15 lessons · read, then do
Each lesson is the written half of a lab: the mental model, the commands it actually uses, and the four things that usually go wrong. Every command listed is one the lab really runs, so nothing here describes something the product cannot demonstrate.
15 lessons
Infrastructure abstraction
A virtual machine is a computer made of files. A hypervisor hands a guest operating system fake hardware, and the guest cannot tell the difference — which is why the same idea runs a laptop demo and half of the public cloud.
The shell is the classroom
Five ideas carry almost everything you will do on a Linux box: everything is a file, paths are absolute or relative, permissions are bits, services are supervised processes, and text is data you can interrogate.
Names before packets
A service is a listener on a port; a name is a pointer to an address. Almost every 'the host is down' is really 'the name resolved wrong', 'the port is not listening', or 'a firewall dropped it' — three different checks.
From image to replica
An image is a filesystem plus a start command. A container is a process wearing that filesystem. Kubernetes adds the part a single machine cannot give you: scheduling, a stable address, and reconciliation to a declared number of replicas.
Snapshots, not diffs
A commit is a snapshot of the whole tree plus a pointer to its parent. A branch is a movable label on one commit. Once you hold those two facts, branching, merging and rebasing stop being magic.
Identity first, state second
In the cloud you never hold the machine — you hold a policy and a record of what should exist. Every failure is either 'who am I' (credentials) or 'what exists' (state), and the CLI answers both.
Declare, plan, apply
Terraform compares what you declared against what it recorded and proposes the difference. The state file is the memory of that comparison — lose it and you are adopting resources by hand.
A contract with the repository
A pipeline is a written promise: every push gets built, tested, and only then deployed. It lives in the repository, so it is reviewed like code and versioned with the thing it ships.
One path, end to end
The capstone is the exam for the whole path: build an image, declare the workload, commit the desired state and let a controller reconcile it. Nothing new — the same tools, in the order production uses them.
Ask the system a question
Metrics tell you that something changed, logs tell you why, and dashboards are only useful if they answer a question someone actually asks. An alert is a promise that a human will act.
Assume the repository is public
A secret committed once is a secret forever: history keeps it, forks copy it, and CI logs quote it. Treat every repository as if it were about to be published, and scan before the merge, not after the incident.
State, not shell history
Ansible describes the end state of a host: this package present, this service running, this file in place. Run it twice and the second run changes nothing — that is idempotence, and it is the point.
Scripts that fail loudly
An automation script has one job: return a status that can be trusted. That means a pinned environment, an explicit interface, and an exit code that is non-zero when the thing it checked is broken.
Ground it, or refuse
A retrieval system earns trust by answering from supplied documents and saying 'no matching context' when there is none. Everything else — chunking, embeddings, prompts — serves those two behaviours.
Reproducibility is the product
A model you cannot rebuild is a demo. MLOps is the discipline of pinning the environment, versioning the data separately from the code, and recording every run so a result can be explained later.
A lesson is not a substitute for the lab. Reading the model makes the first attempt faster; only the lab makes you able to do it. Open the catalog — or open a themed playground.