could someone pls help me with the following clustering (and/or HA/backup scenario):
- N nodes in a symmetric cluster; round-robin of messages to a queue "q1".
- consumers connected to all nodes;
- non-persistent messages
- every node in the cluster has a queue depth > 0 for "q1"
- 1 of the N nodes crashes (H/W dies)
Without having a backup node for every of the N primary nodes, will the remaining cluster nodes pick up the messages that were residing on the "q1" on the node that is no longer avail? I understand there's re-balancing over an internal bridge between the N nodes but I can't tell if the above case is handled after the message have been enqueued.
I am using non-persistent messages, so in the primary/backup scenario, no need to perform data migrating withing journaling or shared disk to the backup broker; in case the above doesn't work w/out backup nodes.
Apologies if this obvious in the doc or discussion group.
I believe there's a Jira for failover on non persistent messages. But basically what u tell when you send non persistent messages is that you could avoid losing them. I suggest you send them persistently but change some of the sync parameters. You would still have pretty good performance on doing that.