Skip to content
KubernetesDockerDelivery

Kubernetes

Kubernetes schedules processes. It does not define your bounded contexts, your data model, or your on-call rotation.

Personal experience

At Armirene, DevOps work included Kubernetes and Docker on an ecommerce platform. Later ProCrédito/Soga v2 work deploys Laravel services, Nginx sidecars, and Python workers on Kubernetes, including Helm for migrate jobs. Cluster topology and node counts are not published (TODO).

Technical perspective

If the application cannot start from a container image, with configuration injected, and declare a health check, Kubernetes will only make that more obvious.

Health probes are part of the contract. A process that reports ready before it can serve is how rolling deploys drop traffic. A process that never reports not-ready is how bad deploys linger.

Start with Deployments, Services, ConfigMaps, and Secrets. Add operators, meshes, and custom schedulers when a measured problem requires them. The control plane is a distributed system of its own; it is not free.

Architecture

Cluster as a runtime, not a domain

Container image

Kubernetes

API pods

Worker pods

Managed data / cache

Technical perspective, consistent with Docker/Kubernetes delivery at Armirene.