[
https://issues.apache.org/jira/browse/FLINK-4973?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15965884#comment-15965884
]
Andrey commented on FLINK-4973:
-------------------------------
Was able to reproduce in 1.2.0.
It looks like components shutdown in wrong sequence. For example
SystemProcessingTimeService has:
{code}
@Override
public void shutdownService() {
if (status.compareAndSet(STATUS_ALIVE, STATUS_SHUTDOWN) ||
status.compareAndSet(STATUS_QUIESCED,
STATUS_SHUTDOWN))
{
timerService.shutdownNow();
}
}
{code}
Which means it won't wait while tasks terminate. And cause the exception above
in the following sequence:
* SystemProcessingTimeService call shutdown now and continue termination
sequence.
** OutputFlusher thread got termination event
* LocalBufferPool got termination request before OutputFlusher
* Exception in logs.
There are several ways to fix that:
* Await thread pool termination
* Properly handle InterruptedException. The following catch is not a correct
way of handling InterruptedException.
{code}
catch (InterruptedException e) {
// ignore on close
}
{code}
Here is good article on how to handle such exceptions:
https://www.ibm.com/developerworks/library/j-jtp05236/index.html
> Flakey Yarn tests due to recently added latency marker
> ------------------------------------------------------
>
> Key: FLINK-4973
> URL: https://issues.apache.org/jira/browse/FLINK-4973
> Project: Flink
> Issue Type: Bug
> Components: Tests
> Affects Versions: 1.2.0
> Reporter: Till Rohrmann
> Assignee: Till Rohrmann
> Priority: Critical
> Labels: test-stability
> Fix For: 1.2.0
>
>
> The newly introduced {{LatencyMarksEmitter}} emits latency marker on the
> {{Output}}. This can still happen after the underlying {{BufferPool}} has
> been destroyed. The occurring exception is then logged:
> {code}
> 2016-10-29 15:00:48,088 INFO org.apache.flink.runtime.taskmanager.Task
> - Source: Custom File Source (1/1) switched to FINISHED
> 2016-10-29 15:00:48,089 INFO org.apache.flink.runtime.taskmanager.Task
> - Freeing task resources for Source: Custom File Source (1/1)
> 2016-10-29 15:00:48,089 INFO org.apache.flink.yarn.YarnTaskManager
> - Un-registering task and sending final execution state
> FINISHED to JobManager for task Source: Custom File Source
> (8fe0f817fa6d960ea33f6e57e0c3891c)
> 2016-10-29 15:00:48,101 WARN
> org.apache.flink.streaming.api.operators.AbstractStreamOperator - Error
> while emitting latency marker
> java.lang.RuntimeException: Buffer pool is destroyed.
> at
> org.apache.flink.streaming.runtime.io.RecordWriterOutput.emitLatencyMarker(RecordWriterOutput.java:99)
> at
> org.apache.flink.streaming.api.operators.AbstractStreamOperator$CountingOutput.emitLatencyMarker(AbstractStreamOperator.java:734)
> at
> org.apache.flink.streaming.api.operators.StreamSource$LatencyMarksEmitter$1.run(StreamSource.java:134)
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471)
> at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:304)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:178)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
> at java.lang.Thread.run(Thread.java:745)
> Caused by: java.lang.IllegalStateException: Buffer pool is destroyed.
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBuffer(LocalBufferPool.java:144)
> at
> org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBlocking(LocalBufferPool.java:133)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:118)
> at
> org.apache.flink.runtime.io.network.api.writer.RecordWriter.randomEmit(RecordWriter.java:103)
> at
> org.apache.flink.streaming.runtime.io.StreamRecordWriter.randomEmit(StreamRecordWriter.java:104)
> at
> org.apache.flink.streaming.runtime.io.RecordWriterOutput.emitLatencyMarker(RecordWriterOutput.java:96)
> ... 9 more
> {code}
> This exception is clearly related to the shutdown of a stream operator and
> does not indicate a wrong behaviour. Since the yarn tests simply scan the log
> for some keywords (including exception) such a case can make them fail.
> Best if we could make sure that the {{LatencyMarksEmitter}} would only emit
> latency marker if the {{Output}} would still be active. But we could also
> simply not log exceptions which occurred after the stream operator has been
> stopped.
> https://s3.amazonaws.com/archive.travis-ci.org/jobs/171578846/log.txt
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
