[
https://issues.apache.org/jira/browse/FLINK-16510?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17154416#comment-17154416
]
Stephan Ewen commented on FLINK-16510:
--------------------------------------
The current approach in Flink is the following:
- fatal errors initiall call System.exit(), which allows shutdown hooks to
run and to clean up stuff (local files)
- one of the Shutdown hooks has a timer that calls Runtime.halt() if the JVM
is not terminated after 10 seconds.
The change you proposed would mean that Flink never attempts to clean up.
For some reason, the shutdown hooks seem to not work reliably on some JVM /
Kernel versions. Could you confirm whether you might be affected by the same
issue as FLINK-17470 ?
If yes, could we try and find another way to go for the "graceful exit first,
hard exit if unsuccessful" approach?
> 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
>
> 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)
