Use these variables to determine if you need to right size your Kubernetes cluster.
đź—ť Key metrics
- CPU
- Memory
- Storage (usually this isn’t a limiting actor)
- GPU (if you have it)
You will need to collect data on these metrics over a period of time. Preferably 3+ months and try to include any peak seasons or events.
We then take the average of the 95th percentile of all requests. We disregard the top 5% of requests in order to get rid of any anomalies while still accounting for peak usage.
If your averages are over 80% or under 60% of your possible utilization, then now is a good time to consider a right sizing.
P.S. When was the last time you did this for your resources?