Hi everybody,
Just to add my experience with network of brokers: we just tried
disconnection/reconnection: most of the time it did not work. The bridge
(DemandBridge...) between brokers does not re-initialize. See my
previous emails on this issue.
Moreover, there is an issue with ClientID - with durable topics. If
somewhere in your network you have a clientID="foo" you can not have
another clientID called ="foo" . This is a problem when your clients are
MDB, because the ClientId is written in the ejb-jar.xml file. So you can
not have the exact same ejb jar file on different servers related with a
network of brokers - why not use the brokerName as a defaultClientID or
something like this?
By the way, why not having brokerName=IP of the machine by default ?
Something else is very annoying: the documentation is ... at least hard
to follow. Say the master/slave feature: I read it several times - and
other persons did that too: we did not understood easily how to do it...
Same for several parts of the documentation. I dream there was a one
piece documentation - a la Springframework for instance (or a la Joram,
which has a very good documentation).
I know it is a lot of complex features, lots of complex developments,
lots of time given by people. And this is an open source project so,
don't get me wrong, I do not complain, because I should rather help
coding... And activeMQ is a good product, no doubt on that.
But as our project is not released yet, and as we tested before to
release it, and had time before going live, we changed the JMS broker.
We switched to Joram and... it just works! Disconnection, re-connection:
ok. Several identical clientID on the same network of brokers: ok too.
etc. We do not have discovery but that's ok, we can live without that
feature - but we can not live with a network of brokers that loose
messages, does not reconnect etc. Sorry guys.
(When I will have more time I will do some bug reports... )
Bernhard Wellhöfer a écrit :
Hello Javier,
Sorry, it was not my intention to make waves. I just want that the whole issue
is taken seriously:
A) If the clustering/network or brokers feature is stable, then the
documentation and the support via the mailing list are not good enough.
B) If the hints (my tests, number of mails for this feature and quality and
quantity of answers) are correct and the feature is not in a production state,
then the ActiveMQ developer should face this and warn the users on the homepage.
At least option A or B is true (and for you I hope it is option A) ....
Regardless whether A or B is true, the whole issue is not good for the trust in
ActiveMQ.
One question: did you ever start/stop brokers and consumer/producer in the
network during your tests. Did all broker and consumer/producer reliable
reconnect and where all message delivered?
Regards,
Bernd
-----Ursprüngliche Nachricht-----
Von: Javier Leyba [mailto:[EMAIL PROTECTED]
Gesendet: Donnerstag, 21. Dezember 2006 09:05
An: activemq-users@geronimo.apache.org
Betreff: Re: Network of brokers
On 12/21/06, Bernhard Wellhöfer
<[EMAIL PROTECTED]> wrote:
Hello Prashanth,
I agree that the clustering/network or brokers feature is
not ready for a productive scenario. The problem is that the
ActiveMQ homepage does not warn you about this case.
Therefore again and again people waste time by trying out
this feature - as each new mail for this issue to the mailing
list proves. It's a pity that this as a whole makes ActiveMQ
less trustworthy...
Is this totally true ???
I've an application that use network of brokers ready to go to test.
I've configured it with the help of this list and nobody told
me what you say...
If this is true I'll be in big problems after four months of
development with this product. I can´t believe it !
If this is true I'm a dead man... :(
--
Javier Leyba
Barcelona - Spain
http://blog.leyba.com.ar
