Cost Optimisation

Kubernetes makes it easy to start resources, and just as easy to forget them. Without care, a cluster slowly becomes a very expensive hobby. The good news is that most savings come from a handful of habits. In this lesson, we walk through them, from the biggest savings, to the smallest.

• Easy to start, easy to forget
• A handful of habits
• Biggest savings first

Where the money goes

The bill has a clear shape. Most of it is worker nodes, the virtual machines that run your pods. Then storage, disks and snapshots. Then network, especially traffic between zones, and out to the internet. And last, load balancers, which are billed by the hour, even when idle. The largest lever is, how efficiently your pods fill your nodes. If the nodes are half empty, you are paying double. So we start there.

1. Worker nodesthe biggest line
2. Storagedisks and snapshots
3. Networkcross-zone and egress
4. Load balancersbilled even when idle

The savings checklist

Here is the checklist, by impact. First, right size requests, using real usage from your metrics, because inflated requests waste whole nodes. Second, autoscale pods, and nodes, so that capacity follows demand. Third, use spot, or preemptible nodes, for fault tolerant workloads, with big discounts. Fourth, remove idle resources, unused load balancers, old volumes, forgotten test namespaces. Fifth, share one ingress, instead of many load balancers. And sixth, set quotas, and labels by team, so that costs are visible, and owned.

Measure before you cut

Never cut blindly. Measure first. Use your monitoring system, to see requested versus actually used resources, per workload. Use cost tools, that break down spending by namespace, and label. Set alerts on sudden jumps. Change one thing at a time, and watch quality, and latency, after each change. A saving that causes an outage is not a saving. Review the numbers monthly, and make cost one of the standing topics, of your team's weekly rhythm.

Summary