You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-issues@hadoop.apache.org by "Ron Bodkin (Created) (JIRA)" <ji...@apache.org> on 2011/10/08 15:50:29 UTC
[jira] [Created] (HDFS-2420) improve handling of datanode timeouts
improve handling of datanode timeouts
-------------------------------------
Key: HDFS-2420
URL: https://issues.apache.org/jira/browse/HDFS-2420
Project: Hadoop HDFS
Issue Type: Improvement
Reporter: Ron Bodkin
If a datanode ever times out on a heart beat, it gets marked dead permanently. I am finding that on AWS this is a periodic occurrence, i.e., datanodes time out although the datanode process is still alive. The current solution to this is to kill and restart each such process independently.
It would be good if there were more retry logic (e.g., blacklisting the nodes but try heartbeats for a longer period before determining they are apparently dead). It would also be good if refreshNodes would check and attempt to recover timed out data nodes.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HDFS-2420) improve handling of datanode
timeouts
Posted by "Suresh Srinivas (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HDFS-2420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13123793#comment-13123793 ]
Suresh Srinivas commented on HDFS-2420:
---------------------------------------
Can you please add Affects Versions field in the jira.
Default keep alive timeout is 600 seconds. That is a long time given default heart beat timeout is 3s. Also the datanode does keep retrying to connect to namenode. So I am not sure about the issue you have reported here. Can you upload logs for the scenario in this bug.
> improve handling of datanode timeouts
> -------------------------------------
>
> Key: HDFS-2420
> URL: https://issues.apache.org/jira/browse/HDFS-2420
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Ron Bodkin
>
> If a datanode ever times out on a heart beat, it gets marked dead permanently. I am finding that on AWS this is a periodic occurrence, i.e., datanodes time out although the datanode process is still alive. The current solution to this is to kill and restart each such process independently.
> It would be good if there were more retry logic (e.g., blacklisting the nodes but try heartbeats for a longer period before determining they are apparently dead). It would also be good if refreshNodes would check and attempt to recover timed out data nodes.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira