Skip to content
Deadline to Die

Module 8: Scaling & Rollouts

🏗️ 1. The Manual Task: Maintenance Window Deployments

Before orchestration, releasing a new version of an application usually required a “maintenance window” where the application was taken offline.

The Manual Steps:

  1. Take App Offline: Tell customers the site will be down.
  2. Download App: SSH into all servers and pull the new version.
  3. Restart App: Manually start version 2 on every machine.
  4. Frantic Rollback: If it crashes, manually revert back to version 1.

📈 2. The Kubernetes Abstraction: Deployments & HPA

In Kubernetes, you rarely use Pods directly. You use a Deployment.

  • Rolling Updates: Automatically starts one V2 pod, waits for it to be “Ready”, then kills one V1 pod. Zero downtime.
  • Rollback: If V2 is buggy, you can undo it with a single command (kubectl rollout undo).
  • Horizontal Pod Autoscaler (HPA): Automatically scales up Pod count based on CPU usage.