[
https://issues.apache.org/jira/browse/HDFS-10808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15440116#comment-15440116
]
Anu Engineer commented on HDFS-10808:
-------------------------------------
[~eddyxu] Sure. I think the change needed was --
{code}
for (Map.Entry<VolumePair, DiskBalancerWorkItem> entry :
workMap.entrySet()) {
blockMover.clearExitFlag();
blockMover.copyBlocks(entry.getKey(), entry.getValue());
}
{code}
The missing line was the clearExitFlag. No real changes are needed in the
copyBlocks path.
Also we seem to introduced some unintended side effects like some lines
becoming no-op. For example,
{code}
if (!shouldRun()) {
continue;
}
{code}
Looks like the intend of the patch was to remove the setExitFlag usage in
copyBlocks, but we reintroduced that in the try \{..} finally \{..} block. I
also wanted to add a unit test with multiple volumes in a single machine where
balancing happens to check for this kind of failure in future.
> DiskBalancer does not execute multi-steps plan- redux
> -----------------------------------------------------
>
> Key: HDFS-10808
> URL: https://issues.apache.org/jira/browse/HDFS-10808
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: balancer & mover
> Reporter: Anu Engineer
> Assignee: Anu Engineer
>
> This is to redo of the fix in HDFS-10598
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
