Blue/Green Upgrades

Using DataPlane

Blue/Green upgrades can be accomplished when working with the DataPlane resource directly.

To enable blue/green deployments set the spec.deployment.rollout.strategy on your DataPlane resource:

 apiVersion: gateway-operator.konghq.com/v1beta1
 kind: DataPlane
 metadata:
   name: dataplane-example
 spec:
   deployment:
     rollout:
       strategy:
         blueGreen:
           promotion:
             strategy: BreakBeforePromotion
     podTemplateSpec:
       spec:
         containers:
         - name: proxy
           image: kong/kong-gateway:3.8.1.0
           env:
           - name: KONG_LOG_LEVEL
             value: debug
           readinessProbe:
             initialDelaySeconds: 1
             periodSeconds: 1

NOTE: Currently only BreakBeforePromotion is available as promotion strategy.

When applied like this, Kong Gateway Operator will deploy new Services through which you’ll be able to access new Pods once they are available.

By default no Pods will be deployed immediately, instead Kong Gateway Operator will observe the DataPlane resource for spec changes and when any configuration drift is detected it will spawn a new “preview” Deployment which will contain the changes applied to the DataPlane resource.

Wait for DataPlane to be ready to accept changes

 kubectl wait dataplane dataplane-example --for=jsonpath='{.status.rollout.conditions[*].reason}'=AwaitingPromotion

Test it out by patching the DataPlane with a new image:

 kubectl patch dataplane dataplane-example --type='json' -p='[{"op": "replace", "path": "/spec/deployment/podTemplateSpec/spec/containers/0/image", "value":"kong:3.3.1"}]'

The output should look like this:

 dataplane.gateway-operator.konghq.com/dataplane-example patched

After this patch gets applied you’ll be able to access the new Kong Gateway Pods via the “preview” ingress Service.

To find the “preview” Service you can look up DataPlane status, and more specifically its rollout field:

 kubectl get dataplane dataplane-example -o jsonpath-as-json='{.status.rollout}'

The output should look like this:

 [
     {
         "conditions": [
             {
                 "lastTransitionTime": "2023-09-21T11:40:25Z",
                 "message": "",
                 "observedGeneration": 2,
                 "reason": "AwaitingPromotion",
                 "status": "False",
                 "type": "RolledOut"
             }
         ],
         "deployment": {
             "selector": "6cf0d993-2319-43d5-bfdc-e2cadd6bd7e3"
         },
         "services": {
             "adminAPI": {
                 "addresses": [
                     {
                         "sourceType": "PrivateIP",
                         "type": "IPAddress",
                         "value": "None"
                     }
                 ],
                 "name": "dataplane-admin-dataplane-example-cx6nq"
             },
             "ingress": {
                 "addresses": [
                     {
                         "sourceType": "PrivateLoadBalancer",
                         "type": "IPAddress",
                         "value": "172.18.0.101"
                     },
                     {
                         "sourceType": "PrivateIP",
                         "type": "IPAddress",
                         "value": "10.96.28.2"
                     }
                 ],
                 "name": "dataplane-ingress-dataplane-example-2249g"
             }
         }
     }
 ]

Here you can see the ingress Service name that was created for you to validate the new set of Pods.

Its addresses (together with their types and sourceTypes) are stored status.rollout.services[].ingress.

Notice that status.rollout.conditions[] contains a condition with Type RolledOut and Reason set to AwaitingPromotion. This means that everything is ready to promote the DataPlane with staged changes. Before we do so, let’s test it.

You can access the spawned “preview” Service by using its LB address (taken from the status.rollout.services[].ingress.addresses[].value field):
```
 $ curl -s -D - -o /dev/null 172.18.0.101
 HTTP/1.1 404 Not Found
 Date: Thu, 21 Sep 2023 11:40:26 GMT
 Content-Type: application/json; charset=utf-8
 Connection: keep-alive
 Content-Length: 52
 X-Kong-Response-Latency: 0
 Server: kong/3.3.1
```
This way we can see that new Kong Gateway Pods have been deployed and are reachable through the “preview” ingress Service using the updated image.

Verify the old Pods are still available and are still serving the traffic. You can verify that by accessing its “live” ingress Service

We can get its addresses with:

 kubectl get dataplane dataplane-example -o jsonpath-as-json='{.status.addresses}'

The output should look like this:

 [
   [
     {
         "sourceType": "PrivateLoadBalancer",
         "type": "IPAddress",
         "value": "172.18.0.100"
     },
     {
         "sourceType": "PrivateIP",
         "type": "IPAddress",
         "value": "10.96.11.156"
     }
   ],
 ]

Knowing the LoadBalancer IP address is 172.18.0.100, you can issue a request:

 $ curl -s -D - -o /dev/null 172.18.0.100
 HTTP/1.1 404 Not Found
 Date: Thu, 21 Sep 2023 11:40:26 GMT
 Content-Type: application/json; charset=utf-8
 Connection: keep-alive
 Content-Length: 52
 X-Kong-Response-Latency: 0
 Server: kong/3.8.1

As you can see the live Service is still serving traffic using 3.8.1

Now you can perform additional validation steps by inspecting the deployed resources.
Once you’ve validated the newly created resources, run kubectl annotate dataplanes.gateway-operator.konghq.com dataplane-example gateway-operator.konghq.com/promote-when-ready=true to allow Kong Gateway Operator to switch the traffic to the new Pods.

This annotation will automatically be cleared by Kong Gateway Operator once the new Pods are promoted to be live.
Once the promotion concludes, the updated Pods start serving traffic and the old Pods and their Deployment will be deleted to conserve the resources.