An implementation of Flinks HighAvailabilityServicesFactory for clusters running on Kubernetes.
This project is currently work in progress.
Assumptions made:
- Only one JobManager is running at a time (achieved through K8s 'Recreate' deployment strategy)
To use multiple JobManagers the Kubernetes Leader Election mechanism (Fabric8 Example) could be implemented.
Not yet implemented:
- CompletedCheckpointStore (currently only in-memory)
- CheckpointIDCounter (currently only in-memory)
The helm-chart-flink repo is used for local Flink + S3 deployment.