[ACCUMULO-2359] compact range hanging indefinately

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Cannot Reproduce
• Affects Version/s: 1.4.4
• Component/s: tserver

Environment

large production cluster

• Assignee: Unassigned
• Reporter: Eric C. Newton
• Created: 12/Feb/14 18:42
• Updated: 18/Jul/14 03:45
• Resolved: 18/Jul/14 03:45

Description

A user pointed out to me that he's seen a range compaction hang.

These compactions are run every day. 99% of the time they complete successfully. But twice, they have hung.

The client is dutifully waiting for the RPC to the master to return.
The master does not have a FATE transaction for the compaction, but the client's thread is waiting for a compaction. There were no table locks.

There are no errors in the master.

A tablet server was known to have died around the time the client requested the compaction. The tablets were all reassigned and recovered.

There were no splits added to the tablet at any time during the request (or after).

There's another very large table on the same system that is range compacted almost continuously, and that has never hung.