2008/5/28 TLFox <[EMAIL PROTECTED]>:
>
>
>
> Robert Greig wrote:
>>
>> 2008/5/28 Robert Godfrey <[EMAIL PROTECTED]>:
>>
>>> It is, indeed, a major omission of AMQP 0-8 / 0-9 that there is no
>>> synchronous publish method in the protocol.  This  is rectified in
>>> 0-10.
>>
>>
>>
>
> That does indeed seem a big omission.
>
> It basically means, that without using transactions, it's not possible to
> get once and only once delivery guarantee with 0.8 or 0.9, and by
> implication any JMS system built on top of AMQP 0.8/0.9 is not possible to
> be spec compliant.

it's possible to pass the TCK :-)  But I agree, to be 100% compliant
with the spirit of the spec then you would want a synchronous publish
in the underlying protocol.  I'm personally reluctant to add more qpid
"extensions" to AMQP0-8/0-9 when the machinery already exists in AMQP
0-10 which is what the trunk code / C++ broker implements.

> What I would suggest is to add a flag so users can choose whether persistent
> messages are sent sync or async. That can default to async so you get better
> performance, but you need some way of letting users get normal JMS
> behaviour. Otherwise you risk losing customer data, since believe me, JMS
> users will assume persistent messages are sent sync (every other system I
> know does it this way).

If we were to add an AMQP extension then I would assume we would also
aysnc publish to be turned on / off on a per producer basis...

-- Rob

Reply via email to