[HBASE-3142] If a master dies and comes back up before his znode expires, the RS heartbeat can lock up

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 0.89.20100924, 0.90.0
• Fix Version/s: 0.90.0
• Component/s: master, regionserver
• Assignee: ryan rawson
• Reporter: Jonathan Gray
• Created: 21/Oct/10 18:55
• Updated: 12/Oct/12 06:17
• Resolved: 09/Nov/10 02:04

Description

During a rolling restart, we ran into a case where a master was shutdown and then brought back up before the znode expired.

On the RS side, while the master was down, it was getting ConnectionRefused exceptions trying to heartbeat to what it thinks is the active master.

Once the master process comes back up, the next heartbeat done by all the RSs just blocks indefinitely.

This is somewhat related to HBASE-3141