[
https://issues.apache.org/jira/browse/HBASE-4001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13052256#comment-13052256
]
Jean-Daniel Cryans commented on HBASE-4001:
-------------------------------------------
Not sure which HDFS version you are running, but in recent ones the DFSClient
just marks the blocks on the dead DN as bad and queries the other ones.
> when region has been OPENING for too long, the master should not reassign
> region
> --------------------------------------------------------------------------------
>
> Key: HBASE-4001
> URL: https://issues.apache.org/jira/browse/HBASE-4001
> Project: HBase
> Issue Type: Bug
> Reporter: Liyin Tang
> Assignee: Liyin Tang
>
> When the region server R1 and data node D1 running in the same machine and
> let's assuem this machine pub into repair,
> the master will try to reassign all the regions in the region server to other
> region server R2. But at that time, the name node hasn't figure out the bad
> data node D1.
> So the region server R2 will try to open the region from the D1, which will
> failed and retried. Then the master believes it took TOO LONG to open this
> region, so it reassign to R3...
> The story continues until the name node figure out the bad datanode D1 and
> finally region server Rn opens the region and do the compaction for the store
> file.
> All previous region servers cannot the region later and find the file doesn't
> exist since it has been compacted.
> So the solution is that when region has been OPENING for too long, the master
> should not reassign region.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
