Matthias J. Sax created KAFKA-20981:
---------------------------------------
Summary: Kafka Streams should handle unknown status code
pro-actively
Key: KAFKA-20981
URL: https://issues.apache.org/jira/browse/KAFKA-20981
Project: Kafka
Issue Type: Bug
Components: streams
Affects Versions: 4.2.0
Reporter: Matthias J. Sax
With KIP-1071, task assignment moved to the broker, and
StreamsGroupHeartbeatResponse now carries status codes. The status field is
present since v0. Adding a new status code is a semantic contract change: it
must be backed by a KIP and a request-version bump, and the broker must gate
the new code so it is only sent to clients that advertise the introducing
version (as it does today for MISSING_CLIENT_TAGS, code 6, which is withheld
from v0 clients). This is subtler than a schema change and easy to get wrong,
so a buggy broker can send a code an older client does not recognize.
The client decodes every status on every heartbeat via Status.fromCode, which
throws on an unknown code. Today that throw is silently swallowed
(https://issues.apache.org/jira/browse/KAFKA-20860): the client keeps
heartbeating, advances its member epoch, and silently ignores the assignment
and every one after it, while the coordinator sees a healthy member. Because
the status is returned on every heartbeat, the condition persists for the
member's lifetime; once https://issues.apache.org/jira/browse/KAFKA-20860 is
addressed so the error surfaces, it would recur on every heartbeat unless
handled here.
We propose to verify the status code pro-actively, and if the client cannot
recognize it, raise a Streams exception that shuts down the StreamThread and
triggers the uncaught-exception handler, so the error surfaces cleanly.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
