[
https://issues.apache.org/jira/browse/KAFKA-17627?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17887235#comment-17887235
]
Spacrocket commented on KAFKA-17627:
------------------------------------
I checked how to implement this and I think the code just requires swapping
restartConnector with restartConnectorAndTasks method that has a parameter
delayMs.
I implemented part of it but before PR I still need to write and repair tests
and finish TODO: mentioned in the commit. I will post it here just in case my
approach is totally incorrect.
[https://github.com/apache/kafka/commit/c8d688b8afb01dc9f865a3ffaa6e55e13609e13b]
I am not really sure whether in standalone mode this is also required but I
think it wouldn't hurt and should keep the implementation easier to read.
Task description mentions:
_Instead, when secrets for a task include a TTL, the task should be restarted
directly._ Why should connector be skipped when it's a part that communicates
with sink/source?
> ConfigProvider TTLs do not restart Tasks
> ----------------------------------------
>
> Key: KAFKA-17627
> URL: https://issues.apache.org/jira/browse/KAFKA-17627
> Project: Kafka
> Issue Type: Bug
> Components: connect
> Affects Versions: 2.0.0
> Reporter: Greg Harris
> Assignee: Spacrocket
> Priority: Major
> Labels: newbie
>
> The ConfigProvider interface allows for implementations to provide
> configurations with an accompanying Time To Live, a lifetime that the
> returned value is valid for. Callers which receive a TTL should periodically
> call the ConfigProvider to refresh and receive any updated value.
> The WorkerConfigTransformer is responsible for accepting TTLs from
> ConfigProviders, and scheduling restarts for connectors to force them to
> refresh their externalized configurations.
> Since the introduction of the ConfigProvider interface in 2.0.0, when the
> WorkerConfigTransformer processes task configurations, it schedules a restart
> of the connector, which may or may not be present on the running machine.
> Instead, when secrets for a task include a TTL, the task should be restarted
> directly.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
