[
https://issues.apache.org/jira/browse/YARN-8381?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
lujie updated YARN-8381:
------------------------
Description:
I started a fresh pseudo-distributed system on an node, then run a job but it
stuck. My first reaction was checking log message to local problem, but
obtaining no error message.
After reading log messages for long time, I waked up to check the node health
. The Yarn web UI showed that the nodemanager is unhealthy, due to "local-dirs
are bad: /tmp/hadoop-hduser/nm-local-dir". I reconfigure the
"{{yarn.nodemanager.disk-health-checker.max-disk-utilization-per-disk-percentage}}"
to 98% and solved this problem.
{color:#d04437}*But I still strongly recommend adding error log messages for
unhealthy nodemanger(especially startup).*{color}
was:
I started a fresh pseudo-distributed system on an node, then run a job but it
stuck. My first reaction was checking log message to local problem, but
obtaining no error message.
After reading log messages for long time, I waked up to check the node health
. The Yarn web UI showed that the nodemanager is unhealthy, due to "local-dirs
are bad: /tmp/hadoop-hduser/nm-local-dir". I reconfigure the
"{{yarn.nodemanager.disk-health-checker.max-disk-utilization-per-disk-percentage}}"
to 98% and solved this problem.
{color:#d04437}*But I still strongly recommend adding error log messages for
unhealthy nodemanger(especially for startup).*{color}
> Job got stuck while node was unhealthy, but without log messages to indicate
> such case
> --------------------------------------------------------------------------------------
>
> Key: YARN-8381
> URL: https://issues.apache.org/jira/browse/YARN-8381
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: lujie
> Priority: Major
>
> I started a fresh pseudo-distributed system on an node, then run a job but
> it stuck. My first reaction was checking log message to local problem, but
> obtaining no error message.
> After reading log messages for long time, I waked up to check the node
> health . The Yarn web UI showed that the nodemanager is unhealthy, due to
> "local-dirs are bad: /tmp/hadoop-hduser/nm-local-dir". I reconfigure the
> "{{yarn.nodemanager.disk-health-checker.max-disk-utilization-per-disk-percentage}}"
> to 98% and solved this problem.
> {color:#d04437}*But I still strongly recommend adding error log messages for
> unhealthy nodemanger(especially startup).*{color}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
