Skip to content

Latest commit

 

History

History
283 lines (217 loc) · 10.2 KB

creating-a-kubernetes-cluster.md

File metadata and controls

283 lines (217 loc) · 10.2 KB
Raw

Creating a Kubernetes Cluster for Knative Serving

Two options:

GKE

To use a k8s cluster running in GKE:

  1. Install gcloud using the instructions for your platform.

  2. Create a GCP project (or use an existing project if you've already created one) at http://console.cloud.google.com/home/dashboard. Set the ID of the project in an environment variable (e.g. PROJECT_ID).

    If you are a new GCP user, you might be eligible for a trial credit making your GKE cluster and other resources free for a short time. Otherwise, any GCP resources you create will cost money.

  3. Enable the k8s API:

    gcloud --project=$PROJECT_ID services enable container.googleapis.com

  4. Create a k8s cluster (version 1.10 or greater):

    gcloud --project=$PROJECT_ID container clusters create \
  --cluster-version=latest \
  --zone=us-east1-d \
  --scopes=cloud-platform \
  --machine-type=n1-standard-4 \
  --enable-autoscaling --min-nodes=1 --max-nodes=3 \
  knative-demo
    • Version 1.10+ is required
    • Change this to whichever zone you choose
    • cloud-platform scope is required to access GCB
    • Knative Serving currently requires 4-cpu nodes to run conformance tests. Changing the machine type from the default may cause failures.
    • Autoscale from 1 to 3 nodes. Adjust this for your use case
    • Change this to your preferred cluster name

    You can see the list of supported cluster versions in a particular zone by running:

    # Get the list of valid versions in us-east1-d
gcloud container get-server-config --zone us-east1-d

  5. Alternately, if you wish to re-use an already-created cluster, you can fetch the credentials to your local machine with:

    # Load credentials for the new cluster in us-east1-d
gcloud container clusters get-credentials --zone us-east1-d knative-demo

  6. If you haven't installed kubectl yet, you can install it now with gcloud:

    gcloud components install kubectl

  7. Add to your .bashrc:

    # When using GKE, the K8s user is your GCP user.
export K8S_USER_OVERRIDE=$(gcloud config get-value core/account)

Minikube

  1. Install and configure minikube with a VM driver, e.g. kvm2 on Linux or hyperkit on macOS.

  2. Create a cluster with version 1.10 or greater and your chosen VM driver.

    The following commands will setup a cluster with 8192 MB of memory and 4 CPUs. If you want to enable metric and log collection, bump the memory to 12288 MB.

    Providing any admission control pluins overrides the default set provided by minikube so we must explicitly list all plugins we want enabled.

    Until minikube makes this the default, the certificate controller must be told where to find the cluster CA certs on the VM.

    For Linux use:

    minikube start --memory=8192 --cpus=4 \
--kubernetes-version=v1.10.5 \
--vm-driver=kvm2 \
--bootstrapper=kubeadm \
--extra-config=controller-manager.cluster-signing-cert-file="/var/lib/localkube/certs/ca.crt" \
--extra-config=controller-manager.cluster-signing-key-file="/var/lib/localkube/certs/ca.key" \
--extra-config=apiserver.admission-control="LimitRanger,NamespaceExists,NamespaceLifecycle,ResourceQuota,ServiceAccount,DefaultStorageClass,MutatingAdmissionWebhook"

    For macOS use:

    minikube start --memory=8192 --cpus=4 \
--kubernetes-version=v1.10.5 \
--vm-driver=hyperkit \
--bootstrapper=kubeadm \
--extra-config=controller-manager.cluster-signing-cert-file="/var/lib/localkube/certs/ca.crt" \
--extra-config=controller-manager.cluster-signing-key-file="/var/lib/localkube/certs/ca.key" \
--extra-config=apiserver.admission-control="LimitRanger,NamespaceExists,NamespaceLifecycle,ResourceQuota,ServiceAccount,DefaultStorageClass,MutatingAdmissionWebhook"

  3. Configure your shell environment to use your minikube cluster:

    export K8S_CLUSTER_OVERRIDE='minikube'
# When using Minikube, the K8s user is your local user.
export K8S_USER_OVERRIDE=$USER

  4. Install Knative Serving

    Before installing knative on minikube, we need to do two things:

    1. Workaround Minikube's lack of support for LoadBalancer type services
    2. Configure ko for local publishing

    After doing those, you can deploy the Knative Serving components to Minikube the same way you would any other Kubernetes cluster.

LoadBalancer Support in Minikube

By default istio uses a LoadBalancer which is not yet supported by Minikube but is required for Knative to function properly (Route endpoints MUST be available via a routable IP before they will be marked as ready). One possible workaround is to install a custom controller that provisions an external IP using the service's ClusterIP, which must be made routable on the minikube host. These two commands accomplish this, and should be run once whenever you start a new minikube cluster:

sudo ip route add $(cat ~/.minikube/profiles/minikube/config.json | jq -r ".KubernetesConfig.ServiceCIDR") via $(minikube ip)
kubectl run minikube-lb-patch --replicas=1 --image=elsonrodriguez/minikube-lb-patch:0.1 --namespace=kube-system

Minikube with ko

You can instruct ko to sideload images into your Docker daemon instead of publishing them to a registry by setting KO_DOCKER_REPO=ko.local:

# Use the minikube docker daemon (among other things)
eval $(minikube docker-env)

# Switch the current kubectl context to minikube
kubectl config use-context minikube

# Set KO_DOCKER_REPO to a sentinel value for ko to sideload into the daemon.
export KO_DOCKER_REPO="ko.local"

Enabling Knative to Use Images in Minikube

In order to have Knative access an image in Minikube's Docker daemon you should prefix your image name with the dev.local registry. This will cause Knative to use the cached image. You must not tag your image as latest since this causes Kubernetes to always attempt a pull.

For example:

eval $(minikube docker-env)
docker pull gcr.io/knative-samples/primer:latest
docker tag gcr.io/knative-samples/primer:latest dev.local/knative-samples/primer:v1

Minikube with GCR

You can use Google Container Registry as the registry for a Minikube cluster.

  1. Set up a GCR repo. Export the environment variable PROJECT_ID as the name of your project. Also export GCR_DOMAIN as the domain name of your GCR repo. This will be either gcr.io or a region-specific variant like us.gcr.io.

    export PROJECT_ID=knative-demo-project
export GCR_DOMAIN=gcr.io

    To publish builds push to GCR, set KO_DOCKER_REPO or DOCKER_REPO_OVERRIDE to the GCR repo's url.

    export KO_DOCKER_REPO="${GCR_DOMAIN}/${PROJECT_ID}"
export DOCKER_REPO_OVERRIDE="${KO_DOCKER_REPO}"

  2. Create a GCP service account:

    gcloud iam service-accounts create minikube-gcr \
  --display-name "Minikube GCR Pull" \
  --project $PROJECT_ID

  3. Give your service account the storage.objectViewer role:

    gcloud projects add-iam-policy-binding $PROJECT_ID \
  --member "serviceAccount:minikube-gcr@${PROJECT_ID}.iam.gserviceaccount.com" \
  --role roles/storage.objectViewer

  4. Create a key credential file for the service account:

    gcloud iam service-accounts keys create \
  --iam-account "minikube-gcr@${PROJECT_ID}.iam.gserviceaccount.com" \
  minikube-gcr-key.json

Now you can use the minikube-gcr-key.json file to create image pull secrets and link them to Kubernetes service accounts. A secret must be created and linked to a service account in each namespace that will pull images from GCR.

For example, use these steps to allow Minikube to pull Knative Serving and Build images from GCR as published in our development flow (ko apply -f config/). This is only necessary if you are not using public Knative Serving and Build images.

  1. Create a Kubernetes secret in the knative-serving and knative-build namespace:

    export [email protected]
kubectl create secret docker-registry "knative-serving-gcr" \
  --docker-server=$GCR_DOMAIN \
  --docker-username=_json_key \
  --docker-password="$(cat minikube-gcr-key.json)" \
  --docker-email=$DOCKER_EMAIL \
  -n "knative-serving"
kubectl create secret docker-registry "build-gcr" \
  --docker-server=$GCR_DOMAIN \
  --docker-username=_json_key \
  --docker-password="$(cat minikube-gcr-key.json)" \
  --docker-email=$DOCKER_EMAIL \
  -n "knative-build"

    The secret must be created in the same namespace as the pod or service account.

  2. Add the secret as an imagePullSecret to the controller and build-controller service accounts:

    kubectl patch serviceaccount "build-controller" \
  -p '{"imagePullSecrets": [{"name": "build-gcr"}]}' \
  -n "knative-build"
kubectl patch serviceaccount "controller" \
  -p '{"imagePullSecrets": [{"name": "knative-serving-gcr"}]}' \
  -n "knative-serving"

Use the same procedure to add imagePullSecrets to service accounts in any namespace. Use the default service account for pods that do not specify a service account.

See also the private-repo sample README.