[
https://issues.apache.org/jira/browse/AMBARI-12951?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yusaku Sako reopened AMBARI-12951:
----------------------------------
Reopening.
The time interval to determine that the NN edit log is too old is hard coded,
and makes it hard to do automated testing.
> Prompt user to save checkpoint before shutdown if last checkpoint is too old
> ----------------------------------------------------------------------------
>
> Key: AMBARI-12951
> URL: https://issues.apache.org/jira/browse/AMBARI-12951
> Project: Ambari
> Issue Type: Task
> Components: ambari-web
> Affects Versions: 2.1.2
> Reporter: Xi Wang
> Assignee: Xi Wang
> Priority: Critical
> Fix For: 2.1.2
>
> Attachments: AMBARI-12951.patch, AMBARI-12951.patch, Popup saying No
> JMX data .png, Popup to create new checkpoint.png
>
>
> If the admin attempts to shutdown the HDFS NameNode(s) through Ambari in
> following situations:
> * HDFS service > Stop / Restart All components,
> * HDFS summary/configs page> Restart Affected,
> * Host Details > NameNode Stop / Restart,
> * Host Details > Stop/Restart all components,
> * Stop All Services
> * Hosts Actions > All/Selected/Filtered Hosts > Hosts > Stop/Restart All
> Components
> A check should be run to see when the last HDFS checkpoint was created.
> If the checkpoint was created more than 12 hours ago, Ambari should prompt
> the user to save/create a checkpoint first.
> "The last HDFS checkpoint is older than 12 hours. Make sure that you have
> taken a checkpoint before proceeding. Otherwise, the NameNode(s) can take a
> very long time to start up."
> The check should be skipped if there's no NameNode to shut down (i.e., all
> NameNodes are already down) as a result of the operation.
> The user should be warned with instructions to create a new checkpoint, it is
> up to User if they would like to create a new checkpoint. User can always
> proceed with button "Next".
> The Active NameNode JMX is checked to figure out when the last checkpoint was
> taken. If the Active NameNode JMX is not available for whatever reason,
> check for any available NameNode JMX (Standby).
> If no JMX info is available, then something is wrong. In this case, display
> warning along the line of "Could not determine the age of the last HDFS
> checkpoint. Please ensure that you have a recent checkpoint. Otherwise, the
> NameNode(s) can take a very long time to start up."
> *Limitations*
> * NameNode shutdown that are part of any Wizard will not be taken care of in
> this RMP (e.g., NameNode HA wizard)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
