Free course · 11 lessons · Advanced
Cloud Engineering in Production
The cloud incidents that actually happen, AWS first with Azure and Google Cloud equivalents: leaked access keys, public buckets, over-privileged roles, forgotten resources, quotas that stop scaling, data-transfer bills, hidden single-zone dependencies, backups that don't restore and untested region failover, with runnable simulations and two capstone case studies.
1. Security Incidents
Most cloud breaches come from the customer side of the shared responsibility model: keys, public data and permissions.
- The Key on GitHub — Long-lived access keys leak into repositories, logs and laptops. Remove them: SSO for people, OIDC for CI, roles for workloads, and guardrails for the rest.
- The Bucket Anyone Could List — A bucket policy with Principal * exposed customer files. Block public access at the account level, serve public content through a CDN, and share files with presigned URLs.
- The Function With Admin Rights — An SSRF bug in one service exposed credentials for a role that could do everything. Scope roles to what's used, add boundaries, and harden metadata access.
2. Cost and Limits
Bills and quotas fail quietly until they fail loudly. Make spend visible, cap the surprises, and know your limits before launch day.
- The Resources Nobody Owned — Untagged, forgotten resources quietly cost thousands a month. Enforce tags, give everything an owner and an expiry, and clean up automatically.
- Scaling Stopped at the Quota — Launch-day traffic hit an account quota nobody had checked. Know your quotas, request increases early, and alarm on usage approaching them.
- Why Is Our Bill Mostly Networking? — Cross-zone chatter, NAT processing and internet egress can outgrow compute costs. Measure traffic, keep paths local, and use endpoints and CDNs.
3. Reliability
Zones fail, regions have bad days and backups are only as good as the last restore test. Design for the failures you'll actually see.
- The Zone Outage We Didn't Survive — Everything was multi-AZ except three small things. Find hidden single-zone dependencies, design for static stability, and practise zonal failover.
- The Backup That Didn't Restore — Backups existed, but the restore took nine hours and missed data. Define RPO and RTO, protect backups from deletion, and test restores on a schedule.
- The Region Failover Nobody Had Practised — A multi-region design existed on paper, but failover took hours. Decide what needs multi-region, automate failover with health checks, and rehearse it.
4. Capstone Case Studies
Design a secure landing zone for a regulated start-up, and investigate a weekend that ended with a five-figure bill and a data exposure.
- Capstone: A Landing Zone for a Regulated Start-up — A health-tech start-up is moving from one shared account to a production-grade cloud foundation. Design the accounts, identity, guardrails, logging, networking and cost controls.
- Capstone: Postmortem — The Weekend That Cost $52,000 — A leaked key, a public bucket and an alert nobody received. Reconstruct the incident and write the remediation plan.