multi-cluster-fan-out

Multi-Cluster Fan-out

This tutorial shows how to manage Namespaces, RoleBindings, and ResourceQuotas across multiple clusters using Anthos Config Management and GitOps.

The resources in this tutorial are identical across both clusters. So Config Sync is configured to pull config from the same directory. If you want your config to be different for every cluster, check out the Multi-Cluster Access and Quota tutorial instead.

Clusters

• cluster-east - A multi-zone GKE cluster in the us-east1 region.
• cluster-west - A multi-zone GKE cluster in the us-west1 region.

Filesystem Hierarchy

Platform Repo (repos/platform/):

└── configsync
    └── all-clusters
        ├── namespaces
        │   ├── tenant-a
        │   │   ├── quota.yaml
        │   │   └── rbac.yaml
        │   ├── tenant-b
        │   │   ├── quota.yaml
        │   │   └── rbac.yaml
        │   └── tenant-c
        │       ├── quota.yaml
        │       └── rbac.yaml
        └── namespaces.yaml

Access Control

This tutorial includes RoleBindings in each namespace to grant view permission to namespace users.

The users are configured to be different for each namespace, but the same across clusters.

Progressive rollouts

This tutorial demonstrates the deployment of resources to multiple clusters at the same time. In a production environment, you may want to reduce the risk of rolling out changes by deploying to each cluster individually and/or by deploying to a staging environment first.

One way to do that is to change the field spec.git.revision in the RootSync resource for each cluster to point to a specific commit SHA or tag. That way, both clusters will pull from a specific revision, instead of both pulling from HEAD of the main branch. This method may help protect against complete outage and allow for easy rollbacks, at the cost of a few more commits per rollout.

Another option is to separate the configuration for each cluster into different directories. See Multi-Cluster Resource Management for an example of this pattern.

To read more about progressive delivery patterns, see Safe rollouts with Anthos Config Management.

Config Sync

This tutorial installs Config Sync on two clusters and configures them both to pull config from the same configsync/all-clusters/ directory in the same Git repository.

Before you begin

1. Follow the Multi-Cluster Anthos Config Management Setup tutorial to deploy two GKE clusters and install ACM.

Create a Git repository for platform config

Github: Create a repo

PLATFORM_REPO_HTTPS="https://github.com/USER_NAME/REPO_NAME/"
PLATFORM_REPO_SSH="[email protected]:USER_NAME/REPO_NAME.git"

Select or create a local workspace directory:

Since you will need to clone multiple repos for this tutorial, select a directory to contain them.

• Replace <WORKSPACE> with the name of the desired workspace directory (ex: ~/workspace)

This value will be stored in an environment variable for later use.

WORKSPACE="<WORKSPACE>"

mkdir -p "${WORKSPACE}"

Clone the tutorial repo:

cd "${WORKSPACE}"

git clone https://github.com/GoogleCloudPlatform/anthos-config-management-samples.git

Clone the platform repo:

cd "${WORKSPACE}"

git clone "${PLATFORM_REPO_SSH}" platform

Copy the platform config from the tutorial repo:

cd "${WORKSPACE}"

cp -r anthos-config-management-samples/multi-cluster-fan-out/repos/platform/* platform/

Push the platform config to the platform repo:

cd "${WORKSPACE}/platform/"

git add .

git commit -m "initialize platform config"

git push

Configure Config Sync

Anthos Config Management (ACM) is used to install and configure Config Sync.

Hub is used to install and configure ACM.

Configure Config Sync using Hub:

cat > config-management.yaml << EOF
applySpecVersion: 1
spec:
  configSync:
    enabled: true
    sourceFormat: unstructured
    syncRepo: ${PLATFORM_REPO_HTTPS}
    syncBranch: main
    syncRev: HEAD
    policyDir: "configsync/all-clusters"
    secretType: none
EOF

gcloud beta container fleet config-management apply \
  --membership "cluster-west" \
  --config config-management.yaml

gcloud beta container fleet config-management apply \
  --membership "cluster-east" \
  --config config-management.yaml

rm config-management.yaml

This triggers the following actions:

1. Hub installs the ACM Operator
2. Hub configures the ACM Operator using a ConfigManagement resource
3. ACM Operator installs ConfigSync
4. Hub configures ConfigSync using a RootSync resources

Validating success

Lookup latest commit SHA:

(cd "${WORKSPACE}/platform/" && git rev-parse --short HEAD)

Lookup the Config Sync status:

gcloud beta container fleet config-management status

Should say "SYNCED" for both clusters with the latest commit SHA.

If not yet SYNCED, you may need to wait and retry.

Alternatively, use the nomos tool:

nomos status --contexts ${CLUSTER_WEST_CONTEXT},${CLUSTER_EAST_CONTEXT}

Should say "SYNCED" for both clusters with the latest commit SHA.

If not yet SYNCED, you may need to wait and retry.

Verify expected namespaces exist:

kubectl get ns --context ${CLUSTER_WEST_CONTEXT}
kubectl get ns --context ${CLUSTER_EAST_CONTEXT}

Should include (non-exclusive):

• tenant-a
• tenant-b
• tenant-c

Verify expected resource exist:

kubectl config use-context ${CLUSTER_WEST_CONTEXT}
kubectl get ResourceQuota,RoleBinding -n tenant-a
kubectl get ResourceQuota,RoleBinding -n tenant-b
kubectl get ResourceQuota,RoleBinding -n tenant-c


kubectl config use-context ${CLUSTER_EAST_CONTEXT}
kubectl get ResourceQuota,RoleBinding -n tenant-a
kubectl get ResourceQuota,RoleBinding -n tenant-b
kubectl get ResourceQuota,RoleBinding -n tenant-c

Should include (non-exclusive):

• resourcequota/hard-limit
• rolebinding.rbac.authorization.k8s.io/namespace-viewer

Cleaning up

If you plan to follow more multi-cluster tutorials, you can clean up these clusters with the following steps. Otherwise, follow the Clean up instructions on the Setup tutorial to delete the clusters, network, and project.

Delete the platform config in the platform repo:

Config Sync prevents you from deleting all resources at once (to prevent accidental deletions). Instead, you will need to teardown your resources in a few steps.

First, delete everything and add a new unused namespace as a tombstone:

cd "${WORKSPACE}/platform/"
rm -rf ./configsync/all-clusters/*

cat > ./configsync/all-clusters/tombstone.yaml <<EOF
apiVersion: v1
kind: Namespace
metadata:
  name: tombstone
EOF

git add .
git commit -m "delete platform config & replace with tombstone"
git push

Then, delete the tombstone and leave an empty invisible file to keep the directory in git:

cd "${WORKSPACE}/platform/"
rm -rf ./configsync/all-clusters/*

touch ./configsync/all-clusters/.gitignore

git add .
git commit -m "delete tombstone namespace"
git push

Lookup latest commit SHA:

(cd "${WORKSPACE}/platform/" && git rev-parse --short HEAD)

Wait for config to be synchronized:

gcloud beta container fleet config-management status

Should say "SYNCED" for both clusters with the latest commit SHA.

If not yet SYNCED, you may need to wait and retry.

Disable Config Sync with Hub:

cat > config-management.yaml << EOF
applySpecVersion: 1
spec:
  configSync:
    enabled: false
EOF

gcloud beta container fleet config-management apply \
  --membership "cluster-west" \
  --config config-management.yaml

gcloud beta container fleet config-management apply \
  --membership "cluster-east" \
  --config config-management.yaml

rm config-management.yaml

Delete the platform repo:

Github: Deleting a repository

cd "${WORKSPACE}/"
rm -rf "${WORKSPACE}/platform/"

Next steps

• To learn how to manage resources separately for each cluster, follow the Multi-Cluster Access and Quota tutorial.