unsubscribing a durable subscriber does not properly clean up undelivered messages

• Key: AMQ-2052
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Major
• Assignee: David Jencks
• Reporter: David Jencks
• Component/s: Message Store
• Affects Version/s: 5.3.0
• Fix Version/s: 5.3.0

Description

Working on the tck I found a lot of messages like
2008-12-25 11:37:08,567 ERROR [RecoveryListenerAdapter] Message id ID:pb.local-56923-1230232035140-2:13:1:1:1 could not be recovered from the data store - already dispatched
in the log. I can reproduce this with the attached test case (It could probably be in a better module, but I started looking here.)

To see the error put a breakpoint at RecoveryListenerAdapter line 61 and run the test. The required steps to reproduce seem to be:

1. create a durable subscriber on a topic.
2. send a message to the topic.
3. close the subscriber and unsubscribe.
4. create a new durable subscriber on the same topic.

Comments

[David Jencks]

Test case to deomonstrate the problem. Note that the test passes, you have to put a breakpoint to see the problem.

[David Jencks]

The actual problem causing this is confusion over when to use a file-system-safened ID. Fixed in rev 730596.

I'm also changing the RecoveryListenerAdapter to throw an exception when it gets into this inconsistent state. Rev 730598.

I may try to simplify when and where the file system safening occurs. I also plan to move the test case to a better location.

[David Jencks]

Code inspection shows there are still a lot more confusions about when to use a file system safened name.

I think rev 730614 is an OK fix.... still need the test case.

[David Jencks]

One more major file name problem... rev 730817. Surprisingly this only caused 13 test failures.

[Jim Taylor]

What is the work around after the store gets into this state? Is there one?

[Gary Tully]

David, are you done with this or is there more work needed to get the tck passing?

[David Jencks]

I've fixed all the problems I can find related to this so I'm closing this.