Skip to content

Latest commit

 

History

History
77 lines (59 loc) · 3.31 KB

ROADMAP.md

File metadata and controls

77 lines (59 loc) · 3.31 KB

schedkit roadmap

Recent Linux kernels ship sched_ext, which lets CPU schedulers be written in eBPF and swapped at runtime. The kernel side is solved; distributing, verifying and operating schedulers across machines is not. schedkit's goal is to make a CPU scheduler as easy to ship, trust and roll out as any other piece of software: pulled from an OCI registry, verified with Sigstore, and managed by tools platform teams already use.

This roadmap is a statement of direction, not a promise of dates. It is reviewed by the maintainers at least once per quarter. To propose a change, open a Discussion or a pull request against this file.

Last updated: October 2026.

Now (next ~3 months)

Project and community

  • Written governance, maintainer list, security policy, contributing guide and Code of Conduct across the organization.
  • DCO enforced on every repository.
  • Onboard at least one additional maintainer, ideally from a different organization.
  • Apply to the CNCF Sandbox and present schedkit to CNCF TAG Workloads Foundation.
  • OpenSSF Scorecard on every repository and an OpenSSF Best Practices badge.

schedctl

  • Automated, reproducible releases: signed binaries (Sigstore keyless), SBOM and provenance attached to every GitHub release.
  • Continue distribution packaging beyond openSUSE and the AUR (Debian, Fedora, Nix, Gentoo).

sked

  • Harden the path to a stable v1.0: upgrade and CRD migration story in the Helm chart, documented compatibility matrix (Kubernetes and kernel versions). (proposed)

plumbing

  • Pin every base and upstream image by digest and automate upstream version bumps. (proposed)
  • Document a license and provenance check for every scheduler in the catalog.

Next (3–9 months)

  • Shared trust policy. schedctl and sked currently each have their own signature verifier and trust-policy format. Converge on one format and one library so a policy can move between a laptop and a cluster unchanged. (proposed)
  • Scheduler image conventions. Turn the OCI labels used today (for example org.schedkit.scheduler.kernel.min-version) into a documented, versioned convention that third-party scheduler authors can target, with a conformance check in schedctl doctor/schedctl info. (proposed)
  • Observability. Expose per-node scheduler state (active scheduler, version, sched_ext kernel state, failures and fallbacks to the default scheduler) as Prometheus metrics and Kubernetes events. (proposed)
  • Safer rollouts in sked. Progressive rollout across node pools with automatic rollback when the kernel ejects a scheduler. (proposed)

Later / ideas

  • Integration with node-level resource management (NRI plugins, CPU manager policies) so workloads can express scheduling intent. (proposed)
  • A browsable catalog of verified scheduler images on schedkit.io. (proposed)
  • Support for additional container runtimes (for example CRI-O). (proposed)

Non-goals

  • Writing or maintaining CPU schedulers ourselves. schedkit packages, verifies, distributes and operates schedulers developed upstream (for example in sched-ext/scx).
  • Replacing the Kubernetes pod scheduler (kube-scheduler). schedkit is about the kernel CPU scheduler on each node, not about placing pods on nodes.