ambari-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nate Cole (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (AMBARI-20736) Allow Potentially Long Running Restart Commands To Have Their Own Timeout
Date Tue, 11 Apr 2017 19:14:41 GMT

     [ https://issues.apache.org/jira/browse/AMBARI-20736?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Nate Cole updated AMBARI-20736:
-------------------------------
    Status: Patch Available  (was: Open)

> Allow Potentially Long Running Restart Commands To Have Their Own Timeout
> -------------------------------------------------------------------------
>
>                 Key: AMBARI-20736
>                 URL: https://issues.apache.org/jira/browse/AMBARI-20736
>             Project: Ambari
>          Issue Type: Bug
>          Components: ambari-server
>            Reporter: Nate Cole
>            Assignee: Nate Cole
>            Priority: Critical
>             Fix For: 2.5.1
>
>         Attachments: AMBARI-20736.patch
>
>
> During an upgrade of a cluster, some commands are expected to take a very long time depending
on what the size of the cluster is and how much data is stored. For example, a NameNode restart
with SafeMode exit may take in excess of 30 minutes. On some clusters, this could take less
than 1 minute.
> Currently today, the only way to adjust these properties is to do so across the board
for all commands by editing {{ambari.properties}} and setting {{agent.task.timeout}}. This
solution doesn't work very well since the majority of restarts during an upgrade are not on
a master component.
> There needs to be a way to instruct Ambari that a restart should be allowed to run for
a relatively long period of time. 
> - Both Java and Python need to be considered here. We don't want Python to give up and
return a {{FAILED}} state and we don't want Ambari server to set the task to {{TIMEDOUT}}.
> - This can be useful in both normal restarts and upgrade scenarios. 
> h3. Upgrade Only
> If considering this functionality in the context of an upgrade only, then it is conceivable
that this logic can be placed inside of the upgrade XML packs:
> {code}
>     <service name="HDFS">
>       <component name="NAMENODE">
>         <upgrade>
>           <task xsi:type="restart-task"  timeout="1800"/>
>         </upgrade>
> {code}
> - This would allow future mpacks to be able to control the restart of components. Perhaps
this can even be slightly abstracted out:
> {code}
>     <service name="HDFS">
>       <component name="NAMENODE">
>         <upgrade>
>           <task xsi:type="restart-task"  timeout="upgrade.parameter.master.restart.long"/>
>         </upgrade>
> upgrade.parameter.slave.restart.short = 300
> upgrade.parameter.slave.restart.long = 900
> upgrade.parameter.master.restart.short = 1500
> upgrade.parameter.master.restart.long = 1800
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message