Hello all,
I’ve got a policy based JMS cluster set up and everytime a broker enters or leaves the cluster all the connections are closed and I’m forced to have my code reconnect to the cluster. I’ve been assured that this is the appropriate behavior.
What I was wondering was whether anyone had experienced any issues regarding timing or message loss or other instabilities associated with the need to reconnect?
Thanks.
Hi Daniel,
You would have created two Broker servers (A, B) that are connected using JMS Cluster, and two other Broker servers (C, D) - one for each A & B that are connected using Broker territory. Later you would define a cluster connection factory & then a JNDI… Now when you define this JNDI, there is an option of defining ‘Failover URL for JNDI’. The actual URL would be pointing to Broker Server C and fail-over url would be pointing to Broker Server D…
Whenever there is a failure in one node, cluster will realize that a node is down, and JNDI fail-over URL comes into place… In order to switch from first to second URL, it takes few seconds where it disables all JMS connections, points to fail-over URL and enables all JMS connections back.
Lets say, you have some data in first broker which went down, and now the failover is pointing the second broker, what happens to the data in first broker? When it comes up again, yes the data will be recovered if it is guaranteed. But what if it doesn’t come up, or it took days to come up for some reason? Business can’t wait till then. Hence, you would need a passive broker server for Broker Server A as well as B. This will take between 1 to 2 mins to bring the Passive broker up and data will be processed without further delay…
-Senthil