[
https://issues.apache.org/jira/browse/HDFS-16062?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17360670#comment-17360670
]
JiangHua Zhu commented on HDFS-16062:
-------------------------------------
We found a phenomenon that when some disks on the DataNode are unavailable, if
the reconfig command is executed, it will take a long time for the DataNode to
process the bad disks (DataNode#removeVolumes()), DataNode#removeVolumes() uses
synchronization control ( synchronized).
When the jmx request is executed on the DataNode, the clusterId information is
obtained (DataNode#getClusterId()), and DataNode#getClusterId() also uses
synchronized control (synchronized). Therefore, DataNode#getClusterId() will
block for a long time.
> When a DataNode hot reload configuration, JMX will block for a long time
> ------------------------------------------------------------------------
>
> Key: HDFS-16062
> URL: https://issues.apache.org/jira/browse/HDFS-16062
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: JiangHua Zhu
> Assignee: JiangHua Zhu
> Priority: Major
>
> When executing the reconfig command on the DataNode, for example:
> hdfs dfsadmin -reconfig datanode hadoopxxxx.jd.163.org:8025 start
> At this time, the JMX of the DataNode is requested. At this time, JMX will
> still block after a long time.
> When this happens, it is mainly caused by some disk exceptions (such as disk
> unavailability) on the DataNode. At this time, DataNode#JMX will block.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
