Don't Be Greedy: Rightsize Your Kubernetes Cluster With Prometheus
summary
Rightsize Kubernetes by measuring actual container usage with Prometheus, pulling metrics from cAdvisor and kube-state-metrics, computing requests by either max (conservative) or average (aggressive), then iterating and re-measuring. Example: a workload with 4 replicas reserved 4 GiB (1 GiB each) was reduced to 128 MiB requests, moving from ~20% utilization of reserved memory to nearly 100% and saving about $75/month for that single workload. Repeatable dashboards and prioritizing the heaviest namespaces keep the effort efficient.
AI-written summary · about
chapters (15)
- 0:00Start
- 1:21The greedy developer problem
- 3:08What are the limits?
- 3:26What are the requests?
- 3:39How looks like in a real scenario
- 4:38QoS Classes
- 7:23Quick review
- 7:40Limit rightsize strategies
- 8:41Request rightsize strategies
- 8:54Rightsizing requests will help to
- 11:46Calculate % unused Memory
- 12:09Detect unused Memory
- 13:13Rightsize the requests
- 16:11Checking the impact
- 18:24Conclusions
related talks
29mRight-Size Your Kubernetes Deployments
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.
39mStaying in Tune: Optimize Kubernetes for Stability and Utilization
32mOptimizing Application Performance on Kubernetes
35mDeep Dive: Kubernetes (Instrumentation) SIG
11mHow Jenga & Tetris Can Help Run More Containers with Lesser Costs...
16mShrinking the Observability Bill
11mK8s Resources Ensuring Performance and Optimization