You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Junrui Li (Jira)" <ji...@apache.org> on 2023/08/08 11:59:00 UTC
[jira] [Comment Edited] (FLINK-32768) SpeculativeSchedulerITCase.testSpeculativeExecutionOfInputFormatSource times out
[ https://issues.apache.org/jira/browse/FLINK-32768?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17751864#comment-17751864 ]
Junrui Li edited comment on FLINK-32768 at 8/8/23 11:58 AM:
------------------------------------------------------------
This issue arises because createDeploymentHandles requires a correspondence between executionsToDeploy and executionSlotAssignments ([here|https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/DefaultExecutionDeployer.java#L140]), but after FLINK-31843, SimpleExecutionSlotAllocator broke this assumption ([here|https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SimpleExecutionSlotAllocator.java#L127]). The error only occurs during schedule slow tasks by SpeculativeScheduler because AdaptiveBatchScheduler will schedule task one-by-one, while SpeculativeScheduler may schedule multiple slow tasks at once. This issue renders speculative execution completely unusable, so I will raise its priority and prepare a PR to fix it.
cc [~wanglijie] [~zhuzh]
was (Author: JIRAUSER293716):
This issue arises because createDeploymentHandles requires a correspondence between executionsToDeploy and executionSlotAssignments ([here|https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/DefaultExecutionDeployer.java#L140]), but after FLINK-31843, SimpleExecutionSlotAllocator broke this assumption ([here|https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SimpleExecutionSlotAllocator.java#L127]). The error only occurs during speculative execution because AdaptiveBatchScheduler will schedule task one-by-one, while speculative execution may schedule multiple slow task nodes at once. This issue renders speculative execution completely unusable, so I will raise its priority and prepare a PR to fix it.
cc [~wanglijie] [~zhuzh]
> SpeculativeSchedulerITCase.testSpeculativeExecutionOfInputFormatSource times out
> --------------------------------------------------------------------------------
>
> Key: FLINK-32768
> URL: https://issues.apache.org/jira/browse/FLINK-32768
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination, Tests
> Affects Versions: 1.18.0
> Reporter: Matthias Pohl
> Assignee: Junrui Li
> Priority: Blocker
> Labels: pull-request-available, test-stability
>
> SpeculativeSchedulerITCase.testSpeculativeExecutionOfInputFormatSource is timing out:
> https://dev.azure.com/apache-flink/apache-flink/_build/results?buildId=52009&view=logs&j=39d5b1d5-3b41-54dc-6458-1e2ddd1cdcf3&t=0c010d0c-3dec-5bf1-d408-7b18988b1b2b&l=9022
> {code}
> 2023-08-06T05:34:27.1867230Z Aug 06 05:34:27 "ForkJoinPool-1-worker-1" #14 daemon prio=5 os_prio=0 tid=0x00007fb7d4e82800 nid=0x6dde7 waiting on condition [0x00007fb7834a4000]
> 2023-08-06T05:34:27.1867541Z Aug 06 05:34:27 java.lang.Thread.State: WAITING (parking)
> 2023-08-06T05:34:27.1867777Z Aug 06 05:34:27 at sun.misc.Unsafe.park(Native Method)
> 2023-08-06T05:34:27.1868191Z Aug 06 05:34:27 - parking to wait for <0x00000000a77360d8> (a java.util.concurrent.CompletableFuture$Signaller)
> 2023-08-06T05:34:27.1868571Z Aug 06 05:34:27 at java.util.concurrent.locks.LockSupport.park(LockSupport.java:175)
> 2023-08-06T05:34:27.1868896Z Aug 06 05:34:27 at java.util.concurrent.CompletableFuture$Signaller.block(CompletableFuture.java:1707)
> 2023-08-06T05:34:27.1869240Z Aug 06 05:34:27 at java.util.concurrent.ForkJoinPool.managedBlock(ForkJoinPool.java:3313)
> 2023-08-06T05:34:27.1869682Z Aug 06 05:34:27 at java.util.concurrent.CompletableFuture.waitingGet(CompletableFuture.java:1742)
> 2023-08-06T05:34:27.1870022Z Aug 06 05:34:27 at java.util.concurrent.CompletableFuture.get(CompletableFuture.java:1908)
> 2023-08-06T05:34:27.1870395Z Aug 06 05:34:27 at org.apache.flink.test.scheduling.SpeculativeSchedulerITCase.executeJob(SpeculativeSchedulerITCase.java:229)
> 2023-08-06T05:34:27.1870858Z Aug 06 05:34:27 at org.apache.flink.test.scheduling.SpeculativeSchedulerITCase.testSpeculativeExecutionOfInputFormatSource(SpeculativeSchedulerITCase.java:165)
> 2023-08-06T05:34:27.1871251Z Aug 06 05:34:27 at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
> [...]
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)