[
https://issues.apache.org/jira/browse/HADOOP-5012?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Brian Bockelman updated HADOOP-5012:
------------------------------------
Attachment: hadoop-5012.patch
Patch for problem as described.
I do have an example of this problem happening in a logfile if someone's
interested.
> addStoredBlock should take into account corrupted blocks when determining
> excesses
> ----------------------------------------------------------------------------------
>
> Key: HADOOP-5012
> URL: https://issues.apache.org/jira/browse/HADOOP-5012
> Project: Hadoop Core
> Issue Type: Bug
> Reporter: Brian Bockelman
> Attachments: hadoop-5012.patch
>
>
> I found another source of corruption on our cluster.
> 0) Three replicas of a block exist
> 1) One is recognized as corrupt (3 reps total)
> 2) Namenode decides to create a new replica. Replication done and
> addStoredBlock is called (4 reps total)
> 3) There are too many replicas, so processOverReplicatedBlock is called by
> addStoredBlock.
> 4) processOverReplicatedBlock is called, and it decides do invalidate the
> newly created replica. [Oddly enough, it decides to invalidate the newly
> created one instead of the one in the corrupted replicas map!]
> 5) We are in the same state as (1) -- 3 replicas total, 1 of which is still
> bad.
> I believe we can fix this easily -- change numCurrentReplica variable to take
> into account the number of corrupt replicas.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
