Skip to main content

// tool

Kubernetes Resource Calculator

Plan Kubernetes requests and limits without guessing. See node-fit, cluster totals, and recommended burst.

// observed peak usage

// per-pod requests
cpu325m
memory666Mi
// per-pod limits
cpu
memory1.30Gi

// cluster totals for 3 replicas

cpu requested

975m

memory requested

1.95Gi

// resources block
resources:
  requests:
    cpu: "325m"
    memory: "666Mi"
  limits:
    memory: "1.30Gi"

How to size Kubernetes Pods

Start from real numbers

Measure actual peak usage from a running Pod with kubectl top pod or Prometheus. Guessing leads to overprovisioning (wastes money) or underprovisioning (evictions and OOMKills).

Then add headroom

Workloads have bursts your metrics didn't capture. 25-50% headroom is typical. If traffic is unpredictable, lean higher. If cost matters more than reliability, lean lower.

Why CPU limits are controversial

Kubernetes throttles containers that exceed their CPU limit, even when the node has idle CPU. This can cause tail-latency spikes. Many teams set CPU requests (for scheduling fairness) and leave CPU limits unset. Memory limits are different because memory can't be throttled and has to be bounded somehow.

// related simulators

Sponsored
Carbon Ads