You can deploy to Kubernetes cluster. Can you fix it when it breaks ? Tutorials show you the happy path. Production pages you for the other 95%: a rollout that stalls halfway, a Pod that restarts with nothing in the logs, a Service with no endpoints, an RBAC rule that looks right and isn't. Kubernetes Incident Lab puts a broken system and a short ticket in front of you. You investigate with your own kubectl, find the cause, and fix it. Then an automatic validator checks the real cluster state and tells you whether you actually recovered it. No self-marking, no "looks about right". Why this is different Real failure patterns, not toy exercises. Every incident is modelled on a failure that actually pages people: quota blocking a rollout, a ConfigMap change that never reaches the Pod, probes that kill a healthy app, a namespace stuck Terminating. Later incidents have two or more layered causes, so fixing the first one reveals the next, like the real thing. Runs entirely on your own machine. A disposable local cluster on Docker, with its own kubeconfig. No accounts, no cloud bill, no website or server to depend on, nothing leaves your laptop, and it can never touch your real clusters. After the one-off setup it works offline. A validator you can't fool. It checks what is really running, and it rejects shortcuts: deleting the probe, raising the quota, granting cluster-admin, or removing the failing component. Spoiler-free tickets. The scenario tells you the symptoms, never the cause. Three progressive hints are there if you need them. The production lesson is revealed only after you pass. Full walkthroughs. After each incident you get the evidence trail, the root cause, the fix, and what not to do, so you learn the method as well as the answer. In the age of AI, understanding is the skill that matters An AI assistant can suggest a fix for any error message you paste. It can't tell you which of five symptoms is the real one, whether its suggestion is safe in your cluster, or what to look at when nothing is erroring at all. The engineers who stay valuable are the ones who can read the evidence, form a hypothesis and verify it themselves. This lab trains exactly that. You won't ask for the answer. You'll find it, and a validator will prove you did. How it works Work in your own IDE. Open the folder in VS Code, IntelliJ, Cursor or any editor, read each ticket there, and run every ./lab.sh command in the IDE's built-in terminal. Start with a broken system. Launch an incident in your local Kubernetes lab. Investigate with kubectl. Inspect Pods, Deployments, Services, events, configuration, resource requests, and other relevant evidence. Form a hypothesis. Use the evidence to explain what is failing and why. Request hints when needed. Each incident includes three progressive hints to help you get unstuck without immediately giving away the diagnosis. Recover and validate. Apply the recovery configuration and run the objective validation command. Assess your reasoning. Use the 10-point self-assessment to reflect on classification, evidence, root cause, recovery, and prevention. Choose your version Free: 5 incidents (workloads, networking, images, scheduling, RBAC), the full tooling, hints and a notes template. Try the method for nothing. Core ($39): 30 incidents (the 5 free ones plus 25 more, five of them at difficulty 4), full solutions and walkthroughs, eight themed learning paths (foundations, new on-call, exam practice, release engineering, security, networking, storage, senior refresher), 12 months of maintenance updates (fixes and compatibility with new Kubernetes/kind releases). Most people start here. Pro ($89): Core plus the 20 hardest multi-cause incidents, 50 in total (fail-closed admission webhooks, token audiences, aggregated RBAC roles, TLS trust chains, paused and stalled rollouts, immutable ConfigMaps and volume claims, evictions where nothing crashes, a CronJob blocked by a hung Job, and more), blameless post-mortem templates with two worked examples, a 150-question interview bank with model-answer outlines, the learning paths, and lifetime maintenance updates. Team ($299): Pro for up to 10 named people in one organisation, plus a ready-to-run team kit: a 90-minute game day, a half-day workshop and a 4-week onboarding plan, a slide deck and printable role cards, a post-mortem workshop pack, a team report tool that combines everyone's completion summaries (by person, area and level, weakest areas), and admin and procurement notes. Lifetime maintenance updates. Who it's for Good fit: Developers, DevOps and platform engineers, SREs and anyone preparing for interviews who knows what a Pod and a Deployment are and wants real practice fixing things when they break. Teams who want a hands-on game day without writing a curriculum. Before you start You’ll need Docker and kubectl installed (kind is the only other tool). If you haven’t used them yet, the free version is the easiest place to begin. ./lab.sh doctor checks your setup and tells you what’s missing, so you’ll know within minutes whether your machine is ready. What this is, and isn’t It’s hands-on practice, not a video course: you fix real failures yourself and get proof. It runs on a local cluster, so the skills carry to EKS, GKE and AKS, but it doesn’t cover cloud-specific services such as load balancers, IAM or managed add-ons. Questions What do I need? Docker running with about 4 GB free, kubectl, kind, and macOS, Linux, or Windows with WSL2 or Git Bash. ./lab.sh doctor checks it for you. Is it safe for my machine and my clusters? It runs on its own kind cluster with its own kubeconfig and never touches your normal one. How long does each incident take? 10 to 45 minutes. Difficulty runs from 1 to 5. Will I get new incidents? Maintenance updates are included. New incidents may be released from time to time as a bonus; they are not part of any version's promise. Can I share it with my team? Free, Core and Pro are for one person. The Team version covers up to 10 named people. Can I get refund ? All sales are final. The free sampler runs on the same tooling as every paid version, so it shows you exactly how the lab looks, feels and installs before you pay. If it runs on your machine and the incident list matches what you want to practise, buy with confidence. The one exception: if you were charged twice, or the file is damaged and we cannot fix it, contact us through your Gumroad receipt and we will put it right. By DecodeDevOps.