Adding a platform provider

This page describes how to add a platform provider to enable its applications.

PipeCD supports multiple platforms and multiple application kinds which run on those platforms. Platform provider defines which platform and where the application should be deployed to. So while registering a new application, the name of a configured platform provider is required.

Currently, PipeCD is supporting these five kinds of platform providers: KUBERNETES, ECS, TERRAFORM, CLOUDRUN, LAMBDA. A new platform provider can be enabled by adding a PlatformProvider struct to the piped configuration file. A piped can have one or multiple platform provider instances from the same or different platform provider kind.

The next sections show the specific configuration for each kind of platform provider.

Configuring Kubernetes platform provider

By default, piped deploys Kubernetes application to the cluster where the piped is running in. An external cluster can be connected by specifying the masterURL and kubeConfigPath in the configuration.

And, the default resources (defined at here) from all namespaces of the Kubernetes cluster will be watched for rendering the application state in realtime and detecting the configuration drift. In case you want to restrict piped to watch only a single namespace, let specify the namespace in the KubernetesAppStateInformer field. You can also add other resources or exclude resources to/from the watching targets by that field.

Below configuration snippet just specifies a name and type of platform provider. It means the platform provider kubernetes-dev will connect to the Kubernetes cluster where the piped is running in, and this platform provider watches all of the predefined resources from all namespaces inside that cluster.

apiVersion: pipecd.dev/v1beta1
kind: Piped
spec:
  ...
  platformProviders:
    - name: kubernetes-dev
      type: KUBERNETES

See ConfigurationReference for the full configuration.

Configuring Terraform platform provider

A terraform platform provider contains a list of shared terraform variables that will be applied while running the deployment of its applications.

apiVersion: pipecd.dev/v1beta1
kind: Piped
spec:
  ...
  platformProviders:
    - name: terraform-dev
      type: TERRAFORM
      config:
        vars:
          - "project=pipecd"

See ConfigurationReference for the full configuration.

Configuring Cloud Run platform provider

Adding a Cloud Run provider requires the name of the Google Cloud project and the region name where Cloud Run service is running. A service account file for accessing to Cloud Run is also required if the machine running the piped does not have enough permissions to access.

apiVersion: pipecd.dev/v1beta1
kind: Piped
spec:
  ...
  platformProviders:
    - name: cloudrun-dev
      type: CLOUDRUN
      config:
        project: {GCP_PROJECT}
        region: {CLOUDRUN_REGION}
        credentialsFile: {PATH_TO_THE_SERVICE_ACCOUNT_FILE}

See ConfigurationReference for the full configuration.

Configuring Lambda platform provider

Adding a Lambda provider requires the region name where Lambda service is running.

apiVersion: pipecd.dev/v1beta1
kind: Piped
spec:
  ...
  platformProviders:
    - name: lambda-dev
      type: LAMBDA
      config:
        region: {LAMBDA_REGION}
        profile: default
        credentialsFile: {PATH_TO_THE_CREDENTIAL_FILE}

You will generally need your AWS credentials to authenticate with Lambda. Piped provides multiple methods of loading these credentials. It attempts to retrieve credentials in the following order:

  1. From the environment variables. Available environment variables are AWS_ACCESS_KEY_ID or AWS_ACCESS_KEY and AWS_SECRET_ACCESS_KEY or AWS_SECRET_KEY.
  2. From the given credentials file. (the credentialsFile field in above sample)
  3. From the pod running in EKS cluster via STS (SecurityTokenService).
  4. From the EC2 Instance Role.

Therefore, you don’t have to set credentialsFile if you use the environment variables or the EC2 Instance Role. Keep in mind the IAM role/user that you use with your Piped must possess the IAM policy permission for at least Lambda.Function and Lambda.Alias resources controll (list/read/write).

See ConfigurationReference for the full configuration.

Configuring ECS platform provider

Adding a ECS provider requires the region name where ECS cluster is running.

apiVersion: pipecd.dev/v1beta1
kind: Piped
spec:
  ...
  platformProviders:
    - name: ecs-dev
      type: ECS
      config:
        region: {ECS_CLUSTER_REGION}
        profile: default
        credentialsFile: {PATH_TO_THE_CREDENTIAL_FILE}

Just same as Lambda platform provider, there are several ways to authorize Piped agent to enable it performs deployment jobs. It attempts to retrieve credentials in the following order:

  1. From the environment variables. Available environment variables are AWS_ACCESS_KEY_ID or AWS_ACCESS_KEY and AWS_SECRET_ACCESS_KEY or AWS_SECRET_KEY.
  2. From the given credentials file. (the credentialsFile field in above sample)
  3. From the pod running in EKS cluster via STS (SecurityTokenService).
  4. From the EC2 Instance Role.

See ConfigurationReference for the full configuration.


Last modified August 23, 2023: Release v0.45.0 (#4556) (600de7e3)