[
https://issues.apache.org/jira/browse/HDFS-17444?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
xiaojunxiang updated HDFS-17444:
--------------------------------
Description:
The JornalNode JVM process is not immediately in a normal state until the
JournalSyncer thread is started.
For some management platforms such as Ambari rolling restart JournalNode, we
need a jmx metric to determine whether the JournalSyncer thread is started and
enter working state for current namespace before restarting the next
JournalNode. Otherwise, restart too quickly and more than half of JournalNodes
will be out of order, causing the NameNode to die.
Extra info: The JournalSyncer thread is started when a NameNode requests
EditLog information.
was:
The JornalNode JVM process is not immediately in a normal state until the
JournalSyncer thread is started.
For some management platforms such as Ambari rolling restart JournalNode, we
need a jmx metric to determine whether the JournalSyncer thread is started
before restarting the next JournalNode. Otherwise, restart too quickly and more
than half of JournalNodes will be out of order, causing the NameNode to die.
Extra info: The JournalSyncer thread is started when a NameNode requests
EditLog information.
> Add getJournalSyncerStatus jmx metrics, to Indicates whether the
> JournalSyncer thread for the every namespace has enter working state.
> --------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-17444
> URL: https://issues.apache.org/jira/browse/HDFS-17444
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: hdfs
> Affects Versions: 3.3.4, 3.3.6
> Reporter: xiaojunxiang
> Assignee: xiaojunxiang
> Priority: Major
>
> The JornalNode JVM process is not immediately in a normal state until the
> JournalSyncer thread is started.
> For some management platforms such as Ambari rolling restart JournalNode,
> we need a jmx metric to determine whether the JournalSyncer thread is started
> and enter working state for current namespace before restarting the next
> JournalNode. Otherwise, restart too quickly and more than half of
> JournalNodes will be out of order, causing the NameNode to die.
> Extra info: The JournalSyncer thread is started when a NameNode requests
> EditLog information.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
