On 11/29/2013 11:17 AM, Toralf Lund wrote:
On 29/11/13 11:19, Gordon Sim wrote:
On 11/29/2013 09:08 AM, Toralf Lund wrote:
Hi,
Just wondering, what gain is there in reusing Sender objects for the
C++
Messaging API? I mean, over doing something like
qpid::messaging::Sender sender=session.createSender(address);
sender.send(message);
sender.close();
every time?
The biggest problem is that the createSender() is synchronous, and
may in some cases involve more than one request to the broker. That
adds to the latency of the response.
That could be a bit of an issue.
I'm thinking about this in conjunction with a request-response type
setup where the reply address is specified via
message::getReplyTo(), so
it's not entirely obvious how long the senders should be kept, if at
all, but I do know that the same "reply-to" will generally be used
several times.
Any thoughts?
One option to reduce the number of distinct senders, and therefore
benefit more from caching, is to use a single a sender per exchange
(this is with AMQP 0-10) and just set the subject of the response
explicitly to match the required routing key.
Right.
It's not entirely clear to me how that might be done, though. I mean,
the reply address is just a queue - there is no explicit binding or
anything.
The one issue here at present is that there is no way to create a
sender for the default exchange. That has been requested before (I
can't find a JIRA though) and it would perhaps be worth exploring.
Ideally it would be done in a way that would translate fairly
naturally to 1.0 as well.
Yes. Maybe this is what I'd need for the "simple queue" address?
I was having the same problem, and someone (sorry for not remember who)
suggested this, that helped:
bool create = false;
Sender sender;
try {
sender = session_.getSender(queue);
} catch (KeyError& e) {
create = true;
}
if (create) {
sender = session_.createSender(queue);
}
The session may already have the sender for the queue, so you just reuse it.
However, if you are able to control the types of reply-addresses you
use (e.g. amq.direct/# where # will be expanded to a uuid),
Yep. I'm still at a stage where I can easily change the addressing
scheme.
So, in this setup, I'd use the "#" bit as subject?
- Toralf
then you could cut down on the distinct number of senders needed,
making caching more likely to help.
--
Bruno Matos
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
