Skip to content

Versioning

How Keystone Core numbers releases, what each milestone signals, and what’s frozen vs. fair-game at each stage.

This document is canonical. When other docs (README, PROJECT-DETAILS, epics, V1X / ROADMAP entries, module doc-comments) talk about versioning, they defer here.

Scheme

Keystone Core follows Semantic Versioning with three milestone tiers before v1.0:

Version lineWhat it means
v0.1.xFirst public release. Linux-only. Epics 01–08 complete. Genuinely try-able: curious operators and early adopters are explicitly invited to install it, run it, and give feedback — expect rough edges and breaking changes between patch releases (minimised, always with a migration note). It is not the formal external-tester hand-off — that bar is the v0.5 checklist below — but it should be useful and interesting enough to attract people to the project.
v0.2.x – v0.4.xIncremental releases as features land. Each release closes one or more epics in some form. Cadence is release-as-ready — no calendar-pre-allocated scope. Breaking changes between minor versions are expected. The v0.x line stays attractive and try-able throughout.
v0.5.xFormal external-tester milestone (a checklist, not a calendar point — see the v0.5 gate below). Cross-Linux-distribution support solid; persistence renderers for the things early adopters actually need (networking, firewall backends); enough of Epics 09–13 to be usable as a multi-user/multi-host control plane. The first release we formally hand to operators outside the dev loop with a “what works / what doesn’t” guarantee. Ships unsigned (sha256sum -c verification) — release signing is the v0.8 milestone (see the v0.6–v0.9 row and RELEASE-PLAYBOOK.md §6). Hosted unsigned apt/dnf repos (apt install / dnf install via repos.keystone-core.io) ship here for convenience, at the same trust level as the direct downloads.
v0.6.x – v0.9.xPolish + remaining epic work toward v1.0. Same release-as-ready cadence. Breaking changes possible but rare; each is documented with a migration note. Release signing lands at v0.8 — the supply-chain milestone (signed tags / checksums / SBOMs, GPG signatures added to the apt/dnf repos, signed-commit enforcement). v0.1–v0.7 ship unsigned (sha256sum -c verification) per RELEASE-PLAYBOOK.md §6; the hosted repos themselves ship unsigned from v0.5.
v1.0.0All 19 epics complete and acceptance-tested. API + CLI + state-file YAML + gRPC frozen. SemVer stability commitment begins here: post-v1.0 minor releases add features, never break existing ones.
v1.xFeature additions on the stable v1.0 line.
v2.0 / v2.xFederation, multi-region, supercluster, cloud KMS. Identity unchanged from the original roadmap — distant.
v3.0+Marketplace, web UI, multi-cloud.

v0.5 gate — “external-tester ready, all Linux”

The v0.5 milestone is not a calendar point; it’s a checklist. v0.5.0 ships when every item below is verifiable on the cross-distro CI matrix.

Platform support

  1. All major Linux distribution families pass the cross-distro CI matrix for applicable stdlib modules:
    • Debian 12, Ubuntu 22.04 + 24.04
    • RHEL 9, Rocky 9
    • Alpine 3.19
    • openSUSE (Leap or Tumbleweed) — stretch target; may slip to v0.6
  2. Idempotency hammered: same state file applies twice produces zero changes on every supported distro for every applicable module.

Network base (the biggest current gap)

  1. At least two persistence renderers for network / route / bond / bridge / vlan:
    • netplan (Ubuntu’s default)
    • systemd-networkd (cross-distro)
    • NetworkManager / RHEL ifcfg-* may follow in v0.6+.
  2. Persistence is opt-in via persist: <renderer>|auto on each network-base decl.

Firewall base

  1. All four firewall modules verified on every distro where each is native:
    • iptables — every distro.
    • nftables — every distro (modern kernels).
    • firewalld — RHEL/Rocky/Fedora native; runs on Debian/Ubuntu when installed.
    • firewall (abstraction) — auto-detects and dispatches correctly; explicit backend: override works.
  2. firewalld rich-rule canonicalisation in the firewalld module (today documented as a v0.x backlog item — required for v0.5 because every RHEL host depends on it).

Package management

  1. package module supports apt + dnf + apk minimum. zypper optional for v0.5 (block on v1.0). v0.1’s apt-only single-backend is alpha-only.

Security

  1. AppArmor backend in the security module — Ubuntu/Debian operators expect it.
  2. Both backends behind the same security module surface (auto-detect + explicit backend: override).

Reboot + system

  1. Cross-distro reboot detection in the system module:
    • Debian: /var/run/reboot-required marker (v0.1).
    • RHEL: dnf needs-restarting -r exit-code parsing.
  2. Reboot result delivery across the disconnect — the kscore-agent coordinates with its reconnect logic to deliver a system.rebooted follow-up event after the host comes back.

Storage

  1. disk filesystem-resize — resize2fs + xfs_growfs at minimum.
  2. lvm VG PV-set + LV resize — vgextend / vgreduce for live VGs, lvextend --resizefs for filesystems-on-LV.

Engine + acceptance

  1. All Epic 08 acceptance criteria checked off (the ones the epic listed but task 13 was scheduled to verify).
  2. Identity + Auth (Epic 09) complete — testers running multi-user/multi-host need real auth.
  3. At least Events (Epic 11) wired through — operators need to see what’s happening.
  4. Engine coverage >85%, module coverage >80%.

Documentation

  1. A “v0.5 — what works, what doesn’t” doc lists every state-file param and each one’s status (stable / experimental / deprecated). Operators evaluating v0.5 can know in advance what’s safe to depend on. Delivered as STATE-SUPPORT-MATRIX.md .
  2. Hugo docs site live under docs/ . The v0.1.x Markdown surface (README.md, docs/project/*.md, docs/runbooks/*.md) graduates into a Hugo + Docsy site with per-page navigation and full-text search. The auto-generated CLI / config / API references regenerate via make docs-sync into the Hugo content tree. Pulled forward from a former v1.x position because v0.5’s “external-tester ready” audience benefits from a polished doc experience; see ROADMAP.md § Hugo docs site.

v1.0 gate — “stable, production-ready”

The v1.0 milestone is the SemVer stability commitment. Once cut, breaking changes require a v2.0.

  1. All 19 epics complete (epics/01-*.md through epics/19-*.md) and their acceptance criteria checked off.
  2. API + CLI + state-file YAML + gRPC contracts frozen for at least one full v0.x cycle before v1.0 cut (i.e. v0.9.x must match v1.0.0 on contracts — no breaking changes in the v0.9→v1.0 jump).
  3. Cross-distro CI matrix green for the full applicable-module set on all supported distros — same matrix as v0.5, expanded to include zypper/SUSE and any post-v0.5 distros.
  4. Engine coverage >85%, module coverage >80% across the board (same standard as v0.5; verified at v1.0).
  5. Performance + load test baseline passing (per Epic 19’s harden-and-release acceptance).
  6. Security review + audit complete (per Epics 12 + 19).
  7. Migration tooling for any state-file or schema changes between v0.x and v1.0 — operators can move data forward without a manual rewrite.
  8. Outstanding “must-haves” from external testers addressed — the v0.5+ tester feedback loop closes before v1.0 cut.
  9. CHANGELOG + RELEASE-PLAYBOOK ready for the long-term v1.x cadence.

The former gate-7 “Documentation site live” moved to the v0.5 gate (above). All other v1.0 gates renumbered down by one.

v0.x roadmap

There is no pre-allocated table of v0.2 contents, v0.3 contents, etc. Pre-allocating that table is what got us into the original “v1.0 contains everything” problem.

Instead:

  • One ranked roadmap in docs/project/ROADMAP.md (replaces ROADMAP.md).
  • Each entry carries a priority: gate-v0.5 | gate-v1.0 | v0.x | v1.x | v2.x+ (the taxonomy in docs/project/ROADMAP.md and tools/trackerctl).
  • v0.x minor releases are cut when natural feature sets land. A v0.3.0 might close one epic; a v0.4.0 might close two and ship a renderer; the boundaries fall where they fall.
  • The v0.5 gate is the only intermediate stop where the version specifically waits for a checklist.

Breaking changes pre-v1.0

This is the whole point of staying on v0.x for now:

  • Patch releases (v0.N.X → v0.N.X+1) prefer compatibility but may break.
  • Minor releases (v0.N → v0.N+1) routinely break.
  • Each breaking change ships with:
    • A CHANGELOG.md entry under ## Breaking
    • A migration note in the release announcement
    • When possible, a deprecation period: the prior version warns + continues to work for one minor cycle before removal.
  • Once v1.0 ships, no breaking changes without a v2.0.

Implications for downstream consumers

  • Do not pin keystone-core@latest in production before v1.0.
  • Pin to exact patch versions (e.g. keystone-core@v0.3.2) and read the CHANGELOG before bumping.
  • State files written for v0.N should be tested against v0.N+1 before bulk upgrade.
  • v0.x → v0.5 is an upgrade path explicitly engineered for; v0.5 → v1.0 is too.

Cross-references