Container

The container action type is an abstraction that can be used with multiple plugins. See here for an in-depth guide on the action type itself. Continue reading for more information on the container deploy action type that can be used with the Kubernetes plugin.

The Kubernetes plugins can deploy container deploy actions.

Garden will take the simplified container deploy specification and convert it to Kubernetes manifests, i.e. Deployment, Service and (if applicable) Ingress resources.

For example:

kind: Build
name: frontend
type: container
---
kind: Deploy
name: frontend
type: container
dependencies: [build.frontend] # <--- Need to specify the build as a dependency
spec:
  image: ${actions.build.frontend.outputs.deploymentImageId} # <--- The output from the Build action defined above
  ports:
    - name: http
      containerPort: 8080
  healthCheck:
    httpGet:
      path: /hello-frontend
      port: http
  ingresses:
    - path: /hello-frontend
      port: http
    - path: /call-backend
      port: http
...

To make sure Garden deploys the frontend with the right image version, we need to reference the output from the container build in the spec.image field.

We also configure a health check, a couple of ingress endpoints, and specify that this deploy depends on the backend deploy. There is a number of other options, which you can find in the container action reference.

If you need to use advanced (or otherwise very specific) features of the underlying platform, you may need to use more platform-specific action types (e.g. kubernetes or helm). The container action type is not intended to capture all those features.

Environment variables

Container services can specify environment variables, using the services[].env field:

kind: Deploy
name: frontend
type: container
spec:
  env:
    MY_ENV_VAR: foo
    MY_TEMPLATED_ENV_VAR: ${var.some-project-variable}
...

env is a simple mapping of "name: value". Template strings can also be used to interpolate values.

Secrets

You can reference secrets in environment variables. For Kubernetes, this translates to valueFrom.secretKeyRef fields in the Pod specs, which direct Kubernetes to mount values from Secret resources that you have created in the application namespace, as environment variables in the Pod.

For example:

kind: Deploy
name: frontend
type: container
spec:
  env:
    MY_SECRET_VAR:
      secretRef:
        name: my-secret
        key: some-key-in-secret
...

This will pull the some-key-in-secret key from the my-secret Secret resource in the application namespace, and make it available as an environment variable.

Note that you must create the Secret manually for the Pod to be able to reference it.

For Kubernetes, this is commonly done using kubectl. For example, to create a basic generic secret you could use:

kubectl --namespace <my-app-namespace> create secret generic --from-literal=some-key-in-secret=foo

Where <my-app-namespace> is your project namespace (which is either set with namespace in your provider config, or defaults to your project name). There are notably other, more secure ways to create secrets via kubectl. Please refer to the official Kubernetes Secrets docs for details.

Also check out the Kubernetes Secrets example project for a working example.

Last updated