Chris Egerton created KAFKA-17130:
-------------------------------------
Summary: Connect workers do not properly ensure group membership
before responding to health checks
Key: KAFKA-17130
URL: https://issues.apache.org/jira/browse/KAFKA-17130
Project: Kafka
Issue Type: Bug
Components: connect
Affects Versions: 3.8.0, 3.9.0
Reporter: Chris Egerton
Initially reported [here|https://github.com/apache/kafka/pull/16585].
When a distributed Connect worker's herder begins an iteration of its tick
loop, it tries to ensure that the worker is still in contact with the Kafka
cluster that's used for cluster coordination and internal topics; see
[here|https://github.com/apache/kafka/blob/0ada8fac6869cad8ac33a79032cf5d57bfa2a3ea/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/distributed/DistributedHerder.java#L419].
However, this method may return even if the Kafka cluster is down. It does not
force a heartbeat request to be sent to the broker, and may return if the time
since the last heartbeat is small enough.
We may want to force at least one request (possibly, specifically a heartbeat)
to the group coordinator to have been sent before returning from
{{WorkerGroupMember::ensureActive}} in order to guarantee that the health check
point only returns 200 if it has explicitly validated the health of the
worker's connection to the group coordinator after the request to the endpoint
was initiated.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
