stella-ops.org/git.stella-ops.org
1
0
Fork 0
Code Issues Pull Requests Actions 2 Releases Wiki Activity
Files
534aabfa2a8c9264f2479541a5c11e48de7b97fa
git.stella-ops.org/docs/operations/multi-tenant-rollout-and-compatibility.md

4.5 KiB
Raw Blame History

Multi-Tenant Same-Key Rollout and Compatibility Policy

Date: 2026-02-22
Source sprint: SPRINT_20260222_053_DOCS_multi_tenant_same_api_key_contract_baseline.md
Related ADR: docs/architecture/decisions/ADR-002-multi-tenant-same-api-key-selection.md

Rollout Goals

  • Deploy one-selected-tenant-per-token behavior with no cross-tenant leakage.
  • Preserve bounded backward compatibility for legacy tenant headers during migration.
  • Enforce strict mode only after module-level validation is complete.

Phase Plan

Phase 0 - Contract and Feature Flags (2026-02-22 to 2026-02-28)

  • Deploy order:
    • Publish ADR + service impact ledger + flow sequences + acceptance matrix.
    • Introduce feature flags required for bounded compatibility paths.
  • Exit criteria:
    • Contract docs accepted by Authority, Router, Platform, Scanner, Graph, and Web owners.
    • Sprint trackers for 053-060 are ready with test evidence plans.
  • Rollback criteria:
    • Contract ambiguity or unresolved cross-module naming conflicts.

Phase 1 - Authority + Router/Gateway Compatibility Mode (2026-03-01 to 2026-03-07)

  • Deploy order:
    • Authority (20260222.054) before Router/Gateway (20260222.055).
    • Enable compatibility aliases while keeping canonical header/claim behavior authoritative.
  • Exit criteria:
    • Token issuance resolves selected tenant deterministically.
    • Router strips spoofed inbound headers and rewrites canonical tenant headers.
    • Targeted Authority/Router tests pass in CI.
  • Rollback criteria:
    • Token issuance regression or tenant mismatch false positives on production traffic.

Phase 2 - Service Migration: Platform, Scanner, Graph (2026-03-08 to 2026-03-17)

  • Deploy order:
    • Platform (20260222.056)
    • Scanner (20260222.057)
    • Graph (20260222.058)
  • Exit criteria:
    • Tenant resolver paths and endpoint policies are consistent.
    • Cross-tenant access attempts are rejected deterministically.
    • Module integration tests for tenant isolation pass.
  • Rollback criteria:
    • Data partition mismatch, endpoint policy regression, or unresolved tenant conflict errors.

Phase 3 - Web Global Selector + Client Unification (2026-03-18 to 2026-03-24)

  • Deploy order:
    • Web selector and runtime state unification (20260222.059).
    • Playwright tenant matrix and QA evidence (20260222.060).
  • Exit criteria:
    • Global tenant selector switches tenant context across primary pages.
    • Canonical tenant injection path is active with legacy usage telemetry.
    • Tier 2c UI evidence is complete.
  • Rollback criteria:
    • Tenant switch causes stale data bleed, broken navigation, or unrecoverable error states.

Phase 4 - Strict Mode + Legacy Removal (earliest start 2026-04-01)

  • Deploy order:
    • Disable legacy tenant header acceptance for new clients first.
    • Remove compatibility aliases from runtime clients and gateway compatibility branches.
  • Exit criteria:
    • Legacy header usage telemetry remains at zero for 14 consecutive days.
    • No dependency remains on scalar-only or header-only tenant override paths.
  • Rollback criteria:
    • Any tenant selection outage or blocked tenant-scoped production path.

Compatibility Window

  • Legacy header aliases (X-Stella-Tenant, X-Tenant-Id) are supported in compatibility mode through 2026-03-31.
  • Starting 2026-04-01, strict mode can be enabled once zero legacy usage is confirmed for 14 days.
  • Canonical header is X-StellaOps-Tenant during and after migration.

Observability Checkpoints

  • Gateway telemetry for stripped/spoofed tenant headers.
  • Legacy header usage counters from Web tenant interceptor telemetry.
  • Authority token issuance/validation audit events for tenant mismatch/ambiguity.
  • Platform/Scanner/Graph tenant conflict and forbidden response rates.

Rollback Playbook

  • If regressions appear:
    • Re-enable compatibility feature flags.
    • Freeze strict-mode rollout.
    • Revert last deployment batch for the affected module only.
    • Re-run tenant isolation acceptance matrix against last-known-good build.
  • Rollback does not change ADR model (one selected tenant per token); it only restores bounded compatibility behavior.

Production Readiness Checklist

  • Authority token issuance + validation tests green with selected tenant model.
  • Router/Gateway spoof/mismatch protections verified in staging.
  • Platform, Scanner, Graph tenant isolation tests green.
  • Web selector tests and Playwright tenant matrix green (desktop + mobile).
  • Legacy header telemetry reviewed with dated cutoff evidence.
  • Go/no-go decision documented by QA and Project Manager.