Bowen Li created FLINK-40710:
--------------------------------
Summary: Dynamic Kafka Source assigns duplicate partition owners
after local failover
Key: FLINK-40710
URL: https://issues.apache.org/jira/browse/FLINK-40710
Project: Flink
Issue Type: Bug
Components: Connectors / Kafka
Affects Versions: 2.3.0
Reporter: Bowen Li
Assignee: Bowen Li
After checkpoint recovery with GLOBAL assignment, a local reader failure before
the next checkpoint can leave the same Kafka partition assigned to multiple
readers. Returned assignments and restored reader reports are not reconciled
consistently.
A deterministic test with 2 readers and 11 partitions produces 19
reader/partition assignments after failover.
Expected:
Exactly one owner per physical partition.
Impact:
Concurrent duplicate consumption and potentially unrestorable checkpoints
containing duplicate assignments.
This was reproduced in a downstream build. The recovery failure persists with
PR305 / FLINK-40615, which addresses a separate discovery race.
We need to build the solution upon Flink-40615 / PR 305.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
