Skip to content

Progressive Delivery in a Hosted Mesh Worker Cluster Based on GitOps

Scenario: How can customers perform a progressive delivery in a hosted mesh worker cluster using GitOps?

Prerequisites

Refer to the documentation Deploy Istio Resources in a Hosted Mesh Worker Cluster Based on GitOps and complete all steps before Create a Hosted Mesh and Manage the Worker Cluster.

Implementing Weight-Based Progressive Delivery with GitOps

  1. Navigate to Workbench, and Create Application in GitOps, with the repository address: https://github.com/amamba-io/rollout-examples. Path should be set to rollouts/v1, and select the namespace in the worker cluster.

  2. After the application is successfully created, synchronize the application resources.

  3. Go to the service mesh and check the vs, dr, and gateway resources synchronized in the worker cluster under the hosted mesh. If not found, check if the notes in the Create a Hosted Mesh and Manage the Worker Cluster section are configured correctly.

  4. By default, the load balancer of the worker cluster exposes port 80, so you need to modify the gateway port of the service to 80 in the worker cluster to access the service.

  5. Access http://<worker cluster node IP>/hello?name=test-v1 in the browser.

  6. Navigate to Workbench, in the Progressive Delivery list, select dubbo3-provider to edit the YAML, and change version: v1 to version: v2.

  7. After the modification, dubbo3-provider begins the progressive process.

    As shown above: You can distinguish based on the workload labels rollouts-pod-template-hash and version.

    Note

    • v1 Pod: dubbo3-provider-5d96bf8bbd-v9c28
    • v2 Pod: dubbo3-provider-f44c7c68f-95pfb

  8. Explanation of the canary rules can be found in the git repository under the directory rollouts-legacy/v1/rollout-provider.yaml.

    ...
strategy:
  canary:
    trafficRouting:
      istio:
        virtualService:
          name: provider
          routes:
            - test-v1
        destinationRule:
          name: dubbo-provider
          stableSubsetName: stable
          canarySubsetName: canary
    # When entering the progressive deivery, the traffic weight of the new version is set to 5%. After 10 minutes, the weight is set to 50%.
    steps:
      - setWeight: 5
      - pause:
          duration: 10m
      - setWeight: 50
      - pause:
          duration: 10m

  9. The terminal simulation results are as follows:

    • When the new version's weight is 5%, the traffic distribution between v1 Pod and v2 Pod is as follows:

      weight 5%

    • When the new version's weight is 50%, the traffic distribution between v1 Pod and v2 Pod is as follows:

      weight 50%

    • After a successful release, all the traffic is directed to the v2 Pod.

      successful release

Implementing Header-Based Progressive Delivery with GitOps

Introduction

Demonstration example repository: https://github.com/amamba-io/rollout-examples

The dubbo3-consumer and dubbo3-provider services are related. When accessing the /hello interface of dubbo3-consumer, dubbo3-consumer will forward the request to dubbo3-provider. The release process will update the dubbo3-provider service and route according to the preset header information.

Steps

  1. Navigate to Workbench, Create Application in GitOps, with the repository address: https://github.com/amamba-io/rollout-examples. Path should be set to rollouts-legacy/v1, and select the namespace in the worker cluster.

  2. After an applcation is successfully created, synchronize the application resources.

  3. Go to the service mesh and check the vs, dr, and gateway resources synchronized in the worker cluster under the hosted mesh. If not found, check if the notes in the Create a Hosted Mesh and Manage the Worker Cluster section are configured correctly.

  4. By default, the load balancer of the worker cluster exposes port 80, so you need to modify the gateway port of the service to 80 in the worker cluster to access the service.

  5. Access http://<worker cluster node IP>/hello?name=test-v1 in the browser.

  6. Navigate to Workbench, in the Progressive Delivery list, select dubbo3-provider to edit the YAML, and change version: v1 to version: v2.

    edit yaml

  7. After the modification, dubbo3-provider begins the progressive process.

    As shown above: You can distinguish based on the workload labels rollouts-pod-template-hash and version.

    Note

    • v1 Pod: dubbo3-provider-5d96bf8bbd-v9c28
    • v2 Pod: dubbo3-provider-f44c7c68f-95pfb

  8. Explanation of the canary rules can be found in the git repository under the directory rollouts-legacy/v1/rollout-provider.yaml.

    ...
strategy:
  canary:
    canaryService: dubbo3-provider-canary
    stableService: dubbo3-provider
    trafficRouting:
      managedRoutes:
        - name: test-v2
      istio:
        virtualService:
          name: provider
          routes:
            - test-v1
    # During the delivery, the vs header is set to test-v2. If the request header contains `lane: test-v2`, it will be routed to the new version.
    steps:
      - setCanaryScale:
          weight: 100
      - setHeaderRoute:
          name: test-v2
          match:
          - headerName: lane
            headerValue:
              exact: test-v2
      - pause: {}

  9. The terminal simulation results are as follows:

    ultimate results

FAQ

Does the above process support multi-cluster progressive delivery?

Yes, it does. The following operations are required:

  1. First, you need to place multiple clusters under the same hosted mesh.

  2. You need to create namespaces with the same name, and deploy the same GitOps application in each cluster's namespace.

  3. During the delivery, you need to operate the corresponding services in both clusters separately.

Comments