[
https://issues.apache.org/jira/browse/HADOOP-4556?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12644434#action_12644434
]
Raghu Angadi commented on HADOOP-4556:
--------------------------------------
yes. E.g. :
- P and Q have a block B. Say B is over replicated to R and P is asked to
delete the block.
- Now Q and R have the block and there is a pending delete-request to P.
- Now P is marked dead and P comes back up again.
- block will become over replicated and NN asks Q to delete it.
- Now the block will be delete from both P and Q! Block is under replicated.
Does it seem right? IOW, why do we not want to remove the pending deletion
requests?
The above example shows loss of one replica unintentionally. The real example
is lot crazier in terms of how the DNs were maked dead and came back alive and
we ended up losing all the replicas. I don't know yet why the heartbeat was
lost.
Note that this in no way means we should not fix HADOOP-4540. These are two
different problems (or two parts of a problem).
> Block went missing
> ------------------
>
> Key: HADOOP-4556
> URL: https://issues.apache.org/jira/browse/HADOOP-4556
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Affects Versions: 0.17.2
> Reporter: Robert Chansler
> Assignee: Hairong Kuang
>
> Suspicion that all replicas of a block were marked for deletion. (Don't
> panic, investigation underway.)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
