[
https://issues.apache.org/jira/browse/KAFKA-10876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17254244#comment-17254244
]
Chris Egerton commented on KAFKA-10876:
---------------------------------------
[~xakassi] they're definitely related, but I think the two issues are slightly
different. KAFKA-7878 is about one bug that may lead to attempted duplicate
task instantiation on a worker; there are other potential bugs that could lead
to that as well. This issue isn't really centered around any of those bugs, and
more about how the framework responds when it runs into that case. Ideally, all
of those bugs would be fixed and this ticket would be irrelevant, but
until/unless we have complete confidence that that's the case, we should still
alter the behavior fo the framework to not incorrectly mark tasks as {{FAILED}}
when that happens.
> Duplicate connector/task create requests lead to incorrect FAILED status
> ------------------------------------------------------------------------
>
> Key: KAFKA-10876
> URL: https://issues.apache.org/jira/browse/KAFKA-10876
> Project: Kafka
> Issue Type: Bug
> Components: KafkaConnect
> Reporter: Chris Egerton
> Priority: Major
>
> If a Connect worker tries to start a connector or task that it is already
> running, an error will be logged and the connector/task will be marked as
> {{FAILED}}. This logic is implemented in several places:
> *
> [https://github.com/apache/kafka/blob/300909d9e60eb1d5e80f4d744d3662a105ac0c15/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/Worker.java#L257-L262]
> *
> [https://github.com/apache/kafka/blob/300909d9e60eb1d5e80f4d744d3662a105ac0c15/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/Worker.java#L299-L306]
> *
> [https://github.com/apache/kafka/blob/300909d9e60eb1d5e80f4d744d3662a105ac0c15/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/Worker.java#L511-L512]
> *
> [https://github.com/apache/kafka/blob/300909d9e60eb1d5e80f4d744d3662a105ac0c15/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/Worker.java#L570-L572]
>
> Although it's certainly abnormal for a worker to run into this case and an
> {{ERROR}}-level log message is warranted when it occurs, the connector/task
> should not be marked as {{FAILED}}, as there is still an instance of that
> connector/task still running on the worker.
>
> Either the worker logic should be updated to avoid marking connectors/tasks
> as {{FAILED}} in this case, or it should manually halt the existing
> connector/task before creating a new instance in its place. The first option
> is easier and more intuitive, but if it's ever possible that the
> already-running connector/task instance has an outdated configuration and the
> to-be-created connector/task has an up-to-date configuration, the second
> option would have correct behavior (while the first would not).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
