• derekwaynecarr
  • janetkuo title: Setting Pod CPU and Memory Limits

By default, pods run with unbounded CPU and memory limits. This means that any pod in the system will be able to consume as much CPU and memory on the node that executes the pod.

Users may want to impose restrictions on the amount of resources a single pod in the system may consume for a variety of reasons.

For example:

  1. Each node in the cluster has 2GB of memory. The cluster operator does not want to accept pods that require more than 2GB of memory since no node in the cluster can support the requirement. To prevent a pod from being permanently unscheduled to a node, the operator instead chooses to reject pods that exceed 2GB of memory as part of admission control.
  2. A cluster is shared by two communities in an organization that runs production and development workloads respectively. Production workloads may consume up to 8GB of memory, but development workloads may consume up to 512MB of memory. The cluster operator creates a separate namespace for each workload, and applies limits to each namespace.
  3. Users may create a pod which consumes resources just below the capacity of a machine. The left over space may be too small to be useful, but big enough for the waste to be costly over the entire cluster. As a result, the cluster operator may want to set limits that a pod must consume at least 20% of the memory and CPU of their average node size in order to provide for more uniform scheduling and limit waste.

This example demonstrates how limits can be applied to a Kubernetes namespace to control min/max resource limits per pod. In addition, this example demonstrates how you can apply default resource limits to pods in the absence of an end-user specified value.

See LimitRange design doc for more information. For a detailed description of the Kubernetes resource model, see Resources

Step 0: Prerequisites

This example requires a running Kubernetes cluster. See the Getting Started guides for how to get started.

Change to the <kubernetes> directory if you're not already there.

Step 1: Create a namespace

This example will work in a custom namespace to demonstrate the concepts involved.

Let's create a new namespace called limit-example:

$ kubectl create namespace limit-example
namespace "limit-example" created

Note that kubectl commands will print the type and name of the resource created or mutated, which can then be used in subsequent commands:

$ kubectl get namespaces
NAME            STATUS    AGE
default         Active    51s
limit-example   Active    45s

Step 2: Apply a limit to the namespace

Let's create a simple limit in our namespace.

$ kubectl create -f docs/admin/limitrange/limits.yaml --namespace=limit-example
limitrange "mylimits" created

Let's describe the limits that we have imposed in our namespace.

$ kubectl describe limits mylimits --namespace=limit-example
Name:   mylimits
Namespace:  limit-example
Type        Resource      Min      Max      Default Request      Default Limit      Max Limit/Request Ratio
----        --------      ---      ---      ---------------      -------------      -----------------------
Pod         cpu           200m     2        -                    -                  -
Pod         memory        6Mi      1Gi      -                    -                  -
Container   cpu           100m     2        200m                 300m               -
Container   memory        3Mi      1Gi      100Mi                200Mi              -

In this scenario, we have said the following:

  1. If a max constraint is specified for a resource (2 CPU and 1Gi memory in this case), then a limit must be specified for that resource across all containers. Failure to specify a limit will result in a validation error when attempting to create the pod. Note that a default value of limit is set by default in file limits.yaml (300m CPU and 200Mi memory).
  2. If a min constraint is specified for a resource (100m CPU and 3Mi memory in this case), then a request must be specified for that resource across all containers. Failure to specify a request will result in a validation error when attempting to create the pod. Note that a default value of request is set by defaultRequest in file limits.yaml (200m CPU and 100Mi memory).
  3. For any pod, the sum of all containers memory requests must be >= 6Mi and the sum of all containers memory limits must be <= 1Gi; the sum of all containers CPU requests must be >= 200m and the sum of all containers CPU limits must be <= 2.

Step 3: Enforcing limits at point of creation

The limits enumerated in a namespace are only enforced when a pod is created or updated in the cluster. If you change the limits to a different value range, it does not affect pods that were previously created in a namespace.

If a resource (CPU or memory) is being restricted by a limit, the user will get an error at time of creation explaining why.

Let's first spin up a Deployment that creates a single container Pod to demonstrate how default values are applied to each pod.

$ kubectl run nginx --image=nginx --replicas=1 --namespace=limit-example
deployment "nginx" created

Note that kubectl run creates a Deployment named "nginx" on Kubernetes cluster >= v1.2. If you are running older versions, it creates replication controllers instead. If you want to obtain the old behavior, use --generator=run/v1 to create replication controllers. See kubectl run for more details. The Deployment manages 1 replica of single container Pod. Let's take a look at the Pod it manages. First, find the name of the Pod:

$ kubectl get pods --namespace=limit-example
NAME                     READY     STATUS    RESTARTS   AGE
nginx-2040093540-s8vzu   1/1       Running   0          11s

Let's print this Pod with yaml output format (using -o yaml flag), and then grep the resources field. Note that your pod name will be different.

$ kubectl get pods nginx-2040093540-s8vzu --namespace=limit-example -o yaml | grep resources -C 8
  resourceVersion: "57"
  selfLink: /api/v1/namespaces/limit-example/pods/nginx-2040093540-ivimu
  uid: 67b20741-f53b-11e5-b066-64510658e388
  - image: nginx
    imagePullPolicy: Always
    name: nginx
        cpu: 300m
        memory: 200Mi
        cpu: 200m
        memory: 100Mi
    terminationMessagePath: /dev/termination-log

Note that our nginx container has picked up the namespace default CPU and memory resource limits and requests.

Let's create a pod that exceeds our allowed limits by having it have a container that requests 3 CPU cores.

$ kubectl create -f docs/admin/limitrange/invalid-pod.yaml --namespace=limit-example
Error from server: error when creating "docs/admin/limitrange/invalid-pod.yaml": Pod "invalid-pod" is forbidden: [Maximum cpu usage per Pod is 2, but limit is 3., Maximum cpu usage per Container is 2, but limit is 3.]

Let's create a pod that falls within the allowed limit boundaries.

$ kubectl create -f docs/admin/limitrange/valid-pod.yaml --namespace=limit-example
pod "valid-pod" created

Now look at the Pod's resources field:

$ kubectl get pods valid-pod --namespace=limit-example -o yaml | grep -C 6 resources
  uid: 3b1bfd7a-f53c-11e5-b066-64510658e388
  - image: gcr.io/google_containers/serve_hostname
    imagePullPolicy: Always
    name: kubernetes-serve-hostname
        cpu: "1"
        memory: 512Mi
        cpu: "1"
        memory: 512Mi

Note that this pod specifies explicit resource limits and requests so it did not pick up the namespace default values.

Note: The limits for CPU resource are enforced in the default Kubernetes setup on the physical node that runs the container unless the administrator deploys the kubelet with the following flag:

$ kubelet --help
Usage of kubelet
  --cpu-cfs-quota[=true]: Enable CPU CFS quota enforcement for containers that specify CPU limits
$ kubelet --cpu-cfs-quota=false ...

Step 4: Cleanup

To remove the resources used by this example, you can just delete the limit-example namespace.

$ kubectl delete namespace limit-example
namespace "limit-example" deleted
$ kubectl get namespaces
NAME            STATUS        AGE
default         Active        12m


Cluster operators that want to restrict the amount of resources a single container or pod may consume are able to define allowable ranges per Kubernetes namespace. In the absence of any explicit assignments, the Kubernetes system is able to apply default resource limits and requests if desired in order to constrain the amount of resource a pod consumes on a node.