[
https://issues.apache.org/jira/browse/HIVE-12046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14946785#comment-14946785
]
Xuefu Zhang commented on HIVE-12046:
------------------------------------
Look at the patch again and it seems that it handles the case where the remote
spark client is in a bad state when trying to submit a spark job. This is good.
However, it's unclear that what's going to happen when either
getDefaultParallelism() or getExecutorCount() is called in such a situation.
Also even in case that exectue() is called, the remote client can become bad
right after the isActive() check.
Therefore, I think we need to define a scope for this JIRA. If we want to be
resilient to the connection loss, then we need to consider more cases and the
way to handle it. However, it's also acceptable in my opinion that we can
detect the error and ask user to log out and log in again to get a valid
session. The latter seems simpler and easier.
> Re-create spark client if connection is dropped
> -----------------------------------------------
>
> Key: HIVE-12046
> URL: https://issues.apache.org/jira/browse/HIVE-12046
> Project: Hive
> Issue Type: Bug
> Components: Spark
> Reporter: Jimmy Xiang
> Assignee: Jimmy Xiang
> Priority: Minor
> Fix For: 1.3.0, 2.0.0
>
> Attachments: HIVE-12046.1.patch
>
>
> Currently, if the connection to the spark cluster is dropped, the spark
> client will stay in a bad state. A new Hive session is needed to re-establish
> the connection. It is better to auto reconnect in this case.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
