[ 
https://issues.apache.org/jira/browse/HIVE-23443?focusedWorklogId=433819&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-433819
 ]

ASF GitHub Bot logged work on HIVE-23443:
-----------------------------------------

                Author: ASF GitHub Bot
            Created on: 15/May/20 18:11
            Start Date: 15/May/20 18:11
    Worklog Time Spent: 10m 
      Work Description: pgaref commented on a change in pull request #1012:
URL: https://github.com/apache/hive/pull/1012#discussion_r425967771



##########
File path: 
llap-server/src/java/org/apache/hadoop/hive/llap/daemon/impl/TaskExecutorService.java
##########
@@ -884,10 +885,20 @@ private void finishableStateUpdated(TaskWrapper 
taskWrapper, boolean newFinishab
         taskWrapper.updateCanFinishForPriority(newFinishableState);
         forceReinsertIntoQueue(taskWrapper, isRemoved);
       } else {
-        taskWrapper.updateCanFinishForPriority(newFinishableState);
-        if (!newFinishableState && !taskWrapper.isInPreemptionQueue()) {
-          // No need to check guaranteed here; if it was false we would 
already be in the queue.
+        // if speculative task, any finishable state change should re-order 
the queue as speculative tasks are always
+        // not-guaranteed (re-order helps put non-finishable's ahead of 
finishable)
+        if (!taskWrapper.isGuaranteed()) {
+          removeFromPreemptionQueue(taskWrapper);
+          taskWrapper.updateCanFinishForPriority(newFinishableState);
           addToPreemptionQueue(taskWrapper);
+        } else {
+          // if guaranteed task, if the finishable state changed to 
non-finishable and if the task doesn't exist
+          // pre-emption queue, then add it so that it becomes candidate to 
kill
+          taskWrapper.updateCanFinishForPriority(newFinishableState);

Review comment:
       Can there be a case where we have a Guaranteed task that changes from 
non-finishable to finishable and is only part of the preemptionQueue?
   Under that scenario our code (and the old code) would remove it from 
preemptionQ and it would not be part of any other Q.
   
   From the code below it seems that this can indeed happen:
   
https://github.com/apache/hive/blob/master/llap-server/src/java/org/apache/hadoop/hive/llap/daemon/impl/TaskExecutorService.java#L776




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Issue Time Tracking
-------------------

            Worklog Id:     (was: 433819)
    Remaining Estimate: 0h
            Time Spent: 10m

> LLAP speculative task pre-emption seems to be not working
> ---------------------------------------------------------
>
>                 Key: HIVE-23443
>                 URL: https://issues.apache.org/jira/browse/HIVE-23443
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Prasanth Jayachandran
>            Assignee: Prasanth Jayachandran
>            Priority: Major
>         Attachments: HIVE-23443.1.patch, HIVE-23443.2.patch
>
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> I think after HIVE-23210 we are getting a stable sort order and it is causing 
> pre-emption to not work in certain cases.
> {code:java}
> "attempt_1589167813851_0000_119_01_000008_0 
> (hive_20200511055921_89598f09-19f1-4969-ab7a-82e2dd796273-119/Map 1, started 
> at 2020-05-11 05:59:22, in preemption queue, can finish)", 
> "attempt_1589167813851_0008_84_01_000008_1 
> (hive_20200511055928_7ae29ca3-e67d-4d1f-b193-05651023b503-84/Map 1, started 
> at 2020-05-11 06:00:23, in preemption queue, can finish)" {code}
> Scheduler only peek's at the pre-emption queue and looks at whether it is 
> non-finishable. 
> [https://github.com/apache/hive/blob/master/llap-server/src/java/org/apache/hadoop/hive/llap/daemon/impl/TaskExecutorService.java#L420]
> In the above case, all tasks are speculative but state change is not 
> triggering pre-emption queue re-ordering so peek() always returns canFinish 
> task even though non-finishable tasks are in the queue. 



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

Reply via email to