You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Andy Coates (Jira)" <ji...@apache.org> on 2020/01/13 18:30:00 UTC
[jira] [Updated] (KAFKA-9416) Streams get stuck in
`PENDING_SHUTDOWN` is underlying topics deleted.
[ https://issues.apache.org/jira/browse/KAFKA-9416?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Andy Coates updated KAFKA-9416:
-------------------------------
Description:
We've noticed that if topics are deleted from under a running topology, e.g. repartition, changelog or sink topics, then the stream threads transition from `RUNNING` to `PENDING_SHUTDOWN`, but then do not transition to `ERROR`.
Likewise, if a Kafka cluster has auto topic create disabled and a topology is started where its sink topic(s) do not exist, then the topology similarly gets stuck in `PENDING_SHUTDOWN`.
Once the query is stuck in `PENDING_SHUTDOWN` any call to close the topology blocks, as per https://issues.apache.org/jira/browse/KAFKA-9398.
We would like to see Kafka Streams handle this case and correctly transition to `ERROR` state.
The root cause of this looks to be that we have a large `max.block.ms` set on the KS Producer and consumer. The producer can get stuck trying to produce to a topic that does not exist. Likewise, a consumer can get stuck trying to commit offsets for a topic that does not exist.
The cause of this is covered in more detail here: [https://github.com/confluentinc/ksql/issues/4268#issuecomment-573020281]
was:
We've noticed that if topics are deleted from under a running topology, e.g. repartition, changelog or sink topics, then the stream threads transition from `RUNNING` to `PENDING_SHUTDOWN`, but then do not transition to `ERROR`.
Likewise, if a Kafka cluster has auto topic create disabled and a topology is started where its sink topic(s) do not exist, then the topology similarly gets stuck in `PENDING_SHUTDOWN`.
Once the query is stuck in `PENDING_SHUTDOWN` any call to close the topology blocks, as per https://issues.apache.org/jira/browse/KAFKA-9398.
We would like to see Kafka Streams handle this case and correctly transition to `ERROR` state.
The cause of this is covered in more detail here: https://github.com/confluentinc/ksql/issues/4268#issuecomment-573020281
> Streams get stuck in `PENDING_SHUTDOWN` is underlying topics deleted.
> ---------------------------------------------------------------------
>
> Key: KAFKA-9416
> URL: https://issues.apache.org/jira/browse/KAFKA-9416
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Affects Versions: 2.4.0
> Reporter: Andy Coates
> Priority: Major
>
> We've noticed that if topics are deleted from under a running topology, e.g. repartition, changelog or sink topics, then the stream threads transition from `RUNNING` to `PENDING_SHUTDOWN`, but then do not transition to `ERROR`.
> Likewise, if a Kafka cluster has auto topic create disabled and a topology is started where its sink topic(s) do not exist, then the topology similarly gets stuck in `PENDING_SHUTDOWN`.
> Once the query is stuck in `PENDING_SHUTDOWN` any call to close the topology blocks, as per https://issues.apache.org/jira/browse/KAFKA-9398.
> We would like to see Kafka Streams handle this case and correctly transition to `ERROR` state.
> The root cause of this looks to be that we have a large `max.block.ms` set on the KS Producer and consumer. The producer can get stuck trying to produce to a topic that does not exist. Likewise, a consumer can get stuck trying to commit offsets for a topic that does not exist.
> The cause of this is covered in more detail here: [https://github.com/confluentinc/ksql/issues/4268#issuecomment-573020281]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)