LearnGlossary

What are resource requests in Kubernetes?

Updated 25 October 2026 · Published by Zoorik

A resource request in Kubernetes is how much CPU, memory or another resource a container asks for. The scheduler uses it to pick a node. Then the kubelet, the node’s agent, reserves at least that much.

A pod goes on a node only if its requests fit in the room left for pods. Low real use doesn’t change that.

The new pod doesn't fit, though most of the node's CPU is idle Illustration with no CPU amounts, for one node. Three rows share one bar width, marked by a bracket labelled CPU the node offers pods. New pod, its request: a block that starts where the free room starts and runs past a dashed line labelled node's edge, marked Doesn't fit. Requested, what the scheduler counts: segments pod 1, pod 2 and pod 3, then a short segment labelled free. Real use, doesn't change the fit: a short segment labelled used, and the rest of the bar labelled idle. The new pod doesn't fit, though most of the node's CPU is idle node's edgeNew podits requestDoesn't fitRequestedwhat the scheduler countspod 1pod 2pod 3freeReal usedoesn't change the fitusedidleCPU the node offers pods
Illustration, with no CPU amounts: the scheduler fits a new pod by its request, so CPU that pods request but don’t use is not free room. Segment lengths are not measurements.

CPU and memory requests

  • A CPU request in Kubernetes is measured in CPU units. 1 CPU is one physical or virtual core, and 100m is 0.1 CPU.
  • Memory is in bytes. 64Mi is 64 mebibytes.
  • Set only a limit, and Kubernetes copies it as the request, unless a default request applies.
  • Since Kubernetes v1.34 (beta), a whole pod can also have requests.

An example from the Kubernetes docs

A pod has two containers that each request 0.25 CPU and 64 MiB. The pod’s request is their sum: 0.5 CPU and 128 MiB.

Why it matters for your bill

The Cluster Autoscaler, which adds and removes nodes, works from Kubernetes resource requests.

  • It adds a node when pods can’t fit.
  • By default, it removes a node only when its pods request under 50% of the node’s CPU and memory for pods.

So your bill follows requests.

Coral rebalances CPU and memory requests on AKS and EKS to match real use, so the same pods fit on fewer nodes. Then it drains the spare nodes, so your autoscaler can remove them. The free assessment works out your own figure.

For requests vs limits, and how to set each one, read Kubernetes pod rebalancing.

Read next: Kubernetes pod rebalancing

Sources: Kubernetes, “Resource Management for Pods and Containers” (opens in a new tab); Kubernetes Cluster Autoscaler on GitHub, “Frequently Asked Questions” (opens in a new tab).

Find your own number

One line in your cloud console. One PDF report with your estimated saving at list price.

Estimate your saving

Talk to the builders

Thirty minutes with the people who build Coral.

Book a demo (opens in a new tab)

Or write to founders@zoorik.com