[
https://issues.apache.org/jira/browse/HIVE-12935?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15143729#comment-15143729
]
Gopal V commented on HIVE-12935:
--------------------------------
Tested this on the cluster for a bit of scale testing and it flexes up/down
fine.
I'm chaining this after HIVE-12967, so the LlapConfiguration class disappears
(but easy to fix).
+1 after the UUID change, but I'll rebase after the config changes go in.
[~sseth] left comments on RB on chaining the listeners for the registry, the
thread-safety for callbacks & possibly about race conditions when an instance
dies, while a task is still scheduled on it.
The API chaining & splitting up the conf out HS2 (hive.llap.zookeeper.*) can be
new JIRAs, but the race-conditions/thread-safety need a closer look before this
goes in tomorrow.
> LLAP: Replace Yarn registry with Zookeeper registry
> ---------------------------------------------------
>
> Key: HIVE-12935
> URL: https://issues.apache.org/jira/browse/HIVE-12935
> Project: Hive
> Issue Type: Improvement
> Affects Versions: 2.0.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Attachments: 12935.1.patch, HIVE-12935.2.patch, HIVE-12935.3.patch
>
>
> Existing YARN registry service for cluster membership has to depend on
> refresh intervals to get the list of instances/daemons that are running in
> the cluster. Better approach would be replace it with zookeeper based
> registry service so that custom listeners can be added to update healthiness
> of daemons in the cluster.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
