[
https://issues.apache.org/jira/browse/FLINK-18983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17248736#comment-17248736
]
Yuan Mei edited comment on FLINK-18983 at 12/14/20, 6:32 AM:
-------------------------------------------------------------
Hey [~liuyufei], thanks for reporting! just to confirm some more details:
# In the case where the task blocks on the close function, if the task fails,
the failed task will block until the close function finishes, and then report
the failure to the JM. Then JM fails the job accordingly instead of *"job will
block at close action and never change to FAILED"*, is that right?
# It is kind of the contract that the close function is called and finished
before exiting, hence I think the close function, in general, is not expected
to do blocking work. May I ask why you would like to block the close function?
And if you indeed want to do this, would you be able to set a timeout in the
close function yourself?
# Cancelation depends on many different things, so it is reasonable to have a
timeout bound.
What do you think?
was (Author: ym):
Hey [~liuyufei], thanks for reporting! just to confirm some more details:
# In the case where the task blocks on the close function, if the task fails,
the failed task will block until the close function finishes, and then report
the failure to the JM. Then JM fails the job accordingly instead of *"job will
block at close action and never change to FAILED"*, is that right?
# It is kind of the contract that the close function is called and finished
before exiting, hence I think the close function, in general, is not expected
to do blocking work. May I ask why you would like to block the close function?
And if you indeed want to do this, would you be able to set a timeout in the
close function yourself?
# Cancelation depends on many different things, so it is reasonable to have a
timeout bound.
> Job doesn't changed to failed if close function has blocked
> -----------------------------------------------------------
>
> Key: FLINK-18983
> URL: https://issues.apache.org/jira/browse/FLINK-18983
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Task
> Affects Versions: 1.11.0, 1.12.0
> Reporter: YufeiLiu
> Priority: Major
>
> If a operator throw a exception, it will break process loop and dispose all
> operator. But state will never switch to FAILED if block in Function.close,
> and JobMaster can't know the final state and do restart.
> Task have {{TaskCancelerWatchDog}} to kill process if cancellation timeout,
> but it doesn't work for FAILED task.TAskThread will allways hang at:
> org.apache.flink.streaming.runtime.tasks.StreamTask#cleanUpInvoke
> Test case:
> {code:java}
> Configuration configuration = new Configuration();
> configuration.set(TaskManagerOptions.TASK_CANCELLATION_TIMEOUT, 10000L);
> StreamExecutionEnvironment env =
> StreamExecutionEnvironment.createLocalEnvironment(2, configuration);
> env.addSource(...)
> .process(new ProcessFunction<String, String>() {
> @Override
> public void processElement(String value, Context ctx,
> Collector<String> out) throws Exception {
> if (getRuntimeContext().getIndexOfThisSubtask() == 0) {
> throw new RuntimeException();
> }
> }
> @Override
> public void close() throws Exception {
> if (getRuntimeContext().getIndexOfThisSubtask() == 0) {
> Thread.sleep(10000000);
> }
> }
> }).setParallelism(2)
> .print();
> {code}
> In this case, job will block at close action and never change to FAILED.
> If change thread which subtaskIndex == 1 to sleep, TM will exit after
> TASK_CANCELLATION_TIMEOUT.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
