[HBASE-10833] Region assignment may fail during cluster start up

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.92.3, 0.98.1, 0.99.0
• Component/s: Region Assignment
• Hadoop Flags: Reviewed
• Assignee: Jeffrey Zhong
• Reporter: Jeffrey Zhong
• Created: 25/Mar/14 21:13
• Updated: 21/Feb/15 23:29
• Resolved: 26/Mar/14 23:28

Description

This is an intermittent & infrequent issue. It happens when a cluster is starting up, only when one region server is available for region assignment temporally and master has a transient issue to talk to the only RS

When the RPC layer has a hiccup to talk to a RS, the RS will be put in failed server list and it will stay in failed server list for 2 sec(default value).

While the 20 region retires completes much shorter than 2 secs, the end result is a failed assignment.

Below are the logging for the total time spent on 20 assignment retires, it only took about 36 ms for all the retries.

2014-03-24 18:14:43,451 WARN  [AM.ZK.Worker-pool2-t59] master.AssignmentManager: Failed assignment of hbase:labels,,1395668310177.f7372ede6c8bd7de4e91bfeda884cffb. to hor15n18.gq1.ygridcore.net,60020,1395684489232, trying to assign elsewhere instead; try=1 of 20
….
2014-03-24 18:14:43,487 WARN  [AM.ZK.Worker-pool2-t59] master.AssignmentManager: Failed assignment of hbase:labels,,1395668310177.f7372ede6c8bd7de4e91bfeda884cffb. to hor15n18.gq1.ygridcore.net,60020,1395684489232, trying to assign elsewhere instead; try=20 of 20