[
https://issues.apache.org/jira/browse/YARN-3860?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14614105#comment-14614105
]
Karthik Kambatla commented on YARN-3860:
----------------------------------------
bq. That is true only if RM restart is enabled and ZKRMStateStore is used but
other settings should be cared.
bq. getServiceState is expected to return or fail immediately if the node is
down.
Do we support other configurations? My only concern is the case when
getServiceState fails because we can't reach the other node, but the node and
the RM are actually active. If someone uses rmadmin -transitionToActive, they
might falsely assume that a split-brain scenario is not possible.
Just to be clear, I agree the patch is benign. I want to make sure we are clear
on why we are adding it.
> rmadmin -transitionToActive should check the state of non-target node
> ---------------------------------------------------------------------
>
> Key: YARN-3860
> URL: https://issues.apache.org/jira/browse/YARN-3860
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Affects Versions: 2.7.0
> Reporter: Masatake Iwasaki
> Assignee: Masatake Iwasaki
> Fix For: 2.8.0
>
> Attachments: YARN-3860.001.patch, YARN-3860.002.patch,
> YARN-3860.003.patch
>
>
> Users can make both ResouceManagers active by {{rmadmin -transitionToActive}}
> even if {{\--forceactive}} option is not given. {{haadmin
> -transitionToActive}} of HDFS checks whether non-target nodes are already
> active but {{rmadmin -transitionToActive}} does not do.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
