom to the subordinate after having received cancelled, and permit the
subordinate to 'resurrect the atom' and reply CONFIRMED -or- if it is not
possible to do so, to reply CANCELLED (or perhaps UNKNOWN).
>

In addition, I seem to remember there was discussion about the possibility
for a coordinator to send messages to participants to check that they are
still able to prepare even if their time-limit has expired. I don't have a
spec. to hand to check this though.

> Hmmm. Maybe the second CANCELLED should be qualified to indicate it is in
response to CONFIRM (for the case where CONFIRM and CANCELLED cross in the
post).
>

Definitely time for a revision task force charter!

Mark.

----------------------------------------------
Dr. Mark Little
Transactions Architect, HP Arjuna Labs
Email: mark@arjuna.com | mark_little@hp.com
Phone: +44 191 2064538
Fax  : +44 191 2064203