[
https://issues.apache.org/jira/browse/HIVE-14003?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15328067#comment-15328067
]
Sergey Shelukhin commented on HIVE-14003:
-----------------------------------------
I think TODO: Reviewer: things need to be investigated (and other TODOs may
need to be fixed). I don't have exact answers to most of them.
Dummy ops are used for mapjoin. Prasanth may know more about merge. The global
map does not need to be cleaned up explicitly; you can check the existing
cleanup, as far as I can tell this patch wouldn't interfere with it.
Overall, I think we should be able to interrupt the execution with interrupt
exception, so I am not sure why some comments say that having a blocking op is
a problem - shouldn't an interrupt there abort properly? Abort flag would only
be an optimization then. The only problem are other library calls that can
swallow interrupt exceptions... as long as abort flag is checked after those it
should be alright. Also we should file JIRAs to fix the code if these are
Hadoop libraries/components. At the very least they should restore the
interrupt flag. There should be not retries on interrupts, etc.
> queries running against llap hang at times - preemption issues
> --------------------------------------------------------------
>
> Key: HIVE-14003
> URL: https://issues.apache.org/jira/browse/HIVE-14003
> Project: Hive
> Issue Type: Bug
> Components: llap
> Affects Versions: 2.1.0
> Reporter: Takahiko Saito
> Assignee: Siddharth Seth
> Attachments: HIVE-14003.01.patch
>
>
> The preemption logic in the Hive processor needs some more work. There are
> definitely windows where the abort flag is completely dropped within the Hive
> processor.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
