1. Secure Model Deployment with GCP IAM Conditions

    Python

    When deploying a model securely on Google Cloud Platform (GCP), one crucial aspect is managing access using Identity and Access Management (IAM) policies. IAM policies define who (users or service accounts) has what access (roles) to which resource. Each policy is composed of multiple bindings, where each binding ties together a single role with a list of members (users, groups, or service accounts), and optionally, role granting can be conditioned on certain aspects of the request (IAM Conditions).

    GCP IAM Conditions are used to fine-tune permissions by specifying attributes that must be true for the role to be granted. Conditions can be based on resource tags, time of day, requester's IP, etc. This allows for a more granular and secure access control mechanism.

    Let's write a Pulumi program that secures a model deployment service with IAM Conditions. For illustrative purposes, this program will:

    • Create a Google Cloud service account that will be used to deploy the model.
    • Define an IAM policy with conditions that constrains the service account's role to specific conditions, like time of day or resource attributes.
    import pulumi import pulumi_google_native as google_native import json # Define the service account for model deployment model_deployment_service_account = google_native.iam.v1.ServiceAccount( "modelDeploymentServiceAccount", account_id="model-deployer", project="my-project-id", # Replace with your project ID description="Service account for model deployment" ) # Define the IAM policy conditions # As an example, this condition will only allow actions between 9am to 5pm time_based_condition = { "title": "Time-based Condition", "description": "Grant access only during business hours", "expression": "request.time < timestamp(\"17:00:00Z\") && request.time > timestamp(\"09:00:00Z\")", } # Define the IAM policy with conditions for the model deployment resource model_deployment_iam_policy = google_native.iam.v1.IamPolicy( "modelDeploymentIamPolicy", bindings=[ { "role": "roles/editor", # Grant the editor role for illustration "members": [model_deployment_service_account.email], # Bind the service account "condition": time_based_condition }, ], project="my-project-id", # Replace with your project ID policy_id="my-policy-id" # Replace with a suitable policy ID ) # Export the service account email and IAM policy ID for reference pulumi.export("service_account_email", model_deployment_service_account.email) pulumi.export("iam_policy_id", model_deployment_iam_policy.id)

    Here’s what each piece of the program does:

    • We've created a service account that is responsible for deploying models.
    • We're using google_native.iam.v1.ServiceAccount to create the service account specifying the project ID and a description.
    • An IAM policy with a condition (time_based_condition) is defined to restrict actions to certain business hours.
    • We're applying this policy to our deployment with google_native.iam.v1.IamPolicy, which binds a role with specific members under the defined condition.
    • The project ID and policy ID are placeholders here and should be replaced with your actual project ID and a defined policy ID respectively.
    • Finally, we export two outputs using pulumi.export, which allows us to retrieve the service account email and IAM policy ID outside of Pulumi, such as within the GCP console or other applications that might need this data.

    In this way, you can leverage Pulumi and GCP IAM Conditions to create secure and conditionally constrained service accounts for deploying models or managing other GCP resources.