4. Where orchestration fits
Orchestration begins when you stop managing one process by hand and start describing the state you want a platform to maintain.
Compose coordinates one application environment
Docker Compose lets you describe an API, database, and cache in one file, connect them on a network, attach storage, and start the stack together. It is especially useful for local development and can be a reasonable deployment tool on one server. Chapter 27 uses it to make the local system match the shape of production.
Compose does not turn one machine into a cluster. If that machine fails, every service on it fails with it. Backups, operating-system updates, external traffic routing, and failover still belong to you.
Managed platforms remove different burdens
A platform such as Railway, Render, Cloud Run, or ECS Fargate can start containers, restart failures, supply configuration, collect logs, and connect deployments to health checks. A managed database takes on backups, patching, storage management, and often failover. "Managed" does not mean there are no servers. It means someone else owns more of their operation.
Kubernetes is a reconciliation system
Kubernetes lets a team describe desired state across a cluster: run three instances of this image, make them reachable behind one service, roll out new versions gradually, replace failed workloads, and place them on machines with available capacity. Controllers continuously compare desired state with observed state and act to close the gap.
That is powerful when an organisation has many services, many teams, specialised networking or scheduling requirements, and enough operational capacity to run a platform. It is overhead when the real requirement is "keep one API and one database online." Kubernetes moves complexity into a consistent control plane; it does not make the complexity disappear.
| Shape | Good first option | What would force a change? |
|---|---|---|
| One small web app | Managed application platform | Platform limits, compliance, unusual networking |
| One API plus database and cache | Managed containers plus managed data services | Scale, cost, or control requirements the platform cannot meet |
| Several services owned by one small team | Managed container service | Coordination burden becomes measurable |
| Many services and platform teams | Managed Kubernetes may fit | The decision depends on organisational as much as technical needs |
Kubernetes is not the graduation ceremony
A smaller architecture that meets its reliability, security, cost, and delivery requirements is not less professional. Choosing it deliberately is the architectural skill.
Check your understanding
Your two-person team runs one API on a managed container platform. It restarts failed containers, deploys gradually, and scales to the current traffic. What problem would Kubernetes solve today?
None has been established. Kubernetes might become useful after concrete platform, scale, scheduling, networking, or organisational requirements appear. Adding it now would replace a working managed control plane with one the team must understand and operate.