5 · Platform engineering

18. Designing a Cloud-Native System End to End

A reference architecture, its failure modes and a production readiness checklist.

10 min read · 3 MCQs

A reference stack

Edge: DNS with health checks, CDN and WAF. Ingress: layer 7 load balancer terminating TLS. Compute: containers on Kubernetes or a managed runtime, spread across three AZs. State: managed relational primary with read replicas, plus object storage and a cache. Async: queue and workers. Everything defined in IaC and deployed by pipeline.

Failure modes to design for

AZ loss, dependency latency spikes, thundering-herd retries, cache stampedes, database connection exhaustion, expired certificates and quota limits. Assume every remote call can hang; enforce timeouts, retry budgets with jitter, and circuit breakers.

Production readiness

Before launch: SLOs and dashboards, alert routing and on-call rota, tested backups and a rehearsed restore, autoscaling verified under load test, least-privilege IAM reviewed, secrets rotating, cost budget alarms, and a documented rollback.

  • Backups are only real once a restore has been rehearsed.
  • Load test before launch, not after the incident.
  • Every dependency needs a timeout.

Chapter quiz

3 questions · pass mark 75%
  1. 1. Retries without jitter can cause…

  2. 2. A backup counts as reliable only when…

  3. 3. Spreading replicas across three AZs protects against…

Answer every question to submit. Progress for cl-18 is saved in this browser.