[
https://issues.apache.org/jira/browse/FLINK-16510?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17163813#comment-17163813
]
Maximilian Michels commented on FLINK-16510:
--------------------------------------------
[~chesnay] Here a stack dump from the stuck task manager: [^stack2-1.txt].
Note the common pattern that after an OOM, all the threads get stuck in native
methods like {{sun.misc.Unsafe.defineAnonymousClass}} or
{{java.lang.Class.getDeclaredConstructors0}}. There does not seem a graceful
way to get out of this situation other than immediately halting the JVM.
I would like to introduce an option to always forcefully terminate the task
manager in case of unrecoverable errors. That's what we have been doing and
it's the only way we have been able to reliably operate Flink in our k8s
deployments in case of OOM errors.
> Task manager safeguard shutdown may not be reliable
> ---------------------------------------------------
>
> Key: FLINK-16510
> URL: https://issues.apache.org/jira/browse/FLINK-16510
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Task
> Reporter: Maximilian Michels
> Assignee: Maximilian Michels
> Priority: Major
> Attachments: stack2-1.txt
>
>
> The {{JvmShutdownSafeguard}} does not always succeed but can hang when
> multiple threads attempt to shutdown the JVM. Apparently mixing
> {{System.exit()}} with ShutdownHooks and forcefully terminating the JVM via
> {{Runtime.halt()}} does not play together well:
> {noformat}
> "Jvm Terminator" #22 daemon prio=5 os_prio=0 tid=0x00007fb8e82f2800
> nid=0x5a96 runnable [0x00007fb35cffb000]
> java.lang.Thread.State: RUNNABLE
> at java.lang.Shutdown.$$YJP$$halt0(Native Method)
> at java.lang.Shutdown.halt0(Shutdown.java)
> at java.lang.Shutdown.halt(Shutdown.java:139)
> - locked <0x000000047ed67638> (a java.lang.Shutdown$Lock)
> at java.lang.Runtime.halt(Runtime.java:276)
> at
> org.apache.flink.runtime.util.JvmShutdownSafeguard$DelayedTerminator.run(JvmShutdownSafeguard.java:86)
> at java.lang.Thread.run(Thread.java:748)
> Locked ownable synchronizers:
> - None
> "FlinkCompletableFutureDelayScheduler-thread-1" #18154 daemon prio=5
> os_prio=0 tid=0x00007fb708a7d000 nid=0x5a8a waiting for monitor entry
> [0x00007fb289d49000]
> java.lang.Thread.State: BLOCKED (on object monitor)
> at java.lang.Shutdown.halt(Shutdown.java:139)
> - waiting to lock <0x000000047ed67638> (a java.lang.Shutdown$Lock)
> at java.lang.Shutdown.exit(Shutdown.java:213)
> - locked <0x000000047edb7348> (a java.lang.Class for java.lang.Shutdown)
> at java.lang.Runtime.exit(Runtime.java:110)
> at java.lang.System.exit(System.java:973)
> at
> org.apache.flink.runtime.taskexecutor.TaskManagerRunner.terminateJVM(TaskManagerRunner.java:266)
> at
> org.apache.flink.runtime.taskexecutor.TaskManagerRunner.lambda$onFatalError$1(TaskManagerRunner.java:260)
> at
> org.apache.flink.runtime.taskexecutor.TaskManagerRunner$$Lambda$27464/1464672548.accept(Unknown
> Source)
> at
> java.util.concurrent.CompletableFuture.uniWhenComplete(CompletableFuture.java:774)
> at
> java.util.concurrent.CompletableFuture$UniWhenComplete.tryFire(CompletableFuture.java:750)
> at
> java.util.concurrent.CompletableFuture.postComplete(CompletableFuture.java:488)
> at
> java.util.concurrent.CompletableFuture.completeExceptionally(CompletableFuture.java:1990)
> at
> org.apache.flink.runtime.concurrent.FutureUtils$Timeout.run(FutureUtils.java:943)
> at
> org.apache.flink.runtime.concurrent.DirectExecutorService.execute(DirectExecutorService.java:211)
> at
> org.apache.flink.runtime.concurrent.FutureUtils.lambda$orTimeout$11(FutureUtils.java:361)
> at
> org.apache.flink.runtime.concurrent.FutureUtils$$Lambda$27435/159015392.run(Unknown
> Source)
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> at java.util.concurrent.FutureTask.run(FutureTask.java:266)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$201(ScheduledThreadPoolExecutor.java:180)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748)
> Locked ownable synchronizers:
> - <0x00000006d5e56bd0> (a
> java.util.concurrent.ThreadPoolExecutor$Worker)
> {noformat}
> Note that under this condition the JVM should terminate but it still hangs.
> Sometimes it quits after several minutes.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
