# Production Engineering

Getting a workload to run on Kubernetes is the easy part. This section is the discipline that separates a cluster that survives a bad Tuesday from one that doesn't: sizing it correctly, controlling its cost, deciding whether you need more than one, planning for the day it fails outright, and proving — with a checklist, not a feeling — that a workload is actually ready for production traffic.

## Read in this order

1. [Cluster Sizing and Capacity Planning](<https://computecentral.in/kubernetes/production-engineering/01-cluster-sizing-and-capacity-planning/index.md>) — the math behind node sizing, reserved overhead, and etcd's hard limits
2. [Cost Optimization](<https://computecentral.in/kubernetes/production-engineering/02-cost-optimization/index.md>) — right-sizing, spot capacity, autoscalers, and where the money actually goes
3. [Multi-Cluster and Multi-Region](<https://computecentral.in/kubernetes/production-engineering/03-multi-cluster-and-multi-region/index.md>) — why teams split clusters, and the patterns that actually get used
4. [Disaster Recovery](<https://computecentral.in/kubernetes/production-engineering/04-disaster-recovery/index.md>) — RTO/RPO, backups, failover, and what a real runbook contains
5. [Production Readiness Checklist](<https://computecentral.in/kubernetes/production-engineering/05-production-readiness-checklist/index.md>) — the gate a workload should pass before it takes real traffic

> [!TIP]
>
> **This section assumes the fundamentals**
>
> If you haven't yet covered workloads, networking, storage, and security individually, work through those sections first — this one is about combining them under real operational pressure, not teaching them from scratch.

## Next

Continue to [AI Workloads](<https://computecentral.in/kubernetes/ai-workloads/index.md>) for GPUs, model serving, and AI agents, or to [OpenShift](<https://computecentral.in/kubernetes/openshift/index.md>) to see how a major enterprise Kubernetes distribution packages many of these same production concerns into the platform itself.
