[rabbitmq-discuss] Scaling

[Emile Joubert]

Hi Reefee,

On 29/10/10 10:51, Reefee wrote:
> 
> I have a Java Servlet that listens to external messages delivered via HTTP
> POST.
> I am using the rabbitmq. Java SDK 
> The servlet then processess them and stores them on to a RabbitMQ queue.
> 
> In order to offer best performance and scalability what is the reccomended
> usage of Connections and channels ?
> i.e. Have a single connection that gets re-used accross all requests and
> with each message being posted to the queue by its own channel instance (as
> it seems Channels being used by multiple threads is not a good idea).

It is possible to share a channel, but you will need to serialise access
to it. It is simpler and recommended to establish a channel per thread
as you suggest.

> Or should I create a seperate Connection for each request ?

>From your description it sounds like this client only acts as an AMQP
producer. The normal pattern here would be use one or many long-lived
connections. Also look at the rabbitmq-java-messagepatterns for a
wrapper that maintains long-lived connections. If the client needs to
act as an AMQP consumer as well then establish separate connections for
this purpose.

Creating a new connection per request would make sense if on average
messages are less frequent than network failures and you can tolerate
the additional connection setup overhead.


Regards

Emile