[jira] [Commented] (HIVE-22686) MapredLocalTask hang for over 9 nine hours

["philipse (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-22686?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17008220#comment-17008220 ] 

philipse commented on HIVE-22686:
---------------------------------

i have find proof to confirm it was caused by wrong decision on small table,i have no idea if we have already fixed the issue in the higher version?Can someone kindly tell me ?

 

Thanks in advance.

> MapredLocalTask hang for over 9 nine hours
> ------------------------------------------
>
>                 Key: HIVE-22686
>                 URL: https://issues.apache.org/jira/browse/HIVE-22686
>             Project: Hive
>          Issue Type: Bug
>          Components: Parser
>    Affects Versions: 1.1.1
>         Environment: HDP 2.5.2
> Hive:1.1
> [^normal_plan.xml]
> [^local_task_hangs_plan.xml] [^hadoop-080-084.dump]
>  
>            Reporter: philipse
>            Priority: Critical
>         Attachments: hadoop-080-084.dump, image-2019-12-31-16-30-14-731.png, image-2019-12-31-16-30-55-652.png, image-2019-12-31-16-31-40-304.png, image-2019-12-31-16-42-11-589.png, image-2019-12-31-16-42-20-416.png, local_task_hangs_plan.xml, normal_plan.xml
>
>
> Hi ,
> Yesterday, i  run a hive on mr task,it hangs for over nine hours,after rerun it with the same data on the same hiveserver,the task finished in 10 mins in time,i am already crazy on it.
> here is the last logs for the first try(which hangs for hours):
>   !image-2019-12-31-16-31-40-304.png!
>  
> here is the second try(which finishs in 10 mins,the local part cost 7 seconds also) :
> !image-2019-12-31-16-30-55-652.png!
>  
> the most weird thing is the plan.xml is different,the plans and jstack infomation are as attached.the cpu and memory is normal,while the process is so slow.
> !image-2019-12-31-16-42-20-416.png!
> !image-2019-12-31-16-42-11-589.png!
> My headache is:
> 1.Do the plan.xml can be different for the same hql and the same hiveserver
> 2.The first try runs quite slow,every records operation(put into hashmap) cost almost 1000ms.while the second try runs quite fast(same data and same hiveserver). 
> Do Anyone has already met similar issues, any points will be appreciated.
> Thanks.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)