[
https://issues.apache.org/jira/browse/HIVE-15217?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15691194#comment-15691194
]
Prasanth Jayachandran commented on HIVE-15217:
----------------------------------------------
I tested the patch with 1s refresh interval on a 2 node llap cluster. The way
status report is implemented right now is, the tool first connects with yarn to
get application status, then to slider to get container status and finally to
llap registry to get daemon status which involves connection to zookeeper. Even
for 2 node llap setup refreshing every second, the cost of connection is high
and the status updates are not instant. If we keep the refresh interval at 1s
we will be creating/closing too many connections in short time with not much
updates to report. Ideally what we want the tool to do is just connect to
zookeeper and watch for new events under the llap namespace. With that we have
have a single connection to registry and watch for notifications until the
desired state is achieved. We are not there yet :)
> Add watch mode to llap status tool
> ----------------------------------
>
> Key: HIVE-15217
> URL: https://issues.apache.org/jira/browse/HIVE-15217
> Project: Hive
> Issue Type: Improvement
> Components: llap
> Affects Versions: 2.2.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Priority: Minor
> Attachments: HIVE-15217.1.patch, HIVE-15217.2.patch
>
>
> There is few seconds overhead for launching the llap status command. To avoid
> we can add "watch" mode to llap status tool that refreshes the status after
> configured interval.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
