Radhika Kundam created ATLAS-5341:
-------------------------------------
Summary: Kafka notification messages permanently dropped during
Atlas HA failover
Key: ATLAS-5341
URL: https://issues.apache.org/jira/browse/ATLAS-5341
Project: Atlas
Issue Type: Bug
Components: atlas-core
Affects Versions: 3.0.0
Reporter: Radhika Kundam
Assignee: Radhika Kundam
The error handling refactoring introduced in *ATLAS-5320* results in the
following failure scenario.
Hive hook notifications corresponding to missing entities can encounter a
transient {{NullPointerException}} or {{IllegalStateException}} during an HA
failover. These exceptions occur when JanusGraph is temporarily unavailable
(for example, during the {{ACTIVE → PASSIVE}} transition).
*Previous behavior*
* These exceptions were treated as transient failures.
* The Kafka offset was *not* committed.
* The notification was automatically retried once the service recovered,
making the failure self-healing.
*Current behavior (after ATLAS-5320)*
* The Kafka offset is committed despite the transient exception.
* The Hive hook notification is skipped permanently.
* As a result, the corresponding entity is never created in Atlas, leading to
missing entities and metadata inconsistency.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
