Doug,
I apologize
if my rant below is a bit to cryptic, let me try again:
1. When
a catastrophic failure occurs (e.g. RMD amnesia), an RMS has to react in some
way; It could return an error to the user or it can engage in a recovery
mechanism of some sort. I don’t believe you are trying to prescribe what
the RMS’s reaction ought to be.
2. As
you’ve said time and again, this proposal is about getting the RMS an
accurate ack set in cases where: 1. A full ack set will never be possible (or
at least not in a reasonable amount of time), there are messages that have been
sent and for which no ack has been received and 3. The problem that prevents a
full ack set doesn’t prevent the exchange of protocol messages.
The point
I was trying to make is that given #1 above, #2 is an optimization for a case
that will be relatively rare. Note that I don’t question for a second the
correctness of your proposal – what concerns me is adding elements to the
protocol for this specific case, #2, especially since apps will have to deal
with #1 anyway.
Note
that thus far, we’ve managed to describe exactly one scenario that fits
the #2 description: RMS has separate state stores for session state and
messages – the latter fails but the former is still operable.
--Stefan
From: Doug Davis
[mailto:dug@us.ibm.com]
Sent: Wednesday, August 31, 2005
3:58 AM
To: ws-rx@lists.oasis-open.org
Subject: RE: [ws-rx] i0019 - a
formal proposal - take 2
I'm having a hard time following this. I
sounds like you're saying because the proposal does not solve all RM related
problems you don't want to have it in our 'bag of tricks' at all. Following
that logic, why should we distinguish between SequenceTerminated Fault and any
other Fault? We do it because we want to provide as much information back
to the RMS as possible. What it uses this information for is up to it.
As I've said may times before, this proposal does not suggest ANY recovery
scheme. What I've done (outside of the proposal itself) is discuss how I
_think_ an RMS might use this information in some error recovery mechanism but
this proposal itself does not suggest one. This proposal simply provides
a mechanism for the RMS to get an accurate accounting of the state of the
sequence - that's it. How the RMS uses this information is up to it. If
for nothing else it may choose to simply log the information - that alone is
invaluable to someone trying to figure out what's going on. And I'm
having a hard time understanding why providing an _optional_ mechanism that
could aide in the RMS getting an accurate accounting of the state of the
sequence (without having to call up the RMD's admin) is a bad thing.
thanks,
-Doug
"Stefan Batres"
<stefanba@microsoft.com>
08/31/2005 01:48 AM
|
To
|
Doug Davis/Raleigh/IBM@IBMUS,
<ws-rx@lists.oasis-open.org>
|
cc
|
|
Subject
|
RE: [ws-rx] i0019 - a formal proposal - take
2
|
|
Doug,
You mention a specific situation: An RMD experiences a failure that
prevents it from receiving application messages. I agree in so far as saying
that in such a failure case this proposal could be helpful in that it helps the
RMS to engage in recovery of some sort (either inform applications that a
specific message was not sent or open a new sequence, assuming ordering is not
important). But this is not the only failure case that applications will want
to deal with (with or without help from the protocol).
Consider the case where connectivity is lost for long enough for both
sequences to expire or consider the case where the destination suffers a loss
of session state. In such failure modes this solution is not helpful –
yet applications will need a recovery strategy of some sort. It might be that
it is application specific, or it might be that a general failure recovery
specification is created and ratified at some point. The important idea is that
the only way to deal with all failure modes is at higher level. This proposal
leverages the protocol to optimize recovery in specific circumstances that
should be relatively rare. RM implementations should not be required to support
failure mode recovery mechanisms that either don’t apply to them or that they
choose to implement in a uniform way at a higher level.
Thanks
--Stefan
From: Doug Davis
[mailto:dug@us.ibm.com]
Sent: Tuesday, August 30, 2005 1:08 PM
To: ws-rx@lists.oasis-open.org
Subject: RE: [ws-rx] i0019 - a formal proposal - take 2
Yet more comments. :-)
-Doug
"Stefan Batres"
<stefanba@microsoft.com>
08/30/2005 03:35 PM
|
To
|
Doug Davis/Raleigh/IBM@IBMUS,
<ws-rx@lists.oasis-open.org>
|
cc
|
|
Subject
|
RE: [ws-rx] i0019 - a formal proposal - take
2
|
|
Doug,
Some more comments and thoughts on your proposal:
<dug>... When or why an RMS uses CloseSequence is up to it to decide.
All we know is that it wants to shut things down and get an accurate ACK from
the RMD.</dug>
I still have not heard of a plausible reason why an RMS “wants to shut
things down” and the current spec presents a problem. Comparing the spec
as it stands today vs. the spec + this proposal:
- TODAY: RMS wants to end the sequence so it
sends a LastMessage and must wait for a complete set of acks; this might
require retransmitting messages. Once a full set of acks is received RMS
sends TerminateSequence.
- TODAY + THIS PROPOSAL: RMS wants to end the
sequence so it sends Close, waits for a CloseResponse, possibly
retransmitting the Close. Once a CloseResponse is received RMS sends
TerminateSequence.
The problem with the TODAY scenario, as I’ve heard it in this forum, is
that the RMS might have to wait unacceptably long between sending LastMessage
and getting a full ack range. But if getting some messages or acks across
proves difficult; why would the RMS expect that getting Close across would be
any easier?
<dug> 1 - I don't believe your text is accurate in that Close is supposed
to be used in cases where the sequence needs to end due to something going
wrong. You've described a case where the sequence is functioning just
fine - and while Close can be used in those cases as well, it provides no
additional value. 2- Sending a Close and sending application data can
have quite a different set of features executed so I don't think its hard to
imagine cases where RM messages can get processed just fine but application
messages run into problems. I believe Chris mentioned on some call the
notion of two different persistent stores - one for RM data and one for
app-data. Its possible that the app-data one is running into problems.
3 - Using the CloseSequence operation is option - if you feel that, as an
RMS implementor, you'll never see its usefulness then you're free to never
implement/send it. However, I'd hate remove this option for those of us
who do see value in it. </dug>
<dug>The case that I keep thinking about is one where the RMD is actually
a cluster of machines and when a sequence gets created it has an affinity to a
certain server in the cluster - meaning it processes all of the messages for
that sequence. If that server starts to have problems, and for some reason it
just can't seem to process any new app messages then the RMS can close down the
sequence and start up a new one. Hopefully, the new sequence will be directed
to a different server in the cluster. </dug>
There are two problems with this scenario and the proposed solution.
1.
If an RMD has sequence-to-machine affinity that should be
strictly the RMDs decision and the RMDs problem. The RMS is autonomous; this
proposal puts expectations on the RMS’ behavior based on particularities
of the RMD implementation. To be clear, I’ll note that affinity can be
achieved in two ways:
i. By
performing stateful routing at the RMD; basically the RMD has to remember every
active sequence and what machine it has affinity to. In this case it would be
simple to change the RMD’s routing table when a machine fails.
ii. By
generating different EPR’s for each machine. For affinity to function
this way two things are necessary:
1.
Some sort of endpoint resolution mechanism would have to
be devised for the RMS to learn the EPR that it should target.
2.
A mechanism for migrating that EPR.
Clearly
1) and 2) are outside the scope of the TC and, in my view, this proposal might
be defining 2) in an informal way that is specific to WS-RM.
2.
If the RMS somehow guesses that there is a problem on the
EPR to which it is sending its messages and somehow decides that Closing the
sequence and starting a new one is the right course of action, ordering
guarantees are compromised.
<dug> I probably didn't state the problem very well. I didn't
intend to claim that the RMS knew about this affinity, but instead it knew that
something was wrong with the current sequence and in order to try to fix the
situation it decided to try another sequence. The affinity bit was thrown
in there to explain why starting a new sequence _might_ fix the problem.
I should also point out that while a lot of these discussions have focused on
InOrder+ExactlyOnce DA, this feature is still useful in other DAs. For
example, if the DA is just ExactlyOnce - having an accurate accounting of the
ACKs allows a subsequent sequence to send just the gaps from the first, so
getting an accurate list of the gaps becomes critical. And this of course
leads us to the discussion of how to determine the DA in use - which I think
might be part of issues 6, 9, 24 and 27.
</dug>
Finally, I agree with you that considering a gap-filling mechanism would be a
good thing for this TC to do.
--Stefan