Knative Serving resource management

This topic tells you how to configure the memory and CPU allocation of resources in the knative-serving namespace.

Overview

By default, Knative deployments are allocated a predefined amount of CPU and memory. These default allocations support general use cases, but you might adjust the allocations for the following reasons:

• Performance Optimization: Customize resource allocations to help fine-tune the load for specific workloads to improve response times and throughput.
• Cost Efficiency: Customize resources to avoid over-provisioning or under-provisioning and verify efficient resource use.
• Improved Stability: Control resource allocation to prevent any deployment from consuming excessive resources and increase the stability of the entire cluster.

You can tailor the memory and CPU of Knative system controllers by using the Cloud Native Runtimes resource_management config option.

Configuring memory and cpu requests and limits for Knative Serving resources

To configure the memory and CPU allocation for the deployments in the knative-serving namespace, you must specify the resource_management config option on Cloud Native Runtimes. You can only configure the following deployments:

• activator
• autoscaler
• autoscaler-hpa
• controller
• net-certmanager-controller
• net-certmanager-webhook
• net-contour-controller
• webhook

Example: updating the activator deployment

The following example shows how to adjust the CPU and memory requests and limits by using the resource_management option of the Knative activator deployment. By default, the activator deployment has the following resources:

resources:
    requests:
        cpu: 300m
        memory: 60Mi
    limits:
        cpu: 1000m
        memory: 600Mi

Specify the following resources to the activator pod:

cnrs:
  resource_management:
    - name: NAME
      requests:
        memory: MEMORY
        cpu: CPU
      limits:
        memory: MEMORY
        cpu: CPU

Where:

• NAME represents the name of the deployment you want to configure.
• requests specifies the amount of resources that are allocated to the pods of this deployment:
  • MEMORY is how many megabytes of memory you want to request.
  • CPU is how many CPU cores, 100 millicpu units, you want to request.
• limits sets the maximum amount of resources that the pods can use:
  • MEMORY is the limit of how many megabytes of memory that the pods can use.
  • CPU is the limit of how many CPU cores, 100 millicpu units, pods can use.

Note

You can provide any of the resource units supported by Kubernetes as explained in the Kubernetes documentation.

After updating Cloud Native Runtimes, get the deployment information to confirm the activator deployment updated:

kubectl get deployment activator -n knative-serving -o=jsonpath='{.spec.template.spec.containers[?(@.name=="activator")].resources}'

{"limits":{"cpu":"1","memory":"1000Mi"},"requests":{"cpu":"100m","memory":"100Mi"}}

All fields, such as the request and limits of CPUs, and the memory’s request and limits, are optional, except for the deployment’s name, causing it to use the default values specified for these deployments.

For example, to change only the limit and the memory’s request of the Knative controller, you can provide the following configuration to resource_management:

cnrs:
  resource_management:
    - name: "controller"
      requests:
        memory: "130M"
      limits:
        cpu: "0.5"