This chapter discusses the reliability features that you can expect out of Onyx. These features are tuneable and require the developer to make trade-offs.
Table of Contents generated with DocToc
Onyx offers transactional processing semantics per batch. Messages are moved from one HornetQ server to the other using non-XA transacted sessions. Messages are eligible for replay if there is machine failure in between the time a batch has been read and committed back to storage.
In the case of a HornetQ server kicking the bucket, you're likely going to want to be able to recover the messages that were lost and continue processing without it. To defend against data loss, you stand up one or more replicas per HornetQ server instance. This will raise the overall latency per message processed. If you're okay with losing some data for some reason, you can run each server without any replication for as low as possible latency. See the section on running a production Onyx environment for information on staging replicas.