wilmerdooley opened a new pull request, #28493: URL: https://github.com/apache/flink/pull/28493
When a Flink Session cluster is deployed on Kubernetes, the TaskManager uses the value returned by `getNamespacedServiceName` (and its related helpers) to resolve the JobManager RPC endpoint via Pekko. The current implementation produces a hostname of the form `<service>.<namespace>`, which is not a fully qualified Kubernetes service DNS name and is treated as an opaque host by Pekko. For namespaces whose name begins with a digit (e.g. `1234-ns`), DNS resolution fails and the TaskManager logs `Actor not found` while trying to connect to the dispatcher. This change appends the `.svc` segment to the namespaced service name in `ExternalServiceDecorator`, `InternalServiceDecorator`, and `KubernetesUtils`, so that the resulting hostname is the standard Kubernetes service FQDN `<service>.<namespace>.svc`. The corresponding unit test in `KubernetesClusterDescriptorTest` is updated to match the new format. ##### Was generative AI tooling used to co-author this PR? - [X] Yes (please specify the tool below) JIRA: https://issues.apache.org/jira/browse/FLINK-39788 Generated-by: Claude Code -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
