[
https://issues.apache.org/jira/browse/YARN-6736?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16119478#comment-16119478
]
Rohith Sharma K S commented on YARN-6736:
-----------------------------------------
Thinking on approach, primarily we can decide to publish into both timeline
server using configuration. Currently we have *yarn.timeline-service.version*
which creates corresponding TimelineClients. How about using same configuration
with comma separated values to start both clients from RM i.e
*yarn.timeline-service.version=1.5,2.0*?
Let's say, before upgrade RM is running with 1.5 which starts only V1 timeline
client. After upgrade based on the configuration, RM starts both V1 and V2
clients as long as admin explicitly change timeline version to 2.
Point to be discussed are
# should we expose new configuration rather than using modifying existing
configuration? Something like *yarn.timeline-service.upgrade-mode=true*. When
this set to true, nevertheless of timeline version, RM will start and
publishing to both v1 and v2.
# When to stop publishing to v1? Does admin decides it? If so, should we
provide a REST interface to disable runtime or need to restart RM by changing
upgrade-mode configuration to false?
> Consider writing to both ats v1 & v2 from RM for smoother upgrades
> ------------------------------------------------------------------
>
> Key: YARN-6736
> URL: https://issues.apache.org/jira/browse/YARN-6736
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Vrushali C
> Assignee: Rohith Sharma K S
>
> When the cluster is being upgraded from atsv1 to v2, it may be good to have a
> brief time period during which RM writes to both atsv1 and v2. This will help
> frameworks like Tez migrate more smoothly.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
