[
https://issues.apache.org/jira/browse/KAFKA-9398?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17013108#comment-17013108
]
Matthias J. Sax commented on KAFKA-9398:
----------------------------------------
The root cause seems to be https://issues.apache.org/jira/browse/KAFKA-3539 –
ie, because `send()` might block, `StreamsThread` is stuck and thus cannot call
`producer.close()`.
> Kafka Streams main thread may not exit even after close timeout has passed
> --------------------------------------------------------------------------
>
> Key: KAFKA-9398
> URL: https://issues.apache.org/jira/browse/KAFKA-9398
> Project: Kafka
> Issue Type: Improvement
> Components: streams
> Reporter: Bill Bejeck
> Assignee: Bill Bejeck
> Priority: Critical
> Fix For: 2.5.0
>
>
> Kafka Streams offers the KafkaStreams.close() method when shutting down a
> Kafka Streams application. There are two overloads to this method, one that
> takes no parameters and another taking a Duration specifying how long the
> close() method should block waiting for streams shut down operations to
> complete. The no-arg version of close() sets the timeout to Long.MAX_VALUE.
> The issue is that if a StreamThread is taking to long to complete or if one
> of the Consumer or Producer clients is in a hung state, the Kafka Streams
> application won't exit even after the specified timeout has expired.
> For example, consider this scenario:
> # A sink topic gets deleted by accident
> # The user sets Producer max.block.ms config to a high value
> In this case, the Producer will issue a WARN logging statement and will
> continue to make metadata requests looking for the expected topic. The
> {{Producer}} will continue making metadata requests up until the max.block.ms
> expires. If this value is high enough, calling close() with a timeout won't
> fix the issue as when the timeout expires, the Kafka Streams application's
> main thread won't exit.
> To prevent this type of issue, we should call Thread.interrupt() on all
> StreamThread instances once the close() timeout has expired.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)