The twelve-factor app predates Kubernetes, yet it still explains why some services are trivial to deploy and scale while others fight you at every step. Here are the factors that matter most in a container world.
Config in the environment
Keep config out of the code and inject it as environment variables. In Kubernetes, that means ConfigMaps and Secrets mounted as env, not values baked into the image.
env:
- name: DATABASE_URL
valueFrom: { secretKeyRef: { name: db, key: url } }
- name: LOG_LEVEL
valueFrom: { configMapKeyRef: { name: app, key: log_level } }Stateless processes and disposability
Processes should be stateless and disposable - fast to start, graceful to stop. Store state in backing services (databases, object storage, Redis), and handle SIGTERM so rolling updates do not drop requests.
- Treat backing services as attached resources swapped via config
- Build, release and run are strictly separated stages
- Logs are event streams to stdout, not files you manage
- Admin tasks run as one-off processes in the same environment
If a service is hard to deploy, it usually violates two or three factors. Fix those and the operations problem quietly disappears.
You do not need to worship all twelve. But when a service is painful to run, walk the list - the culprit is almost always there.