[
https://issues.apache.org/jira/browse/HIVE-11969?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14938940#comment-14938940
]
Gunther Hagleitner commented on HIVE-11969:
-------------------------------------------
How is this going to work with with container prewarm? If you do that async
that actually defeats the purpose of it.
Also in general I think this is a bad idea, because you're just covering up
that there's no capacity to run queries and it makes the code more complicated
because it has to handle async start. The first query you run will now hang and
it will just be less clear to the user what goes on. If the connection
establishment hangs you go look for capacity, if query hangs you don't know if
query is slow, if cluster is busy etc.
I think for instance the establishment of the connection shot up recently which
was due to Tez + ATS interaction. With this it's much harder to find such
things.
A better way to let folks do what you suggest is to connect through beeline and
configure HS2 to run a pool. That way AM and shell are decoupled and you're not
paying the overhead in the first query.
> start Tez session in background when starting CLI
> -------------------------------------------------
>
> Key: HIVE-11969
> URL: https://issues.apache.org/jira/browse/HIVE-11969
> Project: Hive
> Issue Type: Bug
> Components: Tez
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-11969.01.patch, HIVE-11969.patch
>
>
> Tez session spins up AM, which can cause delays, esp. if the cluster is very
> busy.
> This can be done in background, so the AM might get started while the user is
> running local commands and doing other things.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
