Kubernetes Tutorial · Configuration and Resources · Lesson 19 of 26

Resource Requests and Limits

Requests drive scheduling, limits cap usage; set both well to avoid OOMKills, throttling and noisy neighbours.

Create a free account to save this page, take notes and track your progress. Sign in

Requests and limits

resources:
  requests: {cpu: 250m, memory: 256Mi}   # reserved for scheduling
  limits:   {memory: 256Mi}              # hard cap

The scheduler places pods using requests; a pod whose requests don't fit anywhere stays Pending.
Exceeding a memory limit gets the container OOMKilled and restarted.
Exceeding a CPU limit throttles it; many teams set CPU requests but no CPU limit to avoid needless throttling.
Requests equal to limits for every resource give the Guaranteed QoS class (evicted last under pressure).

OOMKilled, Pending and QoS

kubectl create namespace demo-resources > /dev/null
trap 'kubectl delete namespace demo-resources --wait=false > /dev/null' EXIT
cat > hungry.yaml <<'EOF'
apiVersion: v1
kind: Pod
metadata: {name: hungry}
spec:
  restartPolicy: Never
  containers:
    - name: app
      image: busybox:1.37
      command: ["sh", "-c", "tail /dev/zero"]
      resources:
        requests: {memory: 32Mi}
        limits: {memory: 32Mi}
EOF
kubectl -n demo-resources apply -f hungry.yaml > /dev/null
kubectl -n demo-resources wait --for=jsonpath='{.status.phase}'=Failed pod/hungry --timeout=120s > /dev/null
kubectl -n demo-resources get pod hungry -o jsonpath='terminated: {.status.containerStatuses[0].state.terminated.reason}, exit code {.status.containerStatuses[0].state.terminated.exitCode}{"\n"}'
kubectl -n demo-resources run greedy --image=busybox:1.37 --overrides='{"spec":{"containers":[{"name":"greedy","image":"busybox:1.37","command":["sleep","3600"],"resources":{"requests":{"cpu":"500"}}}]}}' > /dev/null
sleep 3
echo "pod asking for 500 CPUs: $(kubectl -n demo-resources get pod greedy -o jsonpath='{.status.phase}') ($(kubectl -n demo-resources get pod greedy -o jsonpath='{.status.conditions[0].reason}'))"
echo "QoS of hungry: $(kubectl -n demo-resources get pod hungry -o jsonpath='{.status.qosClass}')"

kubectl create namespace demo-resources > /dev/null
trap 'kubectl delete namespace demo-resources --wait=false > /dev/null' EXIT
cat > hungry.yaml <<'EOF'
apiVersion: v1
kind: Pod
metadata: {name: hungry}
spec:
  restartPolicy: Never
  containers:
    - name: app
      image: busybox:1.37
      command: ["sh", "-c", "tail /dev/zero"]
      resources:
        requests: {memory: 32Mi}
        limits: {memory: 32Mi}
EOF
kubectl -n demo-resources apply -f hungry.yaml > /dev/null
kubectl -n demo-resources wait --for=jsonpath='{.status.phase}'=Failed pod/hungry --timeout=120s > /dev/null
kubectl -n demo-resources get pod hungry -o jsonpath='terminated: {.status.containerStatuses[0].state.terminated.reason}, exit code {.status.containerStatuses[0].state.terminated.exitCode}{"\n"}'
kubectl -n demo-resources run greedy --image=busybox:1.37 --overrides='{"spec":{"containers":[{"name":"greedy","image":"busybox:1.37","command":["sleep","3600"],"resources":{"requests":{"cpu":"500"}}}]}}' > /dev/null
sleep 3
echo "pod asking for 500 CPUs: $(kubectl -n demo-resources get pod greedy -o jsonpath='{.status.phase}') ($(kubectl -n demo-resources get pod greedy -o jsonpath='{.status.conditions[0].reason}'))"
echo "QoS of hungry: $(kubectl -n demo-resources get pod hungry -o jsonpath='{.status.qosClass}')"

Output

terminated: OOMKilled, exit code 137
pod asking for 500 CPUs: Pending (Unschedulable)
QoS of hungry: Burstable

Run this on your computer — see “Get Started” for setup.

See Resource management.

Test yourself

A pod stays Pending with reason Unschedulable. A likely cause?

Its image is missing
Its resource requests don't fit on any node
Its readiness probe fails
It has no Service

Show answer

Answer: Its resource requests don't fit on any node

kubectl describe pod shows the scheduler's message, for example 'Insufficient cpu'.

Was this page helpful?