[
https://issues.apache.org/jira/browse/HDFS-5070?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13729985#comment-13729985
]
Kihwal Lee commented on HDFS-5070:
----------------------------------
By making SafeModeMonitor the one who always initializes the replication
queues, the long delay may be avoided. Going through addStoredBlock() for an
initial block report while in an initial safe mode with initialized replication
queues seems expensive.
> Do not initialize the replications queues in the middle of block report
> processing
> ----------------------------------------------------------------------------------
>
> Key: HDFS-5070
> URL: https://issues.apache.org/jira/browse/HDFS-5070
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 2.1.0-beta, 0.23.9
> Reporter: Kihwal Lee
>
> While processing an initial block report in the start-up safe mode, namenode
> can reach the safe block threshold in the middle of processing the report.
> This is noticed when checkMode() is called and it causes the replication
> queues to be initialized.
> The safe mode monitor will try to check and leave the safe mode, but can be
> far behind the write lock, if the initialization takes long (e.g. large
> number of blocks) and more block reports come in and get queued before it.
> In this state (replication queue initialized but still in startup safe mode),
> block report processing can take a long time. In one instance, 4 block report
> processing took 13 minutes.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
