Guillaume,
Thank you for your quick reply. Sorry to be so dense, but I just want
to confirm my understanding of what you've said:
If I configure jms flow only with activemq journaled db persistence,
I would get recovery of message-exchanges that were 'in-flight' at
the time of a server crash, but I would NOT get exchanges that had
already hit a particular component, correct?
If I configure jca flow, I will, in addition to above, be able to get
exchanges that were in the middle being processed by a component at
the time of the crash, yes? Is this automatic, or must my components
commit or rollback the transaction? Given that I am using extensions
of servicemix' ComponentSupport, how would I get access to the
transaction to do so?
Thanks again for all your help.
BJ
On May 16, 2006, at 11:47 AM, Guillaume Nodet wrote:
On 5/16/06, William Blackburn <[EMAIL PROTECTED]> wrote:
I will be moving from a single deployment, SEDA flow to a cluster
using the JMSflow. I've run the cluster example and studied up on
ActiveMQ, My goal is as much fault tolerance as load handling,
Configuring ActiveMQ's journaled db persistence will get me most of
the way (I think), but I still have two remaining questions.
First, in the MQ sense, when is an message-exchange considered
'delivered' ? For example, if I have a bunch of servicemix components
running on a node configured as above, and I yank the power while a
component is in the middle of an 'onMessageExchange()' what happens
to that message? Will ActiveMQ redeliver it when the node comes back
up? Can components influence what ActiveMQ considers 'delivered' in
any way.
To be able to use redelivery, the flow should be able to use
acknowledgements, jms transactions or xa transactions. Currenly there
is no callback to be able to use client acks or jms transaction, so
you will have to use xa transactions. Note that there is no real
overhead if the jms resource is the only one enlisted in the xa
transactions. The JCA flow provides high performance concurrent
delivery using. The configuration is the same than for the jms flow.
My second question is probably more an ActiveMQ one, but I hope you
can comment anyway. We will be using a redundant database-server as
well. Should the ActiveMQ brokers be configured to use the same
database? Or should each have its own database?
You should either use master / slave, or have a shared database if you
want high availability, else messages kept by a broker are not
available to other brokers, so if this one fails, the message will be
lost.
Cheers,
Guillaume Nodet
Thanks very much in advance for any help.
B.J.
--
Cheers,
Guillaume Nodet
