The vault’s second-largest domain (~148,000 words) and its largest reference-to-reps gap — you haven’t run Kubernetes, Terraform or Prometheus in anger. This is the section where doing matters most, and every project produces something you can point at.
Difficulty: 🟢 starter (hours–days) · 🟡 intermediate (a week or two) · 🔴 ambitious / portfolio-grade. ⭐ = highest signal.
The ladder
- 🟢 ⭐ Containerize + deploy one of your apps end to end — multi-stage Dockerfile, docker-compose for local, then deploy to a VPS behind Nginx with Let’s Encrypt TLS. The practical capstone.
- 🟡 A real CI/CD pipeline — extend beyond this vault’s deploy workflow: a GitHub Actions pipeline for one of your apps that runs typecheck/lint/tests on PRs and deploys on merge, with a build-once-promote-artifact flow.
- 🟡 Local Kubernetes deploy — stand up a
kind/minikubecluster, write the Deployment/Service/Ingress manifests, deploy a containerized app with health probes, and do a rolling update + rollback. Makes the biggest reference gap concrete. - 🟡 Terraform a small cloud setup — provision a VM + network + managed DB on a throwaway cloud account with Terraform, read the
plan,apply, thendestroy. Learn state the safe way. - 🔴 Observability stack — instrument an app with metrics, run Prometheus + Grafana (+ Loki for logs), build a dashboard and an alert on an SLO. The “you can’t operate what you can’t see” lesson, hands-on.
- 🟢 Secrets, done right — take an app with a committed
.envand move it to SOPS or Vault; wire it into the deploy.
Added reps — the operational ones
- 🟡 Break something on purpose, then write the postmortem — take down one of your own deployed services (fill a disk, exhaust connections, OOM a container), fix it, and write a blameless postmortem with a timeline. Done when: the postmortem names a systemic cause, not a person. Exercises: SRE.
- 🟡 Define an SLO and burn its error budget — pick a real service, define an availability or latency SLO, and alert on the burn rate rather than on raw errors. Done when: you’ve deliberately burned budget and the alert fired at the right time.
- 🟢 Restore from a backup you’ve never tested — back up a real database, then restore it into a fresh environment. Done when: the restore works and you know your actual RTO. Untested backups are the most common operational lie in software.
If you only do one
Containerize and deploy one app end to end. It’s the shortest path from reference knowledge to a running thing with a URL, and every other project here builds on having done it once.
Related
- the devops course · interview bank
- RHCSA exercises — the Linux fundamentals drill
- Project Ideas — the vault-wide index
Jump to a subject
Software engineering
Infrastructure & security
Computer science
Hardware & engineering