LoginSign Up
← Back to all sparks
Kubernetes logo

Kubernetes

DEVOPSINFRA · APIS
Velocity5.0

Container orchestration

kubernetes.io ↗

Kubernetes is rebuilding its core scheduling and hardware model around AI workloads.

ai-infrastructureschedulinghardware-accelerationstoragesecurityoperability
◆Current state
Kubernetes is mid-pivot from a general container orchestrator toward the default substrate for AI/ML and batch compute. Recent releases center on hardware-aware scheduling — Dynamic Resource Allocation reached GA, and workload-aware gang scheduling with a new PodGroup API landed in v1.36 — alongside storage features tuned for stateful and AI pipelines. Operational and security hardening (PSI metrics GA, CVE record corrections, externalIPs deprecation) round out the cadence.
◆Where it's heading
The center of gravity is GPU/accelerator scheduling and multi-node batch workloads. Expect the Workload/PodGroup APIs to mature from alpha toward beta, DRA's ecosystem of drivers and tooling to thicken, and storage work (Volume Health, COSI) to follow AI data-gravity needs. The security posture is shifting from patch-everything toward documenting and mitigating architectural risk.
◆Prediction
Next releases will likely promote the workload-aware scheduling APIs past alpha and expand DRA device-failure handling, with etcd 3.7 moving from beta to a final release that removes the last v2store dependencies.

◆Recent moves

  1. 3d ago

    Spotlight on WG Device Management

    A working-group spotlight recapping Dynamic Resource Allocation's road to GA — the clearest signal yet that hardware-aware scheduling is now core Kubernetes, not an add-on. It frames GPUs, TPUs, and FPGAs as first-class scheduling citizens via the ResourceSlice/ResourceClaim model rather than opaque integer counts.

    View source ↗
  2. 12d ago

    Spotlight on SIG Storage

    An interview-format spotlight on SIG Storage; no release itself, but it surfaces the storage roadmap (VolumeGroupSnapshot GA, Changed Block Tracking beta, COSI, Volume Health) that increasingly tracks AI and stateful-workload data needs.

    View source ↗
  3. 26d ago

    From Kubernetes Dashboard to Headlamp: Understanding the Transition

    The official Kubernetes Dashboard is archived, with Headlamp positioned as the recommended visual UI — adding multi-cluster views, application-centric Projects, and a plugin model. A genuine successor handoff rather than a cosmetic refresh, and a notable end-of-life for a tool that onboarded many users.

    View source ↗
  4. 1mo ago

    Reconciling the Past: Correcting Records for Unfixed Kubernetes CVEs

    The Security Response Committee is correcting CVE records for three older, unfixed issues to drop incorrect 'fixed version' tags, meaning scanners will begin flagging them where they previously didn't. It signals a maturing, transparency-first security posture that documents architectural risk instead of implying patches exist.

    View source ↗
  5. 1mo ago

    Announcing etcd 3.7.0-beta.0

    ⚡ SPARK

    etcd — the datastore under every Kubernetes control plane — ships its 3.7 beta with the long-requested RangeStream RPC and the full removal of the legacy v2store, making this the first 100% v3store release. A foundational dependency taking a breaking step toward leaner, more predictable handling of large resultsets.

    View source ↗
  6. 1mo ago

    Kubernetes v1.36: New Metric for Route Sync in the Cloud Controller Manager

    A new alpha counter metric, route_controller_route_sync_total, added to the cloud controller manager so operators can A/B test watch-based route reconciliation. Narrow observability plumbing in service of an efficiency feature gate, not a user-facing capability shift.

    View source ↗