[
https://issues.apache.org/jira/browse/HDFS-5958?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13904891#comment-13904891
]
Tsz Wo (Nicholas), SZE commented on HDFS-5958:
----------------------------------------------
The balancing policy assumes that there are enough blocks for moving around.
In your case, it may be impossible to satisfy the percentage threshold
requirement for the large the datanode since it remains under utilized even if
it has a replica for all the blocks.
> One very large node in a cluster prevents balancer from balancing data
> ----------------------------------------------------------------------
>
> Key: HDFS-5958
> URL: https://issues.apache.org/jira/browse/HDFS-5958
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: balancer
> Affects Versions: 2.2.0
> Environment: Hadoop cluster with 4 nodes: 3 with 500Gb drives and one
> with 4Tb drive.
> Reporter: Alexey Kovyrin
>
> In a cluster with a set of small nodes and one much larger node balancer
> always selects the large node as the target even though it already has a copy
> of each block in the cluster.
> This causes the balancer to enter an infinite loop and stop balancing other
> nodes because each balancing iteration selects the same target and then could
> not find a single block to move.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
