[
https://issues.apache.org/jira/browse/HADOOP-1184?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12492387
]
Hairong Kuang commented on HADOOP-1184:
---------------------------------------
Here are a few more comments:
1. The indention of the patch seems not the same as the requirement.
2. In isReplicationInProgress, in addiion to to check if neededReplications
contains a block, shall we also check if the pendingReplication queue does not
contain the block?
3. In containingNodeList, the parameter numReplicas works as an output. Is it
clearer if we create a static class as the return class contaning both a
datanode list and an integer? Also since the variable nonCommisiionedNodeList
now contains decommissioning nodes, it would be better to change its name.
4. In TestDecommission, the test case for decomissioning a node that has the
only replica of a block does not 100% guarateen that the decomissioning node
contains any block. A more determininstic testcase is better. One solution
could be first start a minicluster with one datanode and create a file, then
add one more datanode and decomission the old one.
> Decommission fails if a block that needs replication has only one replica
> -------------------------------------------------------------------------
>
> Key: HADOOP-1184
> URL: https://issues.apache.org/jira/browse/HADOOP-1184
> Project: Hadoop
> Issue Type: Bug
> Components: dfs
> Reporter: dhruba borthakur
> Attachments: decommissionOneReplica3.patch,
> decommissionOneReplica4.patch
>
>
> If the only replica of a block resides on a node being decommissioned, then
> the decommission command does not complete. The blocks do not get added to
> neededReplication because neededReplications.update() believes that the
> number of current replicas is zero.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
