[
https://issues.apache.org/jira/browse/YARN-353?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13733164#comment-13733164
]
Jian He commented on YARN-353:
------------------------------
bq. why the catch and re-throw?
The intention was leaving there for future exception handling(HA etc). could be
removed.
bq. why is everything being stored at the top level of the tree and not a
heirarchical structure i.e. attempts of a particular application stored under
that application's dir?
ZK doesn't support directory
bq. How is a connection failure to zk handled? i.e. getNewZooKeeper() throws an
exception. Does the RM fail/shutdown? Is the connection retried at a later
point?
If the connection is lost after RM already connected with ZK, a Disconnected
event will be sent, zkClient sets to null, all operations(create/delete etc.)
will wait for zkClient to be set again (verified this in cluster by stopping ZK
during the connection). But if getNewZooKeeper() itself throws an exception,
for now RM will fail and will not retry
bq.When creating a Zookeeper object, ZK apis support a base root path and all
operations are done relative to the base root path? Any reason why we are not
using that approach by initializing zk with zkRootNodePath ?
didn't know this.
bq.For deleteWithRetries, the return code of exists() could be checked if a
delete is required or not.
this depends on whether RM wants to know the delete operation succeeds or not.
> Add Zookeeper-based store implementation for RMStateStore
> ---------------------------------------------------------
>
> Key: YARN-353
> URL: https://issues.apache.org/jira/browse/YARN-353
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Hitesh Shah
> Assignee: Bikas Saha
> Attachments: YARN-353.10.patch, YARN-353.11.patch, YARN-353.1.patch,
> YARN-353.2.patch, YARN-353.3.patch, YARN-353.4.patch, YARN-353.5.patch,
> YARN-353.6.patch, YARN-353.7.patch, YARN-353.8.patch, YARN-353.9.patch
>
>
> Add store that write RM state data to ZK
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
