W3DevOps Module · Kubernetes: Pods, Deployments, Services, Ingress and kubectl

Most of my deployed projects run on Cloud Run, which is Kubernetes-shaped (Knative) without the cluster to manage. But you can't make good decisions about when to stop using a managed platform without understanding what it hides. This is the Kubernetes module of the course — 26 lab captures across four lessons.

What this module covers

• The control plane (kube-apiserver, etcd, scheduler, controller-manager) and worker components (kubelet, kube-proxy, containerd)
• The three standard interfaces: CRI, CNI and CSI
• Pods → ReplicaSets → Deployments, with health checks and resource limits
• Deployment vs StatefulSet vs DaemonSet — which one, when
• ClusterIP / NodePort / LoadBalancer, and host/path routing with Ingress
• A kubectl cheat sheet organised around debugging

Lessons in this module

1. Kubernetes Cluster Architecture — Kubernetes (often abbreviated K8s) is an open-source container orchestration engine originally designed by Google (Borg) to automate deploying, scaling, healing, and managing containerized applications…
2. Kubernetes Workloads: From Pods to Deployments — In Kubernetes, you almost never deploy individual containers directly. Instead, you work with higher-level abstractions that manage replication, rollouts, health probes, and persistent identity.
3. Kubernetes Services & Ingress Routing — Pods are mortal. They die, scale up, scale down, and get rescheduled onto different nodes, changing their IP addresses continuously. A Kubernetes Service provides a stable, permanent IP and DNS entry that…
4. Essential Kubectl Commands & Production Debugging — kubectl is the official command-line client used to control any Kubernetes cluster. Below is the ultimate reference matrix used during production outages and deployments.

Why my projects mostly don't run on Kubernetes (yet)

Reading this track back, the honest summary is: Kubernetes is the right answer when you have many services, many teams, or stateful workloads that need careful scheduling. For a single FastAPI agent backend with a SQLite file, it's overhead.

• Nora Labs is pinned to --max-instances=1 on Cloud Run because SQLite has a single writer. On Kubernetes that would be a StatefulSet with replicas: 1 and a PVC — more YAML for the same constraint.
• The Part Timer has two independently deployable services (API and web) — the point at which an Ingress with path routing starts to make sense. Today Cloud Run gives each one its own URL.
• The SRE handbook project is where I practice the K8s side: static container + Ingress routing.

Next in the series: AWS — IAM, EC2, S3, VPC and Terraform.

---

Keep going: Start the module → · Full course outline · Self Guide