Quick Start

This guide provides copy-and-paste instructions to try out Admiralty Open Source. We use kind (Kubernetes in Docker) to create Kubernetes clusters, but feel free to use something else—though don't just copy and paste instructions then.

Example Use Case

We're going to model a centralized cluster topology made of a management cluster (named cd) where applications are deployed, and two workload clusters (named us and eu) where containers actually run. We'll deploy a batch job utilizing both workload clusters, and another targeting a specific region. If you're interested in other topologies or other kinds of applications (e.g., micro-services), this guide is still helpful to get familiar with Admiralty in general.

Prerequisites

  1. Install Helm v3 and kind if not already installed.

  2. We recommend you to use a separate kubeconfig for this exercise, so you can simply delete it when you're done:

    export KUBECONFIG=kubeconfig-admiralty-getting-started

  3. Create three clusters (a management cluster named cd and two workload clusters named us and eu):

    for CLUSTER_NAME in cd us eu
    do
    kind create cluster --name $CLUSTER_NAME
    done

  4. Label the workload cluster nodes as if they were in different regions (we'll use these labels as node selectors):

    for CLUSTER_NAME in us eu
    do
    kubectl --context kind-$CLUSTER_NAME label nodes --all topology.kubernetes.io/region=$CLUSTER_NAME
    done
    tip

    Most cloud distributions of Kubernetes pre-label nodes with the names of their cloud regions.

  5. (optional speed-up) Pull images on your machine and load them into the kind clusters. Otherwise, each kind cluster would pull images, which could take three times as long.

    images=(
    # cert-manager dependency
    quay.io/jetstack/cert-manager-controller:v1.7.1
    quay.io/jetstack/cert-manager-webhook:v1.7.1
    quay.io/jetstack/cert-manager-cainjector:v1.7.1
    quay.io/jetstack/cert-manager-ctl:v1.7.1
    # admiralty open source
    quay.io/admiralty/multicluster-scheduler-agent:0.15.1
    quay.io/admiralty/multicluster-scheduler-scheduler:0.15.1
    quay.io/admiralty/multicluster-scheduler-remove-finalizers:0.15.1
    quay.io/admiralty/multicluster-scheduler-restarter:0.15.1
    )
    for image in "${images[@]}"
    do
    docker pull $image
    for CLUSTER_NAME in cd us eu
    do
    kind load docker-image $image --name $CLUSTER_NAME
    done
    done

  6. Install cert-manager in each cluster:

    helm repo add jetstack https://charts.jetstack.io
    helm repo update
    for CLUSTER_NAME in cd us eu
    do
    helm install cert-manager jetstack/cert-manager \
    --kube-context kind-$CLUSTER_NAME \
    --namespace cert-manager --create-namespace \
    --version v1.7.1 --set installCRDs=true \
    --wait --debug
    # --wait to ensure release is ready before next steps
    # --debug to show progress, for lack of a better way,
    # as this may take a few minutes
    done
    note

    Admiralty Open Source uses cert-manager to generate a server certificate for its mutating pod admission webhook.

Installation

Install Admiralty in each cluster:

helm repo add admiralty https://charts.admiralty.io
helm repo update
for CLUSTER_NAME in cd us eu
do
helm install admiralty admiralty/multicluster-scheduler \
--kube-context kind-$CLUSTER_NAME \
--namespace admiralty --create-namespace \
--version 0.15.1 \
--wait --debug
# --wait to ensure release is ready before next steps
# --debug to show progress, for lack of a better way,
# as this may take a few minutes
done

Configuration

Cross-Cluster Authentication

  1. Install jq, the command-line JSON processor, if not already installed.

  2. For each workload cluster,

    1. create a Kubernetes service account in the workload cluster for the management cluster,
    2. extract its default token,
    3. get a Kubernetes API address that is routable from the management cluster—here, the IP address of the kind workload cluster's only (master) node container in your machine's shared Docker network,
    4. prepare a kubeconfig using the token and address found above, and the server certificate from your kubeconfig (luckily also valid for this address, not just the address in your kubeconfig),
    5. save the prepared kubeconfig in a secret in the management cluster:
    for CLUSTER_NAME in us eu
    do
    # i.
    kubectl --context kind-$CLUSTER_NAME create serviceaccount cd
    # ii.
    SECRET_NAME=$(kubectl --context kind-$CLUSTER_NAME get serviceaccount cd \
    --output json | \
    jq -r '.secrets[0].name')
    TOKEN=$(kubectl --context kind-$CLUSTER_NAME get secret $SECRET_NAME \
    --output json | \
    jq -r '.data.token' | \
    base64 --decode)
    # iii.
    IP=$(docker inspect $CLUSTER_NAME-control-plane \
    --format "{{ .NetworkSettings.Networks.kind.IPAddress }}")
    # iv.
    CONFIG=$(kubectl --context kind-$CLUSTER_NAME config view \
    --minify --raw --output json | \
    jq '.users[0].user={token:"'$TOKEN'"} | .clusters[0].cluster.server="https://'$IP':6443"')
    # v.
    kubectl --context kind-cd create secret generic $CLUSTER_NAME \
    --from-literal=config="$CONFIG"
    done
    Security Notice

    Kubernetes service account tokens exposed as secrets are valid forever, or until those secrets are deleted. A leak may go undetected indefinitely. If you use Kubernetes service account tokens as a cross-cluster authentication method in production, we recommend rotating the tokens as often as practical. However, there are other methods, including using Admiralty Cloud.

    Other Platforms

    If you're not using kind, your mileage may vary. The Kubernetes API address in your kubeconfig may or may not be routable from other clusters. If not, the server certificate in your kubeconfig may or may not be valid for the routable address that you'll find instead.

Multi-Cluster Scheduling

  1. In the management cluster, create a Target for each workload cluster:

    for CLUSTER_NAME in us eu
    do
    cat <<EOF | kubectl --context kind-cd apply -f -
    apiVersion: multicluster.admiralty.io/v1alpha1
    kind: Target
    metadata:
    name: $CLUSTER_NAME
    spec:
    kubeconfigSecret:
    name: $CLUSTER_NAME
    EOF
    done

  2. In the workload clusters, create a Source for the management cluster:

    for CLUSTER_NAME in us eu
    do
    cat <<EOF | kubectl --context kind-$CLUSTER_NAME apply -f -
    apiVersion: multicluster.admiralty.io/v1alpha1
    kind: Source
    metadata:
    name: cd
    spec:
    serviceAccountName: cd
    EOF
    done

Demo

  1. Check that virtual nodes have been created in the management cluster to represent workload clusters:

    kubectl --context kind-cd get nodes --watch
    # --watch until virtual nodes are created,
    # this may take a few minutes, then control-C

  2. Label the default namespace in the management cluster to enable multi-cluster scheduling at the namespace level:

    kubectl --context kind-cd label ns default multicluster-scheduler=enabled

  3. Create Kubernetes Jobs in the management cluster, utilizing all workload clusters (multi-cluster scheduling is enabled at the pod level with the multicluster.admiralty.io/elect annotation):

    for i in $(seq 1 10)
    do
    cat <<EOF | kubectl --context kind-cd apply -f -
    apiVersion: batch/v1
    kind: Job
    metadata:
    name: global-$i
    spec:
    template:
    metadata:
    annotations:
    multicluster.admiralty.io/elect: ""
    spec:
    containers:
    - name: c
    image: busybox
    command: ["sh", "-c", "echo Processing item $i && sleep 5"]
    resources:
    requests:
    cpu: 100m
    restartPolicy: Never
    EOF
    done

  4. Check that proxy pods for this job have been created in the management cluster, "running" on virtual nodes, and delegate pods have been created in the workload clusters, actually running their containers on real nodes:

    while true
    do
    clear
    for CLUSTER_NAME in cd us eu
    do
    kubectl --context kind-$CLUSTER_NAME get pods -o wide
    done
    sleep 2
    done
    # control-C when all pods have Completed

  5. Create Kubernetes Jobs in the management cluster, targeting a specific region with a node selector:

    for i in $(seq 1 10)
    do
    cat <<EOF | kubectl --context kind-cd apply -f -
    apiVersion: batch/v1
    kind: Job
    metadata:
    name: eu-$i
    spec:
    template:
    metadata:
    annotations:
    multicluster.admiralty.io/elect: ""
    spec:
    nodeSelector:
    topology.kubernetes.io/region: eu
    containers:
    - name: c
    image: busybox
    command: ["sh", "-c", "echo Processing item $i && sleep 5"]
    resources:
    requests:
    cpu: 100m
    restartPolicy: Never
    EOF
    done

  6. Check that proxy pods for this job have been created in the management cluster, and delegate pods have been created in the eu cluster only:

    while true
    do
    clear
    for CLUSTER_NAME in cd us eu
    do
    kubectl --context kind-$CLUSTER_NAME get pods -o wide
    done
    sleep 2
    done
    # control-C when all pods have Completed

    You may observe transient pending candidate pods in the us cluster.

Cleanup

for CLUSTER_NAME in cd us eu
do
kind delete cluster --name $CLUSTER_NAME
done
rm kubeconfig-admiralty-getting-started
Edit this page