4bd69f56a9
* Bug fix for dockerfile path link |
||
---|---|---|
.github | ||
__tests__ | ||
lib | ||
src | ||
.gitignore | ||
CODE_OF_CONDUCT.md | ||
LICENSE | ||
README.md | ||
SECURITY.md | ||
action.yml | ||
jest.config.js | ||
package-lock.json | ||
package.json | ||
tsconfig.json |
README.md
Deploy manifests action for Kubernetes
This action can be used to deploy manifests to Kubernetes clusters.
This action requires that the cluster context be set earlier in the workflow by using either the Azure/aks-set-context action or the Azure/k8s-set-context action.
If you are looking to automate your workflows to deploy to Azure Web Apps and Azure Web App for Containers, consider using Azure/webapps-deploy
action.
Action capabilities
Following are the key capabilities of this action:
-
Artifact substitution: The deploy action takes as input a list of container images which can be specified along with their tags or digests. The same is substituted into the non-templatized version of manifest files before applying to the cluster to ensure that the right version of the image is pulled by the cluster nodes.
-
Object stability checks: Rollout status is checked for the Kubernetes objects deployed. This is done to incorporate stability checks while computing the action status as success/failure.
-
Secret handling: The secret names specified as inputs in the action are used to augment the input manifest files with imagePullSecrets values before deploying to the cluster. Also, checkout the Azure/k8s-create-secret action for creation of generic or docker-registry secrets in the cluster.
-
Deployment strategy The action supports canary and blue-green deployment strategies:
-
Canary strategy: Choosing canary strategy with this action leads to creation of workloads suffixed with '-baseline' and '-canary'. There are two methods of traffic splitting supported in the action:
- Service Mesh Interface: Service Mesh Interface abstraction allows for plug-and-play configuration with service mesh providers such as Linkerd and Istio. Meanwhile, this action takes away the hard work of mapping SMI's TrafficSplit objects to the stable, baseline and canary services during the lifecycle of the deployment strategy. Service mesh based canary deployments using this action are more accurate as service mesh providers enable granular percentage traffic split (via service registry and sidecar containers injected into pods alongside application containers).
- Only Kubernetes (no service mesh): In the absence of service mesh, while it may not be possible to achieve exact percentage split at the request level, it is still possible to perform canary deployments by deploying -baseline and -canary workload variants next to the stable variant. The service routes requests to pods of all three workload variants as the selector-label constraints are met (KubernetesManifest will honor these when creating -baseline and -canary variants). This achieves the intended effect of routing only a portion of total requests to the canary.
-
Blue-Green strategy: Choosing blue-green strategy with this action leads to creation of workloads suffixed with '-green'. There are three route-methods supported in the action:
*Terminolgy: An identified service is one that is supplied as part of the input manifest(s) and targets a workload in the supplied manifest(s).
- Service route-method: Identified services are configured to target the green deployments.
- Ingress route-method: Along with deployments, new services are created with '-green' suffix (for identified services), and the ingresses are in turn updated to target the new services.
- SMI route-method: A new TrafficSplit object is created for each identified service. The TrafficSplit object is updated to target the new deployments. Note that this works only if SMI is set up in the cluster.
Traffic is routed to the new workloads only after the time provided as
version-switch-buffer
input has passed.promote
action creates workloads and services with new configurations but without any suffix.reject
action routes traffic back to the old workloads and deletes the '-green' workloads.
-
Action inputs
Action inputs | Description |
---|---|
namespace Namespace |
(Optional) Namespace within the cluster to deploy to. |
manifests Manifests |
(Required) Path to the manifest files to be used for deployment |
images Images |
(Optional) Fully qualified resource URL of the image(s) to be used for substitutions on the manifest files. This multiline input accepts specifying multiple artifact substitutions in newline separated form. For example - images: | contosodemo.azurecr.io/foo:test1 contosodemo.azurecr.io/bar:test2 In this example, all references to contosodemo.azurecr.io/foo and contosodemo.azurecr.io/bar are searched for in the image field of the input manifest files. For the matches found, the tags test1 and test2 are substituted. |
imagepullsecrets Image pull secrets |
(Optional) Multiline input where each line contains the name of a docker-registry secret that has already been setup within the cluster. Each of these secret names are added under imagePullSecrets field for the workloads found in the input manifest files |
strategy Strategy |
(Optional) Deployment strategy to be used while applying manifest files on the cluster. Acceptable values: none/canary/blue-green. none - No deployment strategy is used when deploying. canary - Canary deployment strategy is used when deploying to the cluster. blue-green - Blue-Green deployment strategy is used when deploying to cluster. |
traffic-split-method Traffic split method |
(Optional) Acceptable values: pod/smi; Default value: pod SMI: Percentage traffic split is done at request level using service mesh. Service mesh has to be setup by cluster admin. Orchestration of TrafficSplit objects of SMI is handled by this action. Pod: Percentage split not possible at request level in the absence of service mesh. So the percentage input is used to calculate the replicas for baseline and canary as a percentage of replicas specified in the input manifests for the stable variant. |
percentage Percentage |
(Required if strategy == canary) Percentage used to compute the number of replicas of '-baseline' and '-canary' varaints of the workloads found in manifest files. For the specified percentage input, if (percentage * numberOfDesirerdReplicas)/100 is not a round number, the floor of this number is used while creating '-baseline' and '-canary' Example: If Deployment hello-world was found in the input manifest file with 'replicas: 4' and if 'strategy: canary' and 'percentage: 25' are given as inputs to the action, then the Deployments hello-world-baseline and hello-world-canary are created with 1 replica each. The '-baseline' variant is created with the same image and tag as the stable version (4 replica variant prior to deployment) while the '-canary' variant is created with the image and tag corresponding to the new changes being deployed |
baseline-and-canary-replicas Baseline and canary replicas |
(Optional; Relevant only if trafficSplitMethod == smi) When trafficSplitMethod == smi, as percentage traffic split is controlled in the service mesh plane, the actual number of replicas for canary and baseline variants could be controlled independently of the traffic split. For example, assume that the input Deployment manifest desired 30 replicas to be used for stable and that the following inputs were specified for the action - strategy: canary trafficSplitMethod: smi percentage: 20 baselineAndCanaryReplicas: 1 In this case, stable variant will receive 80% traffic while baseline and canary variants will receive 10% each (20% split equally between baseline and canary). However, instead of creating baseline and canary with 3 replicas, the explicit count of baseline and canary replicas is honored. That is, only 1 replica each is created for baseline and canary variants. |
route-method Route Method |
(Optional; Relevant only if strategy==blue-green) Default value: service. Acceptable values: service/ingress/smi. Traffic is routed based on this input.
Service: Service selector labels are updated to target '-green' workloads. Ingress: Ingress backends are updated to target the new '-green' services which in turn target '-green' deployments. SMI: A TrafficSplit object is created for each required service to route traffic to new workloads. |
version-switch-buffer Version Switch Buffer |
(Optional; Relevant only if strategy==blue-green and action == deploy) Default value: 0. Acceptable values: 1-300. Waits for the given input in minutes before routing traffic to '-green' workloads. |
action Action |
(Required) Default value: deploy. Acceptable values: deploy/promote/reject. Promote or reject actions are used to promote or reject canary/blue-green deployments. Sample YAML snippets are provided below for guidance on how to use the same. |
kubectl-version Kubectl version |
(Optional) Version of kubectl client to be used for deploying the manifest to the cluster. If this input is left unspecified, latest version is used. |
force Force |
(Optional) Deploy when a previous deployment already exists. If true then '--force' argument is added to the apply command. Using '--force' argument is not recommended in production. |
Examples YAML snippets
Basic deployment (without any deployment strategy)
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
manifests: |
deployment.yaml
service.yaml
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
kubectl-version: 'latest'
Deployment Strategies - Canary deployment without service mesh
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
strategy: canary
percentage: 20
To promote/reject the canary created by the above snippet, the following YAML snippet could be used:
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
strategy: canary
action: promote # substitute reject if you want to reject
Deployment Strategies - Canary deployment based on Service Mesh Interface
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
strategy: canary
traffic-split-method: smi
percentage: 20
baseline-and-canary-replicas: 1
To promote/reject the canary created by the above snippet, the following YAML snippet could be used:
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }} '
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
strategy: canary
traffic-split-method: smi
action: reject # substitute reject if you want to reject
Deployment Strategies - Blue-Green deployment with different route methods
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
ingress.yml
strategy: blue-green
route-method: ingress # substitute with service/smi as per need
version-switch-buffer: 15
To promote/reject the green workload created by the above snippet, the following YAML snippet could be used:
- uses: Azure/k8s-deploy@v1.3
with:
namespace: 'myapp'
images: 'contoso.azurecr.io/myapp:${{ event.run_id }}'
imagepullsecrets: |
image-pull-secret1
image-pull-secret2
manifests: |
deployment.yaml
service.yaml
ingress-yml
strategy: blue-green
route-method: ingress # should be the same as the value when action was deploy
action: promote # substitute reject if you want to reject
End to end workflows
Following are a few examples of not just this action, but how this action could be used along with other container and k8s related actions for building images and deploying objects onto k8s clusters:
Build container image and deploy to Azure Kubernetes Service cluster
on: [push]
jobs:
build:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- uses: Azure/docker-login@v1
with:
login-server: contoso.azurecr.io
username: ${{ secrets.REGISTRY_USERNAME }}
password: ${{ secrets.REGISTRY_PASSWORD }}
- run: |
docker build . -t contoso.azurecr.io/k8sdemo:${{ github.sha }}
docker push contoso.azurecr.io/k8sdemo:${{ github.sha }}
# Set the target AKS cluster.
- uses: Azure/aks-set-context@v1
with:
creds: '${{ secrets.AZURE_CREDENTIALS }}'
cluster-name: contoso
resource-group: contoso-rg
- uses: Azure/k8s-create-secret@v1
with:
container-registry-url: contoso.azurecr.io
container-registry-username: ${{ secrets.REGISTRY_USERNAME }}
container-registry-password: ${{ secrets.REGISTRY_PASSWORD }}
secret-name: demo-k8s-secret
- uses: Azure/k8s-deploy@v1.3
with:
manifests: |
manifests/deployment.yml
manifests/service.yml
images: |
demo.azurecr.io/k8sdemo:${{ github.sha }}
imagepullsecrets: |
demo-k8s-secret
Build container image and deploy to any Azure Kubernetes Service cluster
on: [push]
jobs:
build:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- uses: Azure/docker-login@v1
with:
login-server: contoso.azurecr.io
username: ${{ secrets.REGISTRY_USERNAME }}
password: ${{ secrets.REGISTRY_PASSWORD }}
- run: |
docker build . -t contoso.azurecr.io/k8sdemo:${{ github.sha }}
docker push contoso.azurecr.io/k8sdemo:${{ github.sha }}
- uses: Azure/k8s-set-context@v1
with:
kubeconfig: ${{ secrets.KUBE_CONFIG }}
- uses: Azure/k8s-create-secret@v1
with:
container-registry-url: contoso.azurecr.io
container-registry-username: ${{ secrets.REGISTRY_USERNAME }}
container-registry-password: ${{ secrets.REGISTRY_PASSWORD }}
secret-name: demo-k8s-secret
- uses: Azure/k8s-deploy@v1.3
with:
manifests: |
manifests/deployment.yml
manifests/service.yml
images: |
demo.azurecr.io/k8sdemo:${{ github.sha }}
imagepullsecrets: |
demo-k8s-secret
Sample workflows for new traceability fields support
- Environment variable
HELM_CHART_PATHS
is a list of helmchart files used in k8s-bake and k8s-deploy - Use script to build image and add
dockerfile-path
label to it. The value expected is the link to the dockerfile. If your dockerfile is in the same repo and branch where the workflow is run, it can be a relative path and it will be converted to a link for traceability. - Run docker login action for each image registry - in case image build and image deploy are 2 distinct jobs in the same or separate workflows.
End to end workflow for building and deploying container images
on: [push]
env:
NAMESPACE: demo-ns2
jobs:
build-and-deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- uses: Azure/docker-login@v1
with:
login-server: contoso.azurecr.io
username: ${{ secrets.REGISTRY_USERNAME }}
password: ${{ secrets.REGISTRY_PASSWORD }}
- run: |
docker build . -t contoso.azurecr.io/k8sdemo:${{ github.sha }} --label dockerfile-path=./Dockerfile
docker push contoso.azurecr.io/k8sdemo:${{ github.sha }}
# Set the target AKS cluster.
- uses: Azure/aks-set-context@v1
with:
creds: '${{ secrets.AZURE_CREDENTIALS }}'
cluster-name: contoso
resource-group: contoso-rg
- uses: Azure/k8s-create-secret@v1
with:
container-registry-url: contoso.azurecr.io
container-registry-username: ${{ secrets.REGISTRY_USERNAME }}
container-registry-password: ${{ secrets.REGISTRY_PASSWORD }}
secret-name: demo-k8s-secret
- uses: Azure/k8s-deploy@v1.2
with:
manifests: |
manifests/deployment.yml
manifests/service.yml
images: |
contoso.azurecr.io/k8sdemo:${{ github.sha }}
imagepullsecrets: |
demo-k8s-secret
CI workflow to build image and add dockerfile-path
label to it. This image can then be used in another CD workflow.
on: [push]
env:
NAMESPACE: demo-ns2
jobs:
build:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- uses: Azure/docker-login@v1
with:
login-server: contoso.azurecr.io
username: ${{ secrets.REGISTRY_USERNAME }}
password: ${{ secrets.REGISTRY_PASSWORD }}
- run: |
docker build . -t contoso.azurecr.io/k8sdemo:${{ github.sha }} --label dockerfile-path=https://github.com/${{github.repo}}/blob/${GITHUB_REF##*/}/Dockerfile
docker push contoso.azurecr.io/k8sdemo:${{ github.sha }}
CD workflow using bake action to get manifests deploying to a Kubernetes cluster
on: [push]
env:
NAMESPACE: demo-ns2
HELM_CHART_PATHS: |
./helmCharts/file1
jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- uses: Azure/docker-login@v1
with:
login-server: contoso.azurecr.io
username: ${{ secrets.REGISTRY_USERNAME }}
password: ${{ secrets.REGISTRY_PASSWORD }}
# Set the target AKS cluster.
- uses: Azure/aks-set-context@v1
with:
creds: '${{ secrets.AZURE_CREDENTIALS }}'
cluster-name: contoso
resource-group: contoso-rg
- uses: Azure/k8s-create-secret@v1
with:
namespace: ${{ env.NAMESPACE }}
container-registry-url: contoso.azurecr.io
container-registry-username: ${{ secrets.REGISTRY_USERNAME }}
container-registry-password: ${{ secrets.REGISTRY_PASSWORD }}
secret-name: demo-k8s-secret
- uses: azure/k8s-bake@v1
with:
renderEngine: 'helm'
helmChart: ${{ env.HELM_CHART_PATHS }}
overrideFiles: './aks-helloworld/values-override.yaml'
overrides: |
replicas:2
helm-version: 'latest'
id: bake
- uses: Azure/k8s-deploy@v1.2
with:
manifests: ${{ steps.bake.outputs.manifestsBundle }}
images: |
contoso.azurecr.io/k8sdemo:${{ github.sha }}
imagepullsecrets: |
demo-k8s-secret
Contributing
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.
When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.