[
https://issues.apache.org/jira/browse/KAFKA-8485?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16858367#comment-16858367
]
kaushik srinivas commented on KAFKA-8485:
-----------------------------------------
I have a query in this scenario, KafkaBasedLog.java uses producers to update
connect config topic within kafka, and this producer object is created at the
runtime bootup.
If the leader broker of the config topic goes down after everything is working
fine, would it mean the metadata which is present with this producer is no more
valid and we hit issues with this producer ?
> Kafka connect worker does not respond when kafka broker goes down with data
> streaming in progress
> -------------------------------------------------------------------------------------------------
>
> Key: KAFKA-8485
> URL: https://issues.apache.org/jira/browse/KAFKA-8485
> Project: Kafka
> Issue Type: Bug
> Components: KafkaConnect
> Affects Versions: 2.2.1
> Reporter: kaushik srinivas
> Priority: Blocker
> Labels: performance
>
> Below is the scenario
> 3 kafka brokers are up and running.
> Kafka connect worker is installed and a hdfs sink connector is added.
> Data streaming started, data being flushed out of kafka into hdfs.
> Topic is created with 3 partitons, one leader on all the three brokers.
> Now, 2 kafka brokers are restarted. Partition re balance happens.
> Now we observe, kafka connect does not respond. REST API keeps timing out.
> Nothing useful is being logged at the connect logs as well.
> Only way to get out of this situation currently is to restart the kafka
> connect worker and things gets normal.
>
> The same scenario when tried without data being in progress, works fine.
> Meaning REST API does not get into timing out state.
> making this issue a blocker, because of the impact due to kafka broker
> restart.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
