Distributed HA clustering

[Jason Barto <ja...@gmail.com>]

All,
I've benchmarked my qpid broker (java) at 70k msg/s with a single producer
/ consumer scenario and am now turning my eye towards ha deployment. I see
from the documentation that a 2 node master / slave is supported on
clustering. What about distributed nodes for load balancing?

Is there any mechanism for having say 3 master / slave setups with
federation like rabbitmq or store/forward like activemq?

The goal is to have an always on 5 9s deployment.

Sincerely,
Jason

Re: Distributed HA clustering

[Maki Camara <Ma...@etu.enseeiht.fr>]

Hi everybody,

I am working on qpid proton-J and I would like to know if there is any 
example in java of a brokered usage for
send/receive scenario. I mean could someone give me the code of a 
broker(in java)
which receive a request and forward a response to the given address (in 
the request)?

Thank you a lot for your help

Maki

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
For additional commands, e-mail: users-help@qpid.apache.org

Re: Distributed HA clustering

[Alan Conway <ac...@redhat.com>]

On 04/17/2013 09:58 AM, Jason Barto wrote:
> All,
> I've benchmarked my qpid broker (java) at 70k msg/s with a single producer
> / consumer scenario and am now turning my eye towards ha deployment. I see
> from the documentation that a 2 node master / slave is supported on
> clustering. What about distributed nodes for load balancing?
>
> Is there any mechanism for having say 3 master / slave setups with
> federation like rabbitmq or store/forward like activemq?
>
> The goal is to have an always on 5 9s deployment.
>
The C++ broker supports federation between multiple clusters, with one 
restriction mentioned in 
http://qpid.apache.org/books/trunk/AMQP-Messaging-Broker-CPP-Book/html/chapter-ha.html#ha-overview

It also supports replication of queues between brokers or between clusters, see 
http://qpid.apache.org/books/trunk/AMQP-Messaging-Broker-CPP-Book/html/ch01s13.html

Cheers,
Alan.

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
For additional commands, e-mail: users-help@qpid.apache.org

Re: Distributed HA clustering

[Robbie Gemmell <ro...@gmail.com>]

Hi Jason,

There currently isn't any support in the Java broker for federation /
store+forward to distribute messages between brokers, if you wanted to do
that you would need to use a separate bridging process.

The optional BDB-HA store replicates the durable configuration and
persistent messages (non-persistent mesages are not replicated) at the
virtualhost level. Current work has mainly been done around the 2-node case
with explicit use of the associated 'designated-primary' flag to control
what happens when a node goes down, there is a split-brain scenario, etc.
Technically there is nothing preventing configuring an N-node BDB-HA
cluster with use of simple majority voting for handling split-brain etc,
but there are some interesting behaviours that occur when doing so which
would require a non-trivial amount of additional work to allow the
virtualhosts to actively monitor the cluster to account for and alter their
own behaviour accordingly (right now its mostly the reverse, with the
virtualhosts simply reacting to notification from the underlying BDB-HA
store nodes regarding their state, which is pretty clear cut in the 2-node
case) to ensure the overall system acted more in line with general
expectation.

Robbie

On 17 April 2013 14:58, Jason Barto <ja...@gmail.com> wrote:

> All,
> I've benchmarked my qpid broker (java) at 70k msg/s with a single producer
> / consumer scenario and am now turning my eye towards ha deployment. I see
> from the documentation that a 2 node master / slave is supported on
> clustering. What about distributed nodes for load balancing?
>
> Is there any mechanism for having say 3 master / slave setups with
> federation like rabbitmq or store/forward like activemq?
>
> The goal is to have an always on 5 9s deployment.
>
> Sincerely,
> Jason
>