Kubernetes (K8s) & AKS

Kubernetes is an open-source container orchestration platform: you declare the desired state (e.g. "run 3 replicas of shop-api:1.4, expose them on port 80"), and Kubernetes continuously makes the actual state match — scheduling containers onto machines, restarting failed ones, scaling, rolling out updates and load balancing.

Created at Google (inspired by its internal Borg system), open-sourced in 2014, donated to the CNCF (Cloud Native Computing Foundation) in 2015. "K8s" = K + 8 letters + s.

Managed offerings: AKS (Azure Kubernetes Service), EKS (AWS), GKE (Google). Local: kind, minikube, k3d, Docker Desktop's Kubernetes.

What Kubernetes gives you: self-healing, horizontal scaling, service discovery and load balancing, rolling updates and rollbacks, config and secret management, storage orchestration, batch jobs, and an extensible API (CRDs and operators).

Kubernetes releases three minor versions per year, each supported upstream for about 14 months (≈ the three latest minors). AKS supports roughly the latest three GA minor versions (plus Long-Term Support for selected versions on Premium tier) — clusters must be upgraded regularly.

Pod lifecycle phases: Pending (accepted, waiting for scheduling/image pull) → Running (bound to a node, at least one container running) → Succeeded (all containers exited 0 — Jobs) or Failed (a container exited non-zero and won't restart) — plus Unknown (node unreachable).

Container states: Waiting (e.g. ContainerCreating, ImagePullBackOff, CrashLoopBackOff), Running, Terminated.

Pod startup sequence: scheduler picks a node → kubelet pulls images → init containers run in order to completion → app containers (and sidecars) start → postStart hook → startup probe → readiness probe passes → pod added to Service endpoints → liveness probe keeps checking.

Pod termination: pod marked Terminating → removed from Service endpoints → preStop hook → SIGTERM sent → app shuts down gracefully → after terminationGracePeriodSeconds (default 30 s) → SIGKILL.

Deployment rollout lifecycle: new ReplicaSet created → pods scaled up/old down according to maxSurge/maxUnavailable → readiness gates progress → rollout complete (or stuck → kubectl rollout undo). Revision history kept (revisionHistoryLimit, default 10).

Cluster lifecycle (AKS): create cluster/node pools → deploy workloads → monitor → upgrade control plane then node pools (node image and Kubernetes version) → scale → eventually decommission.