[CONNECTORS-279] Postgresql load test job delete document cleanup fails sometimes and leaves orphaned documents

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: ManifoldCF 0.4
• Fix Version/s: ManifoldCF 0.4
• Component/s: Framework crawler agent
• Assignee: Karl Wright
• Reporter: Karl Wright
• Created: 21/Oct/11 06:15
• Updated: 28/Oct/11 12:30
• Resolved: 28/Oct/11 12:30

Description

Running the postgresql load test on my laptop, I was surprised when the test did not finish. The UI indicated that the job was being deleted, but there were 49,000 documents and that number was not moving. Further inspection yielded the following:

• Job was in the "DELETING" state
• Documents were in the "BEINGDELETED" state
• No activity of any kind ongoing

The log had no errors.

It was impossible to get a thread dump, but a cursory inspection of the code indicated that either the documents were being marked as "BEINGDELETED" but were not actually being placed on the in-memory queue, or the delete threads were picking up the documents and somehow avoiding marking them as being processed.

Also, probably unrelated, the Document Status report listed these documents as having a status of "Being removed" and a state of "Unknown". The "Unknown" should have been a "Deleting". Since the extended WHEN... ELSE clause has a reasonable condition for the "Deleting" answer, it's hard to see how this could have occurred either.