[
https://issues.apache.org/jira/browse/HIVE-20549?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16619697#comment-16619697
]
Daniel Dai commented on HIVE-20549:
-----------------------------------
In both scenario, killQuery will try to kill applicable yarn application. The
difference is how to deal with exception if the operation found but belong to
other user:
* For queryId, it is clear, if operation found but belong to other user, Hive
will throw "No privilege" exception
* For queryTag, if there's duplicate tag, we need to deal with the case some
operation belong to the user and some are not. The goal is to make
non-duplicate use case more intuitive. If Hive find at least 1 operation belong
to the user, then don't throw exception. For non-duplicate scenario, if the
operation belong to the user, cancel the operation; If the operation belong to
other user, throw "No privilege" exception.
> Allow user set query tag, and kill query with tag
> -------------------------------------------------
>
> Key: HIVE-20549
> URL: https://issues.apache.org/jira/browse/HIVE-20549
> Project: Hive
> Issue Type: Bug
> Reporter: Daniel Dai
> Assignee: Daniel Dai
> Priority: Major
> Fix For: 4.0.0, 3.2.0
>
> Attachments: HIVE-20549.1.patch, HIVE-20549.2.patch
>
>
> HIVE-19924 add capacity for replication job set a query tag and kill the
> replication distcp job with the tag. Here I make it more general, user can
> set arbitrary "hive.query.tag" in sql script, and kill query with the tag.
> Hive will cancel the corresponding operation in hs2, along with Tez/MR
> application launched for the query. For example:
> {code}
> set hive.query.tag=mytag;
> select ..... -- long running query
> {code}
> In another session:
> {code}
> kill query 'mytag';
> {code}
> There're limitations in the implementation:
> 1. No tag duplication check. There's nothing to prevent conflicting tag for
> same user, and kill query will kill queries share the same tag. However, kill
> query will not kill queries from different user unless admin. So different
> user might share the same tag
> 2. In multiple hs2 environment, kill statement should be issued to all hs2 to
> make sure the corresponding operation is canceled. When beeline/jdbc connects
> to hs2 using regular way (zookeeper url), the session will connect to random
> hs2, which might be different than the hs2 where query run on. User can use
> HiveConnection.getAllUrls or beeline --getUrlsFromBeelineSite (HIVE-20507) to
> get a list of all hs2 instances.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
