deployml-core 0.0.21

Infra for academia

deployml

Infrastructure for academia with cost analysis

Features

• 🏗️ Infrastructure as Code: Deploy ML infrastructure using Terraform
• 💰 Cost Analysis: Integrated infracost analysis before deployment
• ☁️ Multi-Cloud Support: GCP, AWS, and more
• 🔬 ML-Focused: Pre-configured for MLflow, experiment tracking, and model registry
• 🛡️ Production Ready: Security best practices and service account management

Instructions

poetry install
poetry run deployml doctor
poetry run deployml deploy --config-path your-config.yaml

docker build --platform=linux/amd64 -t gcr.io/mlops-intro-461805/mlflow/mlflow:latest .

gcloud auth configure-docker docker push gcr.io/PROJECT_ID/mlflow-app:latest

Cost Analysis Integration

deployml integrates with infracost to provide cost estimates before deployment:

Installation

brew install infracost

Once installed you will need to create a free infracost account before creating your API key.

To generate your infracost API key run the following command:

infracost auth login

If you want to retrieve your API key use:

infracost configure get api_key

Cost Analysis Configuration

Add cost analysis settings to your YAML configuration:

name: "my-mlops-stack"
cost_analysis:
  enabled: true              # Enable/disable cost analysis (default: true)
  warning_threshold: 100.0   # Warn if monthly cost exceeds this amount
  currency: "USD"            # Currency for cost display

Cloud Run Service Account Setup

When deploying MLflow, you must specify the service account email that Cloud Run will use. This service account must have permission to access the artifact bucket.

How to get a service account email

1. List your service accounts:

   gcloud iam service-accounts list
   

2. (Recommended) Create a dedicated service account for MLflow if you don't have one:

   gcloud iam service-accounts create mlflow-cloudrun --display-name "MLflow Cloud Run Service Account"
   

3. Find the email for your service account (it will look like mlflow-cloudrun@YOUR_PROJECT.iam.gserviceaccount.com).

Grant the service account permissions

The Terraform module will automatically grant the service account the required permissions on the artifact bucket.

Deploying with Terraform

Pass the service account email as a variable:

terraform apply -var="cloud_run_service_account=mlflow-cloudrun@YOUR_PROJECT.iam.gserviceaccount.com"

Or, if using the CLI, ensure it passes this variable to Terraform.

Why this is needed

The MLflow server (running on Cloud Run) needs permission to list and read artifacts in the GCS bucket. This setup ensures the MLflow UI works for all users without manual permission fixes.