[ 
https://issues.apache.org/jira/browse/YARN-1611?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13888391#comment-13888391
 ] 

Xuan Gong commented on YARN-1611:
---------------------------------

bq. RM_CONF_STORE is actually specific to the FSRemoteConf. Rename it.

Renamed to FS_RM_CONF_STORE

bq. Put audience and stability annotations for the new classes. Mostly keep 
them private and unstable at this point of time.

Added private and unstable annotations for RemoteConfiguration, 
FileSystemBasedRemoteConfiguration, RemoteConfigurationFactory

bq. Create the CS_CONFIGURATION_FILE constant in YarnConfiguration and not 
AdminService. It can be useful beyond, but mark it private for now.

Move CS_CONFIGURATION_FILE from AdminService to YarnConfiguration. But I need 
to use CS_CONFIGURATION_FILE as the input for refreshQueue to get the 
configuration, so mark it as public instead of private.

bq.CapacitySchedulerConfiguration: Why do you need to reloadConfiguration() 
when useRemoteConfiguration is true?

Removed

bq.After this patch is goes in, if HA is enabled, we are forcing users to put 
configs remotely. I agree this is the long term goal, but in the short term, 
can we have an internal non-public flag to enable the remote-configuration 
stuff?

Create YarnConfiguration.RM_HA_REMOTE_CONFIGURATION_ENABLED and set 
YarnConfiguration.DEFAULT_RM_HA_REMOTE_CONFIGURATION_ENABLED as false. 


> Make admin refresh of capacity scheduler configuration work across RM failover
> ------------------------------------------------------------------------------
>
>                 Key: YARN-1611
>                 URL: https://issues.apache.org/jira/browse/YARN-1611
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Xuan Gong
>            Assignee: Xuan Gong
>         Attachments: YARN-1611.1.patch, YARN-1611.2.patch, YARN-1611.2.patch, 
> YARN-1611.3.patch, YARN-1611.3.patch, YARN-1611.4.patch, YARN-1611.5.patch, 
> YARN-1611.6.patch
>
>
> Currently, If we do refresh* for a standby RM, it will failover to the 
> current active RM, and do the refresh* based on the local configuration file 
> of the active RM. 



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to