qpid-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gordon Sim <g...@redhat.com>
Subject Re: svn commit: r1642720 - in /qpid/trunk/qpid/cpp/src/qpid/messaging/amqp: AddressHelper.h ConnectionContext.cpp ReceiverContext.cpp ReceiverContext.h SessionContext.cpp SessionContext.h
Date Tue, 02 Dec 2014 21:13:28 GMT
On 12/02/2014 07:49 PM, Alan Conway wrote:
> On Tue, 2014-12-02 at 17:57 +0000, Gordon Sim wrote:
>> On 12/02/2014 05:38 PM, Gordon Sim wrote:
>>> On 12/01/2014 05:41 PM, aconway@apache.org wrote:
>>>> Author: aconway
>>>> Date: Mon Dec  1 17:41:09 2014
>>>> New Revision: 1642720
>>>>
>>>> URL: http://svn.apache.org/r1642720
>>>> Log:
>>>> QPID-6252: AMQP 1.0 browsing client generates large number of errors
>>>> on broker.
>>>>
>>>> The problem was that messages for browsing receivers were being
>>>> recorded on the
>>>> client SessionContext unacked list. This is incorrect since you don't ack
>>>> browsed messages.  They remained on the list after the browsing
>>>> receiver was
>>>> closed, and every subsequent call to acknowledge() on the client would
>>>> attempt
>>>> to ack these messages for a no-longer-existing link. Fix is to not record
>>>> browsed messages.
>>>
>>> I don't think this is right.
>>>
>>>   From the client's perspective, it should still be able to accept and
>>> settle deliveries, regardless of whether the broker will dequeue in
>>> response.
>>>
>>>   From the protocol perspective, deliveries must always get settled.
>>> Otherwise they need to be tracked (whether within proton or on the
>>> sending peer).
>>>
>>> As regards the original error, the spec says:
>>>
>>>       The disposition performative MAY refer to deliveries on
>>>       links that are no longer attached. As long as the links
>>>       have not been closed or detached with an error then the
>>>       deliveries are still “live” and the updated state MUST
>>>       be applied.
>>>
>>> It doesn't explicitly say what should happen if the link has been closed
>>> (not just detached), so its probably best to just ignore it on the
>>> broker (change the log level from error to info perhaps).
>>>
>>> The client should probably also settle all existing deliveries received
>>> on a link before closing it.
>>
>> Or perhaps at least remove the deliveries from the unacked list if their
>> associated link is closed.
>>
>
> The client definitely needs to do something with these deliveries,
> otherwise they accumulate on the session and you get longer and longer
> lists of bad acknowledgements as time goes on. I don't think anyone on
> the user interface side expects that they can/should ack messages from a
> browsing session, although I would think doing so should be a no-op not
> an error. Are you saying that the protocol requires it?

Yes, proton does also. However we could just settle at the point the 
delivery is received and not add to the unacked list at all (as you do 
now). As the code stands now that would be sufficient. Ideally I think 
acknowledge is compatible with browse, it just serves a different 
purpose. Rather than directly causing a message to be dequeued, it 
signals to the broker that this subscriber has seen it. A future 
implementation on the/a broker could then use that to determine when all 
registered subscribers have seen it.

> From my reading
> the stuff about settlement only applies to "acquired transfers", which I
> took to exclude transfers from a link that is in COPY mode.

You could argue that the accept outcome is only relevant to acquired 
transfers (in my view it's not that clear in the spec text). However 
settlement is distinct from that.


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@qpid.apache.org
For additional commands, e-mail: dev-help@qpid.apache.org


Mime
View raw message