[openamq-dev] Persistence

[Pieter Hintjens]

Carl,

> Are there any plans to support persistent messaging within OpenAMQ at some
> stage?

Our view of reliability is somewhat different from that of other AMQP
implementations.  The classic design is to implement persistence in
the broker, and to use transactions between the clients and the broker
to govern how messages are persisted.

Doing this properly also means implementing clustering with full
failover.  This gets very difficult, and it's one of the reasons the
AMQP spec is getting fatter and more complex that I'd like.

Also, classic reliability does not work across multi-broker
architectures; that is, you can lose messages once your network gets
more complex, even if you have fully-reliable brokers in most of it.

So our architecture - which is still being worked on - treats the
brokers as unreliable, and does end-to-end persistence from client to
client.  This is how we implemented reliability in our production work
for JPMC, it's much simpler than using persistence in the broker.

Overall, OpenAMQ is targetting the very high performance sector, and
we'll be releasing support for multicast and end-to-end reliability
when - as Martin says - we get concrete clients for these
technologies.

-Pieter