Infrastructure & Operations › Kubernetes & Orchestration

Operator

A custom controller that automates running complex software.

Also known as: operator, kubernetes operator, k8s operator

A Kubernetes operator is a controller that encodes the operational knowledge of running a piece of software. Instead of a human following a runbook to back up a database, reshard a queue or promote a replica, an operator watches for the state you declared and reconciles reality to match — the same loop Kubernetes uses for its built-in objects, extended to your application.

An operator is built from two pieces: one or more CustomResourceDefinitions that define the kinds it manages, and a controller that watches those objects and acts.

apiVersion: example.com/v1
kind: Database
metadata: { name: orders }
spec:
  replicas: 3
  version: "16"

From that, the operator creates the underlying StatefulSet, volumes, services and backups, and keeps them healthy — handling upgrades, failover and repair as the declared state changes.

The pattern matters for stateful, complex software: databases, message systems, storage. Running those by hand is exactly the toil that operators remove, and a good operator brings the same declarative, self-healing model that makes stateless workloads easy.

The classic mistakes:

• Expecting an operator to be magic. It automates a specific set of operations the author encoded. If it doesn’t handle the failure you hit, it won’t invent a fix. Read what it actually does.
• Installing one and never watching it. Operators can act destructively — delete volumes, fail over, restart — if misconfigured. They need monitoring and care, not set-and-forget.
• A CRD without the controller. Defining custom objects that nothing reconciles does nothing; the controller is the operator (see custom resource definitions).
• Writing your own too eagerly. Operators are nontrivial to build and maintain. Prefer a maintained operator unless you have highly specific needs.
• Treating backups as covered. An operator’s backup feature is only a backup if restores are tested. Verify (see backups).

When to use one: when you run complex stateful software on Kubernetes and want to automate its day-2 operations. On managed Kubernetes, the provider’s own controllers are operators too. The pattern is a natural extension of Kubernetes’s reconciling model — see container orchestration and configuration drift for the loop it embodies.