Re: [cdi-dev] JMS 2.1: Proposal to allow any CDI managed bean in a Java EE application to listen for JMS messages
To me the proposed approach seems to be very strict about JMS semantics
and bending CDI over it (not using CDI contextual instances properly,
calling @RequestScoped instances from multiple threads in parallel, ..).
A simpler approach would be a bridge where a JMS message would produce a
single CDI event. The semantics of synchronous CDI events would apply
from there. For users this should be straightforward as they would know
that they are working with a CDI event that was raised by the container
when it received a JMS message.
Jozef
On 26.08.2015 11:43, Nigel Deakin wrote:
> On 25.08.2015 16:05, Romain Manni-Bucau wrote:
>>
>> For this last case a really elegant solution would be to just reuse
>> @Observes to fire the message from the jms "container" listener and
>> propagate it to the "user" listener. This would then allow to
decouple
>> the application listener from JMS.
On 25/08/2015 15:26, Jozef Hartinger wrote:
> Agreed. I think we should leverage the existing CDI event/observer
> functionality instead of introducing a completely new
> delivery mechanism.
Can you please say a bit more about what you have in mind?
Romain suggests using events to invoke the "user" listener from the "JMS
container listener".
That's a useful distinction. Just to clarify the terminology:
"user" listener = listener bean provided by the application
"JMS container listener" = JMS consumer provided by the application
server or resource adapter
There needs to be one consumer for every listener bean since the two
need to have the same lifecycle, and also so we can implement JMS queue
sematics which require that a message from a queue is delivered to one
and only one listener.
The transaction needs to be started by the consumer before invoking the
listener and ended after the listener returns. This allows the
acknowledgement of the message (which is performed by the consumer) to
take place in the same transaction as is used by the listener's method.
Currently I'm proposing that the "consumer" invokes the
"listener" by a
simple method call. I suppose instead of simply invoking the method it
could fire a synchronous event, which only the associated listener
instance would receive, but I'm not sure what the benefit of this would
be. Since JMS semantics are very different from CDI event semantics I
think there's a danger that this will be confusing, since the user might
think they were getting CDI event semantics, but they were actually
getting JMS semantics.
Since this is a bit of a FAQ, it might be useful to explore the
differences between the two semantics, but currently they seem
profoundly different to me. That's why my proposals are built on the CDI
bean lifecycle model but not the CDI event observer model.
Nigel