[
https://issues.apache.org/jira/browse/HDFS-6965?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14114065#comment-14114065
]
Kihwal Lee commented on HDFS-6965:
----------------------------------
I think there is a flaw in isGoodTarget(). When checking the available space,
it checks against {{node.getRemaining()}}. It has too check at individual
storage level.
> NN continues to issue block locations for DNs with full disks
> -------------------------------------------------------------
>
> Key: HDFS-6965
> URL: https://issues.apache.org/jira/browse/HDFS-6965
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 2.0.0-alpha, 3.0.0
> Reporter: Daryn Sharp
>
> Encountered issues where DNs have less space than a full block and reject
> incoming transfers. The NN continues giving out locations for these nodes
> for some period of time. It does not appear to be related to the DN's cached
> disk usage.
> One impact is required replications are delayed when a full DN is chosen for
> the pipeline. A DN cannot report a broken pipeline so the replication must
> timeout (5m) before new targets are chosen.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
