On 8/8/06, nkiesel <[EMAIL PROTECTED]> wrote:
Hi all,
this is my first attempt to use ActiveMQ, so please bear with me if I'm
asking stupid questions.
Welcome! :)
I did try to find answers in the FAQ, the
docu, and the forums, but I'm still undecided how/what to do.
Requirements:
I have two (identical) Java applications running on tho machines
connected over the Internet. These two applications act in a
master/slave mode: every action done on the master must be repeated on
the slave. In case either the network or the slave is down, actions
should be queued so that they can be processed once the slave is online
again. On top of that, there is some transaction logic around these
actions: a group of actions should be either executed completely or not
at all by both the master and the slave. Finally, master and slave must
be able to switch roles (by restarting master and slave with e.g. a
command-line option).
Implementation idea:
I have a central place in the application where all actions pass
through. Basic idea is now to splice in a durable queue or topic (see
below) at this point that sends all actions over to the slave. The
transaction logic will be implemented by adding "begin" and "end"
actions. On startup, both apps look at their queue and either purge (if
they find a "begin" without an "end") or process the queue before
resuming execution.
I don't follow why you need a 'begin' and 'end' message? Is it to
process many messages in a single transaction? If so just have some
kinda marker on the last message (e.g. using Message Groups)
http://incubator.apache.org/activemq/message-groups.html
so that the JMSXGroupSeq = 0 for the last message
Then if JMSXGroupSeq is specified as a header you can call
session.commit() if the value is zero to atomically process all of the
messages.
Questions:
1) Should I use embedded or standalone brokers?
It depends on your persistence strategy and how many JMS clients you
have. If you have lots of clients or want reliable persistence I'd
recommend a stand alone broker as its a bit easier to manage and
backup etc
Also what is the
recommended way to XML-configure an embedded broker?
http://incubator.apache.org/activemq/how-do-i-embed-a-broker-inside-a-connection.html
2) Should I use a topic or a composite destination?
Queues are generally easier to work with (you can browse then and load
balance them etc). So I'd recommend using 2 queues using either
Virtual Destinations or Composite Destinations. Though if you don't
need to load balance requests on the queue then durable topics is
fine.
3) Should I use brokers on both machines with a network connector for
bridging them or have the slave app connect to the master broker?
I'd just run 1 broker - depending on your requirements maybe a
master/slave pair of brokers
http://incubator.apache.org/activemq/masterslave.html
4) Can I find out if there is an active sender/publisher for a
queue/topic (so that the slave can decide on startup whether there is
still a chance to get an "end" if he sees a "begin").
You can browse active consumers via JMX
http://incubator.apache.org/activemq/jmx.html
or receive advisories
http://incubator.apache.org/activemq/advisory-message.html
While testing, I had some problems with the "two brokers" approach:
after stopping/starting the slave broker, the slave app would no longer
receive messages although both brokers successfully reconnected. This
was even more the case with durable topics. Is this a known issue?
Not sure - it could be.
--
James
-------
http://radio.weblogs.com/0112098/
