[ 
https://issues.apache.org/jira/browse/YARN-1029?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13862013#comment-13862013
 ] 

Karthik Kambatla commented on YARN-1029:
----------------------------------------

bq. Initialized in AdminService? It can be initialize in ResourceManager class 
itself and it can access AdminService via RMContext.
We initially had it in the RM, but thought AdminService is a better place. 
http://tinyurl.com/qdo2vos

bq. Why is ZK added to hadoop-yarn-client module? It should be only in 
server-common?
TestRMFailover needs it.

bq. yarn-default.xml: "This kind of failover is embedded in the RM and does not 
explicitly fence stores.” - “does not” or “does”?
The elector doesn't explicitly fence (as in the way HDFS does), it is implicit 
and the store is supposed to ensure a single RM can modify it at any point in 
time.

bq. I think we should force admins to set yarn.resourcemanager.cluster-id 
explicitly (only in case HA is enabled for now). Defaults don’t tend to be 
changed and a default cluster-id can potentially cause hard-to-debug issues.
I am okay either way, but I think the fewer configs we *force* admins to set 
the better. If there is a single cluster, it should be perfectly okay to just 
use the default. No? 

Will address remaining suggestions. 

> Allow embedding leader election into the RM
> -------------------------------------------
>
>                 Key: YARN-1029
>                 URL: https://issues.apache.org/jira/browse/YARN-1029
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Bikas Saha
>            Assignee: Karthik Kambatla
>         Attachments: embedded-zkfc-approach.patch, yarn-1029-0.patch, 
> yarn-1029-0.patch, yarn-1029-1.patch, yarn-1029-2.patch, yarn-1029-3.patch, 
> yarn-1029-4.patch, yarn-1029-5.patch, yarn-1029-6.patch, yarn-1029-7.patch, 
> yarn-1029-7.patch, yarn-1029-8.patch, yarn-1029-9.patch, 
> yarn-1029-approach.patch
>
>
> It should be possible to embed common ActiveStandyElector into the RM such 
> that ZooKeeper based leader election and notification is in-built. In 
> conjunction with a ZK state store, this configuration will be a simple 
> deployment option.



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

Reply via email to