docs(): Add ADR for Hub-and-Spoke Partial Mesh MVP topology (#300)

[Shreesha001]

Adds the ADR for the Partial Mesh MVP (Hub-and-Spoke topology) at docs/adr/hub-and-spoke-adr.md, as required by #300. It covers all seven decisions from the issue (CRD schema, resolver placement, edge computation, rewire safety, backward compatibility, desiredPeers delivery, status aggregation), plus validation webhook rules, both sequence diagrams, non-goals, and open issues for rewiring.

This is the prerequisite gate for worker-operator #470 and #471.

Fixes #300

How Has This Been Tested?

Documentation-only change; no code modified. Code references in the ADR were verified against the current source.

Checklist:

• The title of the PR states what changed and the related issues number (used for the release note).
• Does this PR require documentation updates?
• I've updated documentation as required by this PR.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas. (N/A — docs only)
• I have tested it for all user roles. (N/A — docs only)
• I have added all the required unit test cases. (N/A — docs only)

Does this PR introduce a breaking change for other components like worker-operator?

No.

[Copilot]

Pull request overview

Adds an Architecture Decision Record (ADR) documenting the proposed Hub-and-Spoke topology for the Partial Mesh MVP in KubeSlice, intended as a design gate for subsequent controller/worker-operator implementation work.

Changes:

• Introduces a new ADR covering proposed SliceConfig topology schema, edge computation, rewire behavior, status aggregation, webhook validation rules, and convergence sequences.
• Documents operational trade-offs and non-goals for the MVP scope.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[pnavali]

@Shreesha001 - please review and resolve this one.

[pnavali]

This will be a living document with multiple reviews.

[gourishkb]

Strongest of the two ADRs this week. Decision 3's explicit Client/Server role pinning (hub = Server, spoke = Client) is the right call and something the current code doesn't guarantee — good catch. Decision 7's spoke-to-spoke routing via a slice-CIDR-scoped default route is clever and self-maintaining. A few things need resolution before this gates #470/#471.

Must fix

1. Decision 2 contradicts Decision 7 on worker-side changes. Decision 2 says "the worker-operator needs little or no change for the core feature to work" because gating happens controller-side. Decision 7 then proposes adding a default-route injection to the sliceRouterInjectRoute path in router-sidecar (worker-side). These two claims conflict. Please clarify: is spoke-to-spoke routing via hub in scope for this MVP, and if so, acknowledge explicitly that it requires worker-side changes. If it is out of scope, move it to Non-Goals and remove Decision 7.

2. Decision 4's concurrent edit guard requires stateful webhook validation — the webhook must read the existing SliceConfig's status.conditions (to check TopologyConverged) during an UPDATE call. This is more complex than the stateless validations already in the file. Please address: (a) does the webhook have a client to read the existing object? (it receives old object in the admission request, so yes — use old.Status.Conditions, no extra fetch needed); (b) what is the behavior when TopologyConverged condition is absent (first topology update on a new slice)? Document the answer so the implementer doesn't have to guess.

3. SliceConfigStatus.Conditions field addition is an implicit CRD schema change. Decision 6 adds Conditions []metav1.Condition to SliceConfigStatus. This requires a Go type change, make generate && make manifests, and a separate PR for the CRD update. The ADR should list this as an explicit prerequisite for #303 and reference the PR that will carry it. Right now it reads like a free addition but it isn't — it must be done before any status-aggregation code can ship.

4. Missing: FullMesh → HubAndSpoke transition in sequence diagrams. The hub-change sequence is covered, but the initial transition from an existing full-mesh slice to HubAndSpoke is not. Specifically: when an operator applies topology.mode: HubAndSpoke to an existing full-mesh slice, cleanupObsoleteGateways will delete a large number of WorkerSliceGateway objects simultaneously. What is the expected connectivity impact? Is there a maximum safe rate of deletion? Please add a note in Decision 4 and a short sequence for this case.

Minor

5. Issues #301, #302, #303, #304 appear in the header's Related field but are not in the original downstream issue list from #300 (which only lists worker-operator #470 and #471). Please confirm these are newly created issues, link them from #300, and update the issue body to reference them so the tracking is complete.

6. Missing trailing newline at end of file.

[gourishkb]

This paragraph says "the worker-operator needs little or no change for the core feature to work". Decision 7 later proposes adding a default-route injection to sliceRouterInjectRoute in router-sidecar, which IS a worker-side change. These two claims conflict — please reconcile.

[gourishkb]

The Client/Server role pinning (hub=Server, spoke=Client) is the right call for NAT-traversal. Suggest adding it as a webhook validation rule: reject HubAndSpoke if the hub cluster has no NodeIP or no external endpoint configured (i.e. ClusterStatus.NodeIPs is empty). Otherwise an operator can designate a NAT'd spoke as hub and not realize it won't work until tunnels fail to come up.

[gourishkb]

The concurrent-edit guard is correct in principle, but the webhook receives the old object in the admission request — use oldObj.Status.Conditions to read TopologyConverged without an extra API call. Please also specify the nil case: if the condition is absent (slice was just created or never transitioned), treat it as True (no in-progress rewire) so the first topology update is always allowed.

[gourishkb]

Adding Conditions []metav1.Condition to SliceConfigStatus requires a Go type change and CRD regeneration (make generate && make manifests). This is not free — it needs its own PR (can be part of the #301 CRD PR). Please call this out as an explicit dependency so whoever implements #303 knows they need the schema change landed first.

[gourishkb]

Missing trailing newline.