On 07/20/2011 04:57 PM, Ken Giusti wrote:
----- Original Message -----
On 07/01/2011 07:04 AM, Ken Giusti wrote:
For #2:
The key for the header that contains the group identifier would be
provided to
the broker via configuration.
A fixed name like 'qpid.group' seems less likely to confuse. Perhaps
have id
default to 'qpid.group' but be cofigurable to another name in the
(unlikely?)
event that it is required.
That's a good idea. The current AMQP 1.0 draft defines a message property header called
"group-id", perhaps we should use "qpid.group-id" as the default instead of
"qpid.group"?
The advantage of a configurable header is that the group can be keyed
off some application meaningful property and the application need not be
altered just to introduce a particular arbitrary property.
I actually think that having a default makes it more confusing. You will
still have to turn grouping on explicitly, right? So why not have the
mechanism for so doing be the identification of a header from which the
message's group can be determined? I don't find that confusing.
2. ensure that messages are not selectively acknowledged or released
- order
must be preserved in both cases.
What happens if the consuming application fails to respect these
requirements?
Hmmm... good question. My initial thoughts are that this would be considered
an application error, since preserving order is important for this feature.
Perhaps the broker should respond with an error exception?
I think that is an application error. The aim is to allow processing of
the messages in order. If the application fails to sensibly communicate
regarding that processing, it is itself responsible for the results.
(I could imagine in the future ways of preventing this or alerting
applications about it, but I think for now that should be out of scope).
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
