History

Origin

Karmada grew out of the multi-cluster work led by Huawei Cloud, co-initiated with a group of large-scale users: First Automobile Works, ICBC, SPD Bank, Qutoutiao, VIPKid, and xiaohongshu. The project positions itself as the successor to the deprecated Kubernetes KubeFed (federation v2) effort, keeping Kubernetes-native APIs while adding independent propagation and override policies plus cross-cluster scheduling (CNCF blog).

Timeline

Year	Milestone
2021	Project launched, co-initiated by Huawei Cloud and several large users.
2021	Accepted as a CNCF Sandbox project (2021-09-14).
2023	Promoted to CNCF Incubating (2023-12-12).
2025	Formal Adopter Group program launched (2025-03).
2026	v1.18.0 released as the recent stable line; v1.19.0-alpha.0 cut on master.

How it evolved

Karmada moved from a basic federation tool toward an automation-heavy control plane. The scheduler gained the ability to divide replicas across clusters by static weight or by real available capacity, backed by the karmada-scheduler-estimator and karmada-descheduler for dynamic rebalancing. The Lua-based Resource Interpreter Framework was added so arbitrary CRDs can be taught to Karmada without recompiling, and the repo now ships third-party interpreters for Flux, Argo, Ray, Kubeflow, and Flink. An operator/ was introduced to manage Karmada instances declaratively through a Karmada CRD. These shifts are visible in the codebase under pkg/scheduler, pkg/resourceinterpreter, and operator/ (karmada-io/karmada).

Where it stands now

At the documented commit (658499d, 2026-06-22), Karmada is a CNCF Incubating project with active release cadence: v1.18.0 is the recent stable release (2026-05-30) and v1.19.0-alpha.0 is in progress. At the time of incubation CNCF reported more than 500 contributors from 60+ organizations across 20+ countries, with 7 maintainers. As of 2026-06-24 the GitHub repo shows 5,503 stars and 1,149 forks (CNCF blog, GitHub API).