Release Readiness & Deployment Verification
Deciding whether to ship — go/no-go criteria, release checklists, canary/blue-green/progressive rollout verification, post-deploy smoke, rollback rehearsal, DORA metrics, change-risk scoring, and release sign-off reports.
Answer "Is this safe to ship, and how will we know fast if it isn't?" with evidence, not vibes.
Go / No-Go scorecard
| Gate | Evidence | Pass rule |
|---|---|---|
| Functional | Regression + new-feature results | 100% Must pass, 0 Sev1/Sev2 open |
| Non-functional | Perf vs baseline, security scan, a11y | p95 within +10%; 0 Critical/High vulns; no new WCAG AA fails |
| Data | Migration rehearsed; backward-compat N/N-1 | Reconciled; rollback tested |
| Operability | Dashboards, alerts, runbook, on-call aware | SLO alerts exist for new paths |
| Rollout | Flag/kill-switch, canary plan | Halt criteria defined |
| Compliance/Docs | Legal, privacy, release notes, support briefed | Signed |
| Risk | Change risk score (size × criticality × novelty × blast radius) | Below threshold or mitigated |
Change-risk heuristics: touches auth/payments/data schema (+), large diff (+), first-time contributor to area (+), late Friday (+), weak test coverage on changed lines (+), feature-flagged (–), easily reversible (–), automated rollback (–).
Pre-release checklist
Build reproducible & signed · versions/changelog · config diff prod vs staging (configuration-feature-flag-testing) · migrations ordered & backward-compatible · secrets present · dependencies scanned (supply-chain-dependency-testing) · smoke suite green on the release candidate artifact · perf smoke · feature flags default OFF · rollback command written & tested · comms plan.
Progressive delivery verification
- Deploy dark (0% traffic) → post-deploy smoke against prod URL (
smoke-sanity-testing). - Canary 1–5% for 10–30 min: compare canary vs baseline on error rate, p95 latency, saturation, business KPIs (orders/min, sign-ups) — automated analysis (Argo Rollouts/Flagger/Kayenta).
- Ramp 25% → 50% → 100% with bake time; auto-rollback on SLO breach.
- Blue/green: keep old env warm; switch back = instant rollback; verify DB compat both directions.
- Post-release 24–48 h: error budget burn, support tickets, crash-free rate, analytics parity.
kubectl argo rollouts get rollout app --watch
kubectl rollout status deploy/app --timeout=5m && ./smoke.sh https://prod.example.com
kubectl rollout undo deploy/app # rehearse this!
Release report (1 page)
Recommendation (GO / GO-with-risks / NO-GO) first → scope → test evidence links → open defects & accepted risks (owner, mitigation) → rollout & rollback plan → monitoring links → approvers.
DORA metrics to trend
Deployment frequency · lead time for changes · change failure rate · time to restore (MTTR). Improve by shrinking batch size, not adding gates.
Related
smoke-sanity-testing, regression-testing, production-monitoring-synthetic-testing, test-planning-documentation, cicd-testing