Karmada Graduates CNCF Project and Releases Version 1.19
Multi-cluster orchestration tool Karmada has achieved CNCF graduation status alongside the launch of version 1.19, which introduces enhanced scheduling capabilities for AI training workloads.
Karmada is now a graduated CNCF project. The graduation was announced in September 2026 at the KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China in Shanghai.
Alongside this status change, version 1.19 arrives with multi-component scheduling improvements for AI training workloads. This release promotes priority-based scheduling to Beta and enables it by default.
Shortages.
Karmada extends the standard Kubernetes API instead of replacing it. Existing manifests, controllers, and operational tools work on a Karmada control plane without modifications. The architecture contains a dedicated API server, a controller manager, a scheduler, and an etcd instance for state. It replaces the archived KubeFed project.
Two primary custom APIs handle configuration and deployment. The PropagationPolicy maps policies to workloads to define constraints for cluster affinity, rebalancing, and high availability across regions, availability zones, or cloud providers. The OverridePolicy permits cluster-specific configuration changes, including modifying container image prefixes or storage classes, without altering underlying resource templates. This supports hybrid cloud use cases and regional failover while helping organizations avoid vendor lock-in.