History
Origin
Kthena grew out of Volcano, the CNCF batch scheduler. Volcano traces back to kube-batch, which added gang scheduling to Kubernetes, was open-sourced at KubeCon Shanghai in 2019, entered the CNCF Sandbox in April 2020, and was promoted to Incubating on 2022-03-21 (CNCF blog).
Volcano's scheduling was built for AI training. Kthena extends it to inference so the project covers the full AI lifecycle rather than training alone. The Kthena repository was created on 2025-05-08, and the Volcano community formally announced Kthena on 2026-01-28, with Huawei Cloud (Volcano's originator) leading the effort (CNCF blog).
Kthena is not a standalone CNCF project. It is a subproject under Volcano, so its maturity follows Volcano's Incubating status.
Timeline
| Year | Milestone |
|---|---|
| 2019 | kube-batch open-sourced at KubeCon Shanghai, later renamed Volcano |
| 2020 | Volcano accepted into the CNCF Sandbox (April) |
| 2022 | Volcano promoted to CNCF Incubating (2022-03-21) |
| 2025 | Kthena repository created (2025-05-08); v0.1.0 released 2025-10-31; v0.2.0 released 2025-12-10 |
| 2026 | Kthena announced by the Volcano community (2026-01-28); v0.3.0 (2026-01-31); v0.4.0 (2026-04-21) |
How it evolved
Kthena shipped four releases in roughly six months. v0.1.0 landed on 2025-10-31, v0.2.0 on 2025-12-10, v0.3.0 on 2026-01-31 alongside the public announcement, and v0.4.0 on 2026-04-21. The pinned commit affd5be is the main HEAD from 2026-06-24, about two months past v0.4.0, and carries no tag of its own.
By March 2026 the Volcano community described an AI-native unified scheduling platform bundling Volcano v1.14, Kthena v0.3.0, and AgentCube, positioning Kthena as the inference-serving member of that platform (Beyond Batch).
Where it stands now
Releases have been frequent, roughly every one to two months from v0.1.0 to v0.4.0. The project is young and pre-1.0. Development is led by Huawei Cloud under the Volcano community (CNCF blog). The stated direction is to make Volcano's gang scheduling and Kthena's inference orchestration parts of one AI-native scheduling platform (Beyond Batch).