Practical guide
How to Right-Size Kubernetes Ephemeral Storage
A practical method for sizing ephemeral-storage requests and limits from logs, writable layers, emptyDir data, and eviction headroom.
Updated · Sources checked
Right-sizing ephemeral storage starts with measurement, not a generic per-pod number. Record peak writable-layer, log, and emptyDir usage for a representative workload window, then add explicit headroom for deploys, retries, and log bursts.
Requests and limits are different
The request influences scheduling; the limit caps the pod’s configured ephemeral-storage usage. Set the request near a measured peak plus operational headroom, and set the limit high enough to avoid normal bursts without allowing unbounded disk growth.
Use the ephemeral storage calculator to aggregate those per-pod values across replicas. Then compare the total with node allocatable storage and disk-pressure eviction thresholds. The pod capacity calculator can help with CPU and memory, but it does not model disk pressure.
Three checks before production
- Confirm logs have rotation and retention limits.
- Give temporary files and
emptyDirdata an explicit size limit where appropriate. - Load-test a rollout and failure retry, because those events often create the largest short-lived disk spike.