KafkaListenerEndpointRegistry.stop(Runnable) does not call callback when there are no containers to stop

[wilkinsona]

Boot's tests for its Kafka auto-configuration are taking 30 seconds each to run as we have a context with a KafkaListenerEndpointRegistry that has no MessageListenerContainers. This causes it to fail to call the callback that's passed into stop(Runnable) so we have to wait for Framework's 30 second timeout to pop.

[artembilan]

Good catch, Andy!

Will fix soon...

[artembilan]

Do you need the fix until Spring Boot 1.5.x ?

[wilkinsona]

Boot 1.5 doesn't appear to be affected. For example, this test takes less than 1 second. That may be due to the tests being different in 1.5, I'm not sure.

[artembilan]

Yeah... I think so. We will back-port any way until the version which is pulled by Initializr (seems for me 1.3.x), because we haven't committed into 1.1.x for a while already...

[garyrussell]

Yeah I found and fixed this in my experimental reactive @KafkaListener because in that environment, the MLC list was empty.

https://github.com/spring-projects/spring-kafka/pull/1123/files#diff-f23a91d9c56afd55946555b6ec1a150fR331

[wilkinsona]

Thanks, both. What's your schedule for the next Spring Kafka maintenance releases? I'd like to switch to snapshots if you have releases planned before the next round of Boot releases. We have 2.2 M5 scheduled for 24 July. Given the Framework releases planned for 17 July, we'll probably have 1.5 and 2.1 releases around then as well.

[wilkinsona]

I've figured out why we're only seeing the 30 second delay in Boot 2.2.

It is due to #1102. In Boot 2.1 (Spring Kafka 2.2), isRunning() only returns true if one of the MLCs is running. As there are no MLCs in the affected tests, it returns false. As a result, stop(Runnable) is never called and the 30 second delay doesn't occur.

[garyrussell]

Thanks Andy, yes we'll have a new milestone (M4). I've scheduled it for July 23 (it's the day before Artem goes on extended PTO, so we may go a few days earlier - as long as SF stays where it is).

[garyrussell]

@wilkinsona Do you plan to change the version in s-b-dependencies? (I just hit the problem while testing a patch and I see the version is still M3). Working around it with a version override for now.

Thanks.

[wilkinsona]

Oops. Thanks for the reminder. I'd intended to move to snapshots and then got distracted by other things. I've just pushed the change: spring-projects/spring-boot@b67fedd.