[Discuss] Error interrupt LoadRetriever loop

Hi, AutoMQ community, an Error will causes the loop to exit, maybe need catch it let the task continue or provide metrics for task state.

Detail see,

automq/core/src/main/java/kafka/autobalancer/LoadRetriever.java

Lines 431 to 435 in baf30f0

    
           } catch (Exception e) { 
        
               logger.error("Consumer poll error", e); 
        
               this.mainExecutorService.schedule(this::retrieve, 1, TimeUnit.SECONDS); 
        
               return; 
        
           }

@Wzy19930507 good point, although it's unlikely to be able to continue the task when any Error occurs during runtime (such as OOM or stackoverflow), it's worth log it for diagnose.

BTW pls let us know if you're insterested in solving this issue by yourself (so I can assign it to you), new contributor is always welcomed in AutoMQ community

Why is there a problem with this code? schedule has made it reconsume after 1 second.

Why is there a problem with this code? schedule has made it reconsume after 1 second.

The current implementation only catches Exception, not Error

Why is there a problem with this code? schedule has made it reconsume after 1 second.

The current implementation only catches Exception, not Error

Could there be an exception in this logic that would throw a non-inherited Exception?

KafkaException unified inheritance to RuntimeException

Why is there a problem with this code? schedule has made it reconsume after 1 second.

The current implementation only catches Exception, not Error

Could there be an exception in this logic that would throw a non-inherited Exception?

It's possible, although very rare, to throw an Error, which is an inheritance of Throwable, for example OutOfMemoryError

BTW pls let us know if you're insterested in solving this issue by yourself (so I can assign it to you), new contributor is always welcomed in AutoMQ community

I'm insterested, please assign it to me, thanks!

@SCNieh Hello, Please look at it for me, i run ./gradlew core:test failed with some exception.

when quorum=kraft

java.lang.IllegalStateException: AwsCredentialsProvider has not been initialized
	at com.automq.s3shell.sdk.auth.CredentialsProviderHolder.getAwsCredentialsProvider(CredentialsProviderHolder.java:25)
	at org.apache.kafka.controller.QuorumController.<init>(QuorumController.java:2098)
	at org.apache.kafka.controller.QuorumController$Builder.build(QuorumController.java:484)
	at kafka.server.ControllerServer.startup(ControllerServer.scala:296)

BrokerApiVersionsCommandTest.checkBrokerApiVersionCommandOutput, when quorum=zk

expected: <	DeleteStreams(504): UNSUPPORTED,> but was: <	OpenStreams(502): UNSUPPORTED,>
Expected :	DeleteStreams(504): UNSUPPORTED,
Actual   :	OpenStreams(502): UNSUPPORTED,

Did I miss some configuration?

@SCNieh Hello, Please look at it for me, i run ./gradlew core:test failed with some exception.

when quorum=kraft

java.lang.IllegalStateException: AwsCredentialsProvider has not been initialized
	at com.automq.s3shell.sdk.auth.CredentialsProviderHolder.getAwsCredentialsProvider(CredentialsProviderHolder.java:25)
	at org.apache.kafka.controller.QuorumController.<init>(QuorumController.java:2098)
	at org.apache.kafka.controller.QuorumController$Builder.build(QuorumController.java:484)
	at kafka.server.ControllerServer.startup(ControllerServer.scala:296)

BrokerApiVersionsCommandTest.checkBrokerApiVersionCommandOutput, when quorum=zk

expected: <	DeleteStreams(504): UNSUPPORTED,> but was: <	OpenStreams(502): UNSUPPORTED,>
Expected :	DeleteStreams(504): UNSUPPORTED,
Actual   :	OpenStreams(502): UNSUPPORTED,

Did I miss some configuration?

@Wzy19930507 Some of the unit tests from Apacha Kafka has not yet been made compatible with AutoMQ and we're still working on it. You can try something like ./gradlew core:S3UnitTest to run tests that're related to AutoMQ under core module, or ./gradlew s3stream:test to run tests under s3stream

It's possible, although very rare, to throw an Error, which is an inheritance of Throwable, for example OutOfMemoryError

Some unrecoverable exceptions, I believe, should not continue the loop. Of course, we can wait to see how they are handled after the PR is submitted before making a decision.

	} catch (Exception e) {
	logger.error("Consumer poll error", e);
	this.mainExecutorService.schedule(this::retrieve, 1, TimeUnit.SECONDS);
	return;
	}