☰
Navigation :

Canary Upgrade

You can upgrade your Gloo Edge deployments by following a canary model. In the canary model, you make two different gloo deployments in your data plane, one that runs your current version and one for the target version to upgrade to. Then, you check that the deployment at the target version handles traffic as you expect before upgrading to run the target version. This approach helps you reduce potential downtime for production upgrades.

Before you begin

  1. Install Gloo Edge Open Source or Enterprise version 1.9.0 or later, or Federation version 1.13.0 or later.
  2. Upgrade your installation to the latest patch version for your current minor version. For example, you might upgrade your 1.14 installation to the latest 1.14.6 patch.
  3. If you have Gloo Edge Enterprise or Federation, set your license key as an environment variable. To request a license, contact Sales. 
    export GLOO_LICENSE=<license>
  4. Set the target version that you want to upgrade to as an environment variable. To find available versions, check the changelog. The following commands include the latest versions for each of the following Gloo Edge installation options. 
    
export TARGET_VERSION=1.14.6

    
export TARGET_VERSION=1.14.3

    
export TARGET_VERSION=1.14.3
  5. Upgrade your glooctl CLI to the version that you want to upgrade to.
  6. Check the upgrade notice for the minor version and the changelogs for the patch version that you want to upgrade to. In particular, review the following changes:
    • CRD changes: Each patch version might add custom resource definitions (CRDs), update existing CRDs, or remove outdated CRDs. When you perform a canary upgrade, the existing Gloo Edge CRDs are not updated to the newer version automatically. You must manually apply the new CRDs first. The Gloo Edge CRDs are designed to be backward compatible, so the new CRDs should not impact the performance of your older installation. However, if after evaluating the newer installation you decide to continue to use the older installation, you can easily remove any added CRDs by referring to the upgrade notices for the CRD names and running kubectl delete crd <CRD>. Then, to re-apply older versions of CRDs, you can run helm pull gloo/gloo --version <older_version> --untar and kubectl apply -f gloo/crds.
    • Breaking changes: Occasionally, breaking changes are introduced between patch versions. For example, Gloo custom resources might get renamed or have conflicting fields. Modify your resources to use the new version that you upgrade to.

Now you’re ready to upgrade. The steps vary depending on your Gloo Edge installation.

Upgrade Gloo Edge Open Source or Enterprise in a canary model

  1. Update and pull the Gloo Edge Helm chart for the target version. 
    
helm repo add gloo https://storage.googleapis.com/solo-public-helm
helm repo update
helm pull gloo/gloo --version $TARGET_VERSION --untar

    
helm repo add glooe https://storage.googleapis.com/gloo-ee-helm
helm repo update
helm pull glooe/gloo-ee --version $TARGET_VERSION --untar
  2. Apply the new Gloo CRDs from the target version Helm chart to your cluster. 
    
kubectl apply -f gloo/crds

    
kubectl apply -f gloo-ee/charts/gloo/crds
  3. Install the target version of Gloo Edge in the new namespace in your cluster. 
    
glooctl install gateway --version $TARGET_VERSION -n gloo-system-$TARGET_VERSION

    
glooctl install gateway enterprise --version $TARGET_VERSION -n gloo-system-$TARGET_VERSION --license-key $GLOO_LICENSE
  4. Verify that your current and target versions of Gloo Edge are running. 
    kubectl get all -n gloo-system
kubectl get all -n gloo-system-$TARGET_VERSION
  5. Modify any custom resources for any changes or new capabilities that you noticed in the upgrade notice and the changelogs for the target version.
  6. Test your routes and monitor the metrics of the newer version. 
    glooctl check
  7. Uninstall the older version of Gloo Edge so that your cluster uses the newer version going forward. 
    gloooctl uninstall -n gloo-system

Upgrade Gloo Edge Federation in a canary model

You can upgrade Gloo Edge Federation in a canary model in version 1.13 or later. In Gloo Edge Federation, you have a management cluster that runs Gloo Edge Federation (and optionally, other Enterprise components). Then you register remote clusters with Gloo Edge Enterprise.

In the canary upgrade model, you start with an older version of Gloo Edge Federation in your management cluster that manages one or more remote Gloo Edge instances. To test a newer target version, you install the target version of Gloo Edge Federation in a new namespace on the management cluster. Then, you install one or more Gloo Edge instances at the target version in remote test clusters, register the new clusters, and check your setup. Finally, you deregister the remote clusters from the older version of Gloo Edge Federation and uninstall the older Gloo Edge Federation version from the management cluster.

  1. Update and pull the Gloo Fed Helm chart for the target version.

    helm repo add gloo-fed https://storage.googleapis.com/gloo-fed-helm
helm repo update
helm pull gloo-fed/gloo-fed --version $TARGET_VERSION --untar

  2. Apply the new Gloo Fed CRDs from the target version Helm chart to your management cluster.

    kubectl apply -f gloo-fed/crds

  3. Install the target version of Gloo Edge Federation in the new namespace in your management cluster.

    helm install -n gloo-system-$TARGET_VERSION gloo-fed gloo-fed/gloo-fed --create-namespace --set-string license_key=$GLOO_LICENSE --version $TARGET_VERSION

  4. Verify that your current and target versions of Gloo Edge Federation are running.

    kubectl get all -n gloo-system-$OLD_VERSION
kubectl get all -n gloo-system-$TARGET_VERSION

  5. Install and register Gloo Edge Enterprise on one of your remote clusters. Make sure that the version of Gloo Edge Enterprise matches the version of Gloo Edge Federation that you installed earlier.

  6. In your management cluster, create or modify any federated custom resources needed to test your target version of Gloo Federation/Edge. Because Gloo Edge Federation scans all federated resources on the management cluster, you might want to reuse the existing federated resources in the gloo-system-$OLD_VERSION namespace. Include any changes or new capabilities that you noticed in the upgrade notice and the changelogs for the target version.

    Example: The following configuration creates a federated virtual service with different features across versions 1.12 and 1.13.
    • Creates a shared gloo-fed namespace to show that both the old and new Gloo Edge Federation instances can watch for changes to a resource in the same namespace.
    • Places the federated virtual service resource in both the remote1 old version cluster (1.12) and the remote2 new target version cluster (1.13).
    • Configures settings such as numRetries: 10 that are shared across versions.
    • Adds the retryBackOff that is available in version 1.13 but not in 1.12. As such, the 1.12 federated resource ignores this new configuration.
    kubectl create ns gloo-fed
kubectl apply -f - <<EOF
apiVersion: fed.gateway.solo.io/v1
kind: FederatedVirtualService
metadata:
  name: my-fed-vs
  namespace: gloo-fed
spec:
  placement:
    clusters:
      - remote1
      - remote2
    namespaces:
      - gloo-system
  template:
    metadata:
      name: my-vs
    spec:
      virtualHost:
        domains:
        - '*'
        routes:
        - matchers:
          - prefix: /
          routeAction:
            single:
              upstream:
                name: default-petstore-8080
                namespace: gloo-system
          options:
            retries:
              retryOn: 'connect-failure'
              numRetries: 10
              perTryTimeout: '5s'
              retryBackOff:
                baseInterval: 1s
                maxInterval: 3s
EOF

  7. Check that the updated, federated custom resources from the management cluster are propagated to the remote test cluster.

    1. To find the clusters and names of the propagated resources, review the federated resource configuration. The following example is based off the federated virtual service that you created in the previous step. The PLACED state in the status section confirms that the resource is federated. 
      kubectl get federatedvirtualservice -n gloo-fed -o yaml

      Example truncated output: 

       
...
status:
  namespacedPlacementStatuses:
    gloo-system-$TARGET_VERSION:
      clusters:
        remote2:
          namespaces:
            gloo-system:
              state: PLACED
    gloo-system-$OLD_VERSION:
      clusters:
        remote1:
          namespaces:
            gloo-system:
              state: PLACED

    2. In the remote clusters, confirm that the federated resources are created. The following commands are based on the previous step’s example.
      • In the old version’s remote cluster, verify that the resource is federated with only the old version’s configuration. In the example, the old 1.12 version has an updated numRetries: 10, but no retryBackOff section, which is only available in version 1.13 or later. 
        kubectl get virtualservices -n gloo-system --context remote1 -o yaml
...
options:
  retries:
    numRetries: 10
    perTryTimeout: 5s
    retryOn: connect-failure
      • In the target version’s remote cluster, verify that the resource is federated with only the new version’s configuration. In the example, the new 1.13 version has an updated numRetries: 10, as well as a retryBackOff section. 
         
kubectl get virtualservices -n gloo-system --context remote2 -o yaml
...
options:
  retries:
    numRetries: 10
    perTryTimeout: 5s
    retryOn: connect-failure
    retryBackOff:
      baseInterval: 1s
      maxInterval: 3s
    3. Repeat the previous steps for each federated resource that you want to check.

  8. Test your routes and monitor the metrics of the target version in your remote test cluster.

    glooctl check

  9. Shift traffic to the target version of Gloo Edge Federation by uninstalling the older version.

    1. Deregister your other remote clusters that still use the old version of Gloo Edge Federation.
    2. Uninstall the older version of Gloo Edge Federation so that your management cluster uses the newer version going forward. 
      gloooctl uninstall -n gloo-system-$OLD_VERSION
    3. Optionally delete the old version namespace from each remote cluster. The deregister command does not clean up the namespace and custom resources in the old version. 
      kubectl delete ns gloo-system-$OLD_VERSION

Roll back a canary upgrade

As you test your environment with the new version of Gloo Edge, you might need to roll back to the previous version. You can follow a canary model for the rollback.

  1. Set the previous version that you want to roll back to as an environment variable. 
    export ROLLBACK_VERSION=<version>
  2. If you already removed the previous version of Gloo from your cluster, re-install Gloo Edge.
    1. Update and pull the Gloo Edge Helm chart for the target version. The Helm charts vary depending on the Gloo Edge installation option. 
      
helm repo update
helm pull gloo/gloo --version $ROLLBACK_VERSION --untar

      
helm repo update
helm pull glooe/gloo-ee --version $ROLLBACK_VERSION --untar

      
helm repo update
helm pull gloo-fed/gloo-fed --version $ROLLBACK_VERSION --untar
    2. Install the rollback version of Gloo Edge in the new namespace in your cluster. 
      
glooctl install gateway --version $ROLLBACK_VERSION-n gloo-system-$ROLLBACK_VERSION

      
glooctl install gateway enterprise --version $ROLLBACK_VERSION -n gloo-system-$ROLLBACK_VERSION --license-key $GLOO_LICENSE

      
helm install -n gloo-system-$ROLLBACK_VERSION gloo-fed gloo-fed/gloo-fed --create-namespace --set-string license_key=$GLOO_LICENSE --version $ROLLBACK_VERSION
  3. Revert any changes to custom resources that you previously modified during the upgrade to the newer target version. For differences between versions, check the upgrade notice and the changelogs.
  4. Gloo Edge Federation: Shift traffic from the newer target version to the rollback version of Gloo Edge Federation.
    1. Deregister your remote clusters that still use the newer target version of Gloo Edge Federation.
    2. Install and register Gloo Edge Enterprise for the rollback version’s Gloo Edge Federation on each remote cluster.
    3. Optionally delete the newer target version namespace. The deregister command does not clean up the namespace and custom resources in the old version. 
      kubectl delete ns gloo-system-$TARGET_VERSION
  5. Uninstall the newer target version of Gloo Edge from your cluster. 
    gloooctl uninstall -n gloo-system-$TARGET_VERSION
  6. Apply the Gloo CRDs from the rollback version Helm chart to your cluster. In Gloo Edge Federation, this cluster is the local management cluster. 
    
kubectl apply -f gloo/crds

    
kubectl apply -f gloo-ee/charts/gloo/crds

    
kubectl apply -f gloo-fed/crds
  7. Test your routes and monitor the metrics of the rollback version. 
    glooctl check