Lianet Magrans created KAFKA-18034:
--------------------------------------
Summary: CommitRequestManager should fail pending requests on
fatal coordinator errors
Key: KAFKA-18034
URL: https://issues.apache.org/jira/browse/KAFKA-18034
Project: Kafka
Issue Type: Bug
Components: clients, consumer
Reporter: Lianet Magrans
The commit request manager requires a known coordinator to send requests fetch
and commit offsets (sync and async).
If there is no known coordinator, the commit manager simply returns no requests
on poll, but it keeps pending requests in hand, to be send as soon as there is
a known coordinator. This ensures that on recoverable errors the requests are
sent. But if there is a fatal error looking up a coordinator (ie.
GroupAuthorizationFailed), the commit manager does not fail the pending
requests it has in hand. This means that api calls that generate commit/fetch
requests will not be aware of the error.
We should consider if there is a fatal error with the coordinator when there is
a commitManager poll that finds not coordinator
[https://github.com/apache/kafka/blob/50c15b94c94fbe8f964703c057963b38100b0bd6/clients/src/main/java/org/apache/kafka/clients/consumer/internals/CommitRequestManager.java#L181]
In that case, we should fail all pendingRequests to ensure that the calling API
funcs know about the error. This should allow us to enable auth tests that are
currently failing for the new consumer due to this gap (expecting
GroupAuthException on consumer api calls). Ex.
https://github.com/apache/kafka/blob/50c15b94c94fbe8f964703c057963b38100b0bd6/core/src/test/scala/integration/kafka/api/AuthorizerIntegrationTest.scala#L1277
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
