[
https://issues.apache.org/jira/browse/KAFKA-19608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18101148#comment-18101148
]
Yunseop Eom commented on KAFKA-19608:
-------------------------------------
Opened PR #23040: https://github.com/apache/kafka/pull/23040
The PR adds classic and consumer group integration coverage for KAFKA-19608. It
verifies that a subscribed consumer continues consuming after the brokers swap
client listener ports and the server requests a rebootstrap, including the
group-coordinator path that was not covered by the existing manually assigned
consumer tests.
The current trunk implementation already contains the connection reset behavior
from the KIP-1242 work; this PR adds regression coverage rather than
duplicating that implementation.
Verification: both new tests pass normally and with --rerun-tasks; clients
integration-test Spotless, Checkstyle, and git diff --check pass locally.
GitHub label checks pass and the PR is awaiting maintainer review.
> REBOOTSTRAP_REQUIRED leave connections to group coordinator untouched
> ---------------------------------------------------------------------
>
> Key: KAFKA-19608
> URL: https://issues.apache.org/jira/browse/KAFKA-19608
> Project: Kafka
> Issue Type: Bug
> Components: clients
> Affects Versions: 4.0.0
> Reporter: David Dufour
> Priority: Major
>
> KIP-1102 has introduced a new error code to instruct clients to rebootstrap.
> This error has been introduced to help proxies to instruct the clients to
> reboostrap when brokers have changed.
> However, the current implementation in group consumers leave the connections
> to the group coordinator intact.
> As a result, when doing a cluster failover:
> - the clients receive REBOOTSTRAP_REQUIRED
> - the broker connections are reboostrapped, connections to the primary
> cluster are closed and connections to the secondary clusters are opened
> - the consumers are still connected to the group coordinator of the primary
> cluster
> ==> the consumers are:
> - fetching data from the secondary cluster
> - sending hearbeats to the primary cluster
> The consumers are in an inconsistent state
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
