Design issues in Java client failover
-------------------------------------
Key: QPID-2876
URL: https://issues.apache.org/jira/browse/QPID-2876
Project: Qpid
Issue Type: Bug
Components: Java Client
Reporter: Rajith Attapattu
Assignee: Rajith Attapattu
Fix For: 0.7
The java client keeps AMQP commands in it's replay buffer in case it needs to
replay them during failover.
However this could cause several issues. For example,
1. The failover code at the JMS layer recreates subscriptions after failing
over to a new node. However if there were message subscriptions present in the
replay buffer and were replayed after failing over it can cause the
subscription attempt by the JMS layer to fail if using exclusive subscriptions.
2. If there were ExchangeQuery/QueueQuery (or any command that expects a
response) gets replayed, it will cause the broker to respond, but the JMS
client may not have the command in it's internal map that manages
request-response coordination. This can cause errors. See QPID-2861
Since we are not implementing full session resume for the 0-10 client, I don't
think there is any advantage in replaying anything other than the message
transfers.
The JMS client should also attempt to shrink it's in doubt window by
periodically calling flush to get the broker to confirm the completion of the
commands.
This will prevent in message being duplicated unnecessarily.
There maybe a potential bug where the Java client may not be removing commands
even after the broker confirming them. I will create a new JIRA if this is
confirmed.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
