[ 
https://issues.apache.org/jira/browse/KYLIN-6078?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18019308#comment-18019308
 ] 

Longfei Jiang commented on KYLIN-6078:
--------------------------------------

h2. Root Cause
h2. 
Currently, partition refresh is directly executed by the KE query engine 
Sparder, without submitting a job.
As a result, there is no job-level overlap detection for the job_range.

h2. Dev Design

Adjust the logic of partition refresh to follow the normal job submission flow.

Partition refresh will submit a job, and the corresponding time partition will 
occupy the job_range, consistent with the handling of build/refresh jobs.

The job UI will display the refreshed partition range for better visibility.

> [Internal Table] job_range was occupied by another task, but the time 
> partitions within job_range were refreshed successfully.
> ------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: KYLIN-6078
>                 URL: https://issues.apache.org/jira/browse/KYLIN-6078
>             Project: Kylin
>          Issue Type: Bug
>    Affects Versions: 5.0.2
>            Reporter: Longfei Jiang
>            Assignee: Longfei Jiang
>            Priority: Major
>
> *Description*
> When loading data into an internal table, the job_range is already occupied 
> by a running task. However, a refresh job for an overlapping time partition 
> still succeeds, which is not the expected behavior.
> *Steps to Reproduce*
> 1. Load data for 1996-01-04 into table A (succeeds).
> 2. Submit another job to load data for the range 1996-01-03 ~ 1996-01-31 (job 
> is running).
> 3. While step 2 is still in progress, submit a refresh job for 1996-01-04.
> 4. The refresh job for 1996-01-04 succeeds.
> *Expected Behavior*
> If the job_range overlaps with an existing running task, the new job should 
> not be allowed to run (task submission should be rejected).
> *Actual Behavior*
> The refresh job for 1996-01-04 succeeds even though the job_range overlaps 
> with a running job.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to