Configuring the Knative Serving Operator custom resource

The Knative Serving Operator can be configured with the following options:

Version configuration

Cluster administrators can install a specific version of Knative Serving by using the spec.version field.

For example, if you want to install Knative Serving v0.23.0, you can apply the following KnativeServing custom resource:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. version: 0.23.0

If spec.version is not specified, the Knative Operator installs the latest available version of Knative Serving. If users specify an invalid or unavailable version, the Knative Operator will do nothing. The Knative Operator always includes the latest 3 minor release versions. For example, if the current version of the Knative Operator is v0.24.0, the earliest version of Knative Serving available through the Operator is v0.22.0.

If Knative Serving is already managed by the Operator, updating the spec.version field in the KnativeServing resource enables upgrading or downgrading the Knative Serving version, without needing to change the Operator.

Important

The Knative Operator only permits upgrades or downgrades by one minor release version at a time. For example, if the current Knative Serving deployment is version v0.22.0, you must upgrade to v0.23.0 before upgrading to v0.24.0.

Knative Serving configuration by ConfigMap

The Operator manages the Knative Serving installation. It overwrites any updates to ConfigMaps which are used to configure Knative Serving. The KnativeServing custom resource (CR) allows you to set values for these ConfigMaps by using the Operator. Knative Serving has multiple ConfigMaps that are named with the prefix config-. The spec.config in the KnativeServing CR has one <name> entry for each ConfigMap, named config-<name>, with a value which will be used for the ConfigMap data.

In the setup a custom domain example, you can see the content of the ConfigMap config-domain is:

  1. apiVersion: v1
  2. kind: ConfigMap
  3. metadata:
  4. name: config-domain
  5. namespace: knative-serving
  6. data:
  7. example.org: |
  8. selector:
  9. app: prod
  10. example.com: ""

Using the operator, specify the ConfigMap config-domain using the operator CR:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. config:
  8. domain:
  9. example.org: |
  10. selector:
  11. app: prod
  12. example.com: ""

You can apply values to multiple ConfigMaps. This example sets stable-window to 60s in config-autoscaler as well as specifying config-domain:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. config:
  8. domain:
  9. example.org: |
  10. selector:
  11. app: prod
  12. example.com: ""
  13. autoscaler:
  14. stable-window: "60s"

All the ConfigMaps are created in the same namespace as the operator CR. You can use the operator CR as the unique entry point to edit all of them.

Private repository and private secrets

You can use the spec.registry section of the operator CR to change the image references to point to a private registry or specify imagePullSecrets:

  • default: this field defines a image reference template for all Knative images. The format is example-registry.io/custom/path/${NAME}:{CUSTOM-TAG}. If you use the same tag for all your images, the only difference is the image name. ${NAME} is a pre-defined variable in the operator corresponding to the container name. If you name the images in your private repo to align with the container names ( activator, autoscaler, controller, webhook, autoscaler-hpa, net-istio-controller, and queue-proxy), the default argument should be sufficient.

  • override: a map from container name to the full registry location. This section is only needed when the registry images do not match the common naming format. For containers whose name matches a key, the value is used in preference to the image name calculated by default. If a container’s name does not match a key in override, the template in default is used.

  • imagePullSecrets: a list of Secret names used when pulling Knative container images. The Secrets must be created in the same namespace as the Knative Serving Deployments. See deploying images from a private container registry for configuration details.

Download images in a predefined format without secrets:

This example shows how you can define custom image links that can be defined in the CR using the simplified format docker.io/knative-images/${NAME}:{CUSTOM-TAG}.

In the following example:

  • the custom tag v0.13.0 is used for all images
  • all image links are accessible without using secrets
  • images are pushed as docker.io/knative-images/${NAME}:{CUSTOM-TAG}

First, you need to make sure your images pushed to the following image tags:

ContainerDocker Image
activatordocker.io/knative-images/activator:v0.13.0
autoscalerdocker.io/knative-images/autoscaler:v0.13.0
controllerdocker.io/knative-images/controller:v0.13.0
webhookdocker.io/knative-images/webhook:v0.13.0
autoscaler-hpadocker.io/knative-images/autoscaler-hpa:v0.13.0
net-istio-controllerdocker.io/knative-images/net-istio-controller:v0.13.0
queue-proxydocker.io/knative-images/queue-proxy:v0.13.0

Then, you need to define your operator CR with following content:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. registry:
  8. default: docker.io/knative-images/${NAME}:v0.13.0

Download images individually without secrets

If your custom image links are not defined in a uniform format by default, you will need to individually include each link in the CR.

For example, to given the following images:

ContainerDocker Image
activatordocker.io/knative-images-repo1/activator:v0.13.0
autoscalerdocker.io/knative-images-repo2/autoscaler:v0.13.0
controllerdocker.io/knative-images-repo3/controller:v0.13.0
webhookdocker.io/knative-images-repo4/webhook:v0.13.0
autoscaler-hpadocker.io/knative-images-repo5/autoscaler-hpa:v0.13.0
net-istio-controllerdocker.io/knative-images-repo6/prefix-net-istio-controller:v0.13.0
net-istio-webhookdocker.io/knative-images-repo6/net-istio-webhooko:v0.13.0
queue-proxydocker.io/knative-images-repo7/queue-proxy-suffix:v0.13.0

The Operator CR should be modified to include the full list:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. registry:
  8. override:
  9. activator: docker.io/knative-images-repo1/activator:v0.13.0
  10. autoscaler: docker.io/knative-images-repo2/autoscaler:v0.13.0
  11. controller: docker.io/knative-images-repo3/controller:v0.13.0
  12. webhook: docker.io/knative-images-repo4/webhook:v0.13.0
  13. autoscaler-hpa: docker.io/knative-images-repo5/autoscaler-hpa:v0.13.0
  14. net-istio-controller: docker.io/knative-images-repo6/prefix-net-istio-controller:v0.13.0
  15. net-istio-webhook/webhook: docker.io/knative-images-repo6/net-istio-webhook:v0.13.0
  16. queue-proxy: docker.io/knative-images-repo7/queue-proxy-suffix:v0.13.0

Note

If the container name is not unique across all Deployments, DaemonSets and Jobs, you can prefix the container name with the parent container name and a slash. For example, istio-webhook/webhook.

Download images with secrets

If your image repository requires private secrets for access, include the imagePullSecrets attribute.

This example uses a secret named regcred. You must create your own private secrets if these are required:

After you create this secret, edit the Operator CR by appending the following content:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. registry:
  8. ...
  9. imagePullSecrets:
  10. - name: regcred

The field imagePullSecrets expects a list of secrets. You can add multiple secrets to access the images as follows:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. registry:
  8. ...
  9. imagePullSecrets:
  10. - name: regcred
  11. - name: regcred-2
  12. ...

SSL certificate for controller

To enable tag to digest resolution, the Knative Serving controller needs to access the container registry. To allow the controller to trust a self-signed registry cert, you can use the Operator to specify the certificate using a ConfigMap or Secret.

Specify the following fields in spec.controller-custom-certs to select a custom registry certificate:

  • name: the name of the ConfigMap or Secret.
  • type: either the string “ConfigMap” or “Secret”.

If you create a ConfigMap named testCert containing the certificate, change your CR:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. controller-custom-certs:
  8. name: testCert
  9. type: ConfigMap

Configuration of Knative ingress gateway

To set up custom ingress gateway, follow Step 1: Create Gateway Service and Deployment Instance.

Step 2: Update the Knative gateway

Update spec.ingress.istio.knative-ingress-gateway to select the labels of the new ingress gateway:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. ingress:
  8. istio:
  9. enabled: true
  10. knative-ingress-gateway:
  11. selector:
  12. custom: ingressgateway

Step 3: Update Gateway ConfigMap

Additionally, you will need to update the Istio ConfigMap:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. ingress:
  8. istio:
  9. enabled: true
  10. knative-ingress-gateway:
  11. selector:
  12. custom: ingressgateway
  13. config:
  14. istio:
  15. gateway.knative-serving.knative-ingress-gateway: "custom-ingressgateway.istio-system.svc.cluster.local"

The key in spec.config.istio is in the format of gateway...

Configuration of cluster local gateway

Update spec.ingress.istio.knative-local-gateway to select the labels of the new cluster-local ingress gateway:

Default local gateway name:

Go through the guide here to use local cluster gateway, if you use the default gateway called knative-local-gateway.

Non-default local gateway name:

If you create custom local gateway with a name other than knative-local-gateway, update config.istio and the knative-local-gateway selector:

This example shows a service and deployment knative-local-gateway in the namespace istio-system, with the label custom: custom-local-gw:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. ingress:
  8. istio:
  9. enabled: true
  10. knative-local-gateway:
  11. selector:
  12. custom: custom-local-gateway
  13. config:
  14. istio:
  15. local-gateway.knative-serving.knative-local-gateway: "custom-local-gateway.istio-system.svc.cluster.local"

High availability

By default, Knative Serving runs a single instance of each controller. The spec.high-availability field allows you to configure the number of replicas for the following leader-elected controllers: controller, autoscaler-hpa, net-istio-controller. This field also configures the HorizontalPodAutoscaler resources for the data plane (activator):

The following configuration specifies a replica count of 3 for the controllers and a minimum of 3 activators (which may scale higher if needed):

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. high-availability:
  8. replicas: 3

System Resource Settings

The operator custom resource allows you to configure system resources for the Knative system containers. Requests and limits can be configured for the following containers: activator, autoscaler, controller, webhook, autoscaler-hpa, net-istio-controller and queue-proxy.

To override resource settings for a specific container, create an entry in the spec.resources list with the container name and the Kubernetes resource settings.

For example, the following KnativeServing resource configures the activator to request 0.3 CPU and 100MB of RAM, and sets hard limits of 1 CPU, 250MB RAM, and 4GB of local storage:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. resources:
  8. - container: activator
  9. requests:
  10. cpu: 300m
  11. memory: 100Mi
  12. limits:
  13. cpu: 1000m
  14. memory: 250Mi
  15. ephemeral-storage: 4Gi

If you would like to add another container autoscaler with the same configuration, you need to change your CR as follows:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: knative-serving
  5. namespace: knative-serving
  6. spec:
  7. resources:
  8. - container: activator
  9. requests:
  10. cpu: 300m
  11. memory: 100Mi
  12. limits:
  13. cpu: 1000m
  14. memory: 250Mi
  15. ephemeral-storage: 4Gi
  16. - container: autoscaler
  17. requests:
  18. cpu: 300m
  19. memory: 100Mi
  20. limits:
  21. cpu: 1000m
  22. memory: 250Mi
  23. ephemeral-storage: 4Gi

Override system deployments

If you would like to override some configurations for a specific deployment, you can override the configuration by using spec.deployments in CR. Currently replicas, labels, annotations and nodeSelector are supported.

Override replicas, labels and annotations

The following KnativeServing resource overrides the webhook deployment to have 3 Replicas, the label mylabel: foo, and the annotation myannotataions: bar, while other system deployments have 2 Replicas by using spec.high-availability.

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: ks
  5. namespace: knative-serving
  6. spec:
  7. high-availability:
  8. replicas: 2
  9. deployments:
  10. - name: webhook
  11. replicas: 3
  12. labels:
  13. mylabel: foo
  14. annotations:
  15. myannotataions: bar

Note

The KnativeServing resource label and annotation settings override the webhook’s labels and annotations for both Deployments and Pods.

Override nodeSelector

The following KnativeServing resource overrides the webhook deployment to use the disktype: hdd nodeSelector:

  1. apiVersion: operator.knative.dev/v1alpha1
  2. kind: KnativeServing
  3. metadata:
  4. name: ks
  5. namespace: knative-serving
  6. spec:
  7. deployments:
  8. - name: webhook
  9. nodeSelector:
  10. disktype: hdd