Aljoscha Krettek created FLINK-17514:
----------------------------------------

             Summary: TaskCancelerWatchdog does not kill TaskManager
                 Key: FLINK-17514
                 URL: https://issues.apache.org/jira/browse/FLINK-17514
             Project: Flink
          Issue Type: Bug
          Components: Runtime / Task
    Affects Versions: 1.10.1, 1.11.0
            Reporter: Aljoscha Krettek
             Fix For: 1.10.1, 1.11.0


The watchdog reports a fatal error using {{taskManager.notifyFatalError(msg, 
null)}}. This should normally lead to the TaskManager being terminated. The 
code introduced in FLINK-16255 tries to look at the passed exception and will 
eventually fail with a {{NullPointerException}}, which prevents the TaskManager 
from being terminated.

Stacktrace:
{code}
2020-05-05 09:43:01,588 ERROR org.apache.flink.runtime.taskmanager.Task         
            - Task did not exit gracefully within 180 + seconds.
2020-05-05 09:43:01,588 ERROR 
org.apache.flink.runtime.taskexecutor.TaskExecutor            - Task did not 
exit gracefully within 180 + seconds.
2020-05-05 09:43:01,588 ERROR org.apache.flink.runtime.taskmanager.Task         
            - Error in Task Cancellation Watch Dog
java.lang.NullPointerException
        at 
org.apache.flink.util.ExceptionUtils.isOutOfMemoryErrorWithMessageStartingWith(ExceptionUtils.java:186)
        at 
org.apache.flink.util.ExceptionUtils.isMetaspaceOutOfMemoryError(ExceptionUtils.java:170)
        at 
org.apache.flink.util.ExceptionUtils.enrichTaskManagerOutOfMemoryError(ExceptionUtils.java:144)
        at 
org.apache.flink.runtime.taskexecutor.TaskManagerRunner.onFatalError(TaskManagerRunner.java:249)
        at 
org.apache.flink.runtime.taskexecutor.TaskExecutor$TaskManagerActionsImpl.notifyFatalError(TaskExecutor.java:1751)
        at 
org.apache.flink.runtime.taskmanager.Task$TaskCancelerWatchDog.run(Task.java:1514)
        at java.lang.Thread.run(Thread.java:748)
{code}



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

Reply via email to