[JAMES-3721] Browse on the Pulsar mailqueue fails with ObjectNotFoundException

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.7.0
• Component/s: pulsar, Queue
• Labels: Pulsar
• External issue URL: https://issues.apache.org/jira/browse/JAMES-3695
• Assignee: Unassigned
• Reporter: Jean Helou
• Created: 04/Mar/22 10:02
• Updated: 10/Mar/22 12:03
• Resolved: 10/Mar/22 08:38

Description

We have identified a couple interrelated issues around the puslar mailqueue :

Upon acknowledging a message, the corresponding mime message was not removed from the blob store.

Upon fixing that issue (and adding the corresponding verification), we started getting more failures on the browse (the failures could also be observed on the remove).

 

Because of how pulsar works and because we allow parallel processing of messages, the browse operation can end up reading meta data of messages that have been removed or acknowledged. The metadata of acknowledged messages is not immediately  purged from pulsar. It is somewhat configurable but for a normally operating cluster acked messages can remain readable for hours (and forcing that to the order of seconds would badly affect the cluster operation)

 

We chose to catch the specific exception and simply ignore the corresponding mails during the browse operation.

Issue Links

• links to: GitHub Pull Request #905