Capstone: A Production-Grade Platform, End to End

You will hand in

• Three repos (infra, app, config) with READMEs and a top-level architecture document with ADRs (architecture decision records)
• Infrastructure fully in Terraform, including DNS, CDN/WAF, EKS, data services, and IAM
• Secure CI (Project 11) + GitOps CD with canary (Project 9) + observability and SLOs (Project 10)
• Load-test report (k6) at 2× expected sale traffic, with the bottleneck found and fixed
• DR plan with measured RTO/RPO, a game-day report, and a cost breakdown per service

1. 1

   Architecture and decisions

   Done when: A written design the team could review before you build.

   • Architecture diagram and request flow (Production Architecture guide)
   • ADRs for key choices: EKS vs ECS, managed vs self-hosted Kafka, Argo CD vs Flux, Helm vs Kustomize, DR strategy
   • SLOs and RPO/RTO agreed per service

   Stuck? Hints

   • An ADR is short: context, decision, consequences. Write down what you rejected and why; it's what interviewers ask about.

   Prove it works

   terminalcopy cmd

   $ ls docs/adr

   ── expected output ──

   0001-eks-over-ecs.md

   0002-managed-data-services.md

   0003-argocd-gitops.md

   0004-warm-standby-not-active-active.md

   0005-kyverno-for-policy.md

2. 2

   Foundation: infra, delivery, security

   Done when: Terraform-built AWS + EKS, secure CI, GitOps CD, guardrails.

   • Combine Projects 7 and 8 into the infra repo (edge: Route 53, CloudFront, WAF)
   • Add Karpenter, Kyverno policies, External Secrets, Gateway/Ingress, cert-manager, ExternalDNS
   • Wire Projects 9 and 11: signed images only, canary to prod

   Prove it works

   terminalcopy cmd

   $ argocd app list -o name | wc -l; kubectl get clusterpolicy -o name | wc -l

   ── expected output ──

   14

   6

3. 3

   Observe and load test

   Done when: Full telemetry, SLO alerts, and proof it holds 2× peak.

   • Project 10 across all services, with a checkout SLO dashboard
   • k6 test at 2× forecast peak with arrival-rate executor (Stateful course, Unit 5.1)
   • Find the first bottleneck with USE/RED (Unit 5.3), fix it, re-run, and write the report

   Prove it works

   terminalcopy cmd

   $ k6 run --out experimental-prometheus-rw sale-2x.js | grep -E 'http_req_duration|http_req_failed'

   ── expected output ──

   ✓ http_req_duration p(95)=318ms

   ✓ http_req_failed rate=0.07%

4. 4

   Game day: incident → runbook → recovery

   Done when: A rehearsed failure, handled end to end, with a blameless postmortem.

   • Write runbooks for the top 5 alerts
   • Run a scheduled game day: inject payments latency and an AZ failure (Chaos Engineering guide)
   • Run incident command (SRE course), recover using runbooks, and write the postmortem with action items

   Prove it works

   terminalcopy cmd

   $ cat docs/postmortems/2026-10-gameday.md | head -8

   ── expected output ──

   # Game day postmortem — payments latency + AZ-a loss

   Detected: 2m 10s (CheckoutLatencyBurnFast)

   Mitigated: 11m (circuit breaker opened; traffic shifted from AZ-a)

   Customer impact: 0.4% of checkouts slowed > 2s; 0 failed orders

   What went well: alert → trace → runbook in < 5 min

   What didn't: runbook step 3 referenced a removed dashboard

   Action items: 4 (owners + dates)

5. 5

   DR and cost

   Done when: Measured recovery and an understood bill.

   • Execute the DR runbook (restore/fail over per your chosen strategy) and measure RTO/RPO
   • Tag everything; OpenCost + Cost Explorer breakdown per service
   • Optimisation plan: Savings Plans baseline, Spot for workers, VPC endpoints, right-sized requests (Cost guide)

   Prove it works

   terminalcopy cmd

   $ cat docs/dr-rehearsal.md | grep -E 'RTO|RPO'

   ── expected output ──

   Measured RTO: 38 min (target 60)

   Measured RPO: 1 min (target 5)

• Backstage portal with a golden-path template for new services (Platform course, Stage 1)
• Crossplane self-service databases (Platform course, Mission 2.2)
• Multi-region warm standby with Route 53 failover
• DORA metrics dashboard (Platform course, Mission 2.4)

Résumé bullet

Designed and built a production-grade AWS/EKS platform end to end: Terraform infrastructure, secure CI with signed images and SBOMs, GitOps canary delivery, OpenTelemetry-based observability with SLO alerting, load-tested at 2× peak, and validated through a chaos game day and a timed DR rehearsal.

Demo script

• 10-minute walkthrough: architecture diagram → one commit flowing to prod as a canary
• Live incident: inject a failure, show the alert, the trace, the runbook, and recovery
• Show ADRs, the load-test report, the DR rehearsal results, and the cost breakdown