🤖 AI Training Workflow

Simple GCP Infrastructure as Code for machine learning training workflows.

What Gets Created

• VM (e2-medium) - AI/ML training compute instance
• GCS bucket - dataset storage and results
• Labels - resource organization and tracking

Quick Start

Prerequisites:

• Terraform ~> 1.7
• Google Cloud CLI
• GCP project with billing

Setup:

gcloud auth login
gcloud config set project YOUR_PROJECT_ID

Run Workflow:

.\run-workflow.ps1 -DurationHours 2 -ProjectId your-gcp-project

Notes:

• The launcher auto-sanitizes the owner label from your gcloud account (emails like user@example.com become user-example-com).
• If running Terraform manually, ensure owner in terraform.tfvars matches GCP label rules: lowercase letters, digits, - or _, max 63 chars.

State storage:

• Terraform state is stored remotely in GCS at gs://firefly-tfstate-backend/vm/default.tfstate.
• The launcher ensures the bucket exists and migrates any local state automatically.

What Happens

1. Deploy - VM + bucket created with proper labels
2. Train - VM runs ML training workflow (~15 minutes)
3. Monitor - Background monitoring for remainder of time
4. Cleanup - Auto-destroy after workflow completion

Files

• main.tf - Core infrastructure
• variables.tf - Configuration options
• outputs.tf - Resource info
• startup-script.sh - AI training simulation
• run-workflow.ps1 - One-command launcher

Cost

~$0.10 for 2-hour workflow (e2-medium + minimal storage)

Manual Operation

If you prefer manual control:

# 1. Setup
terraform init -reconfigure -migrate-state   # uses remote GCS backend
cp terraform.tfvars.example terraform.tfvars
# Edit terraform.tfvars with your project_id (and optional labels). Ensure `owner` is label-compliant.

# 2. Deploy  
terraform plan
terraform apply

# 3. Wait for training completion (2 hours)

# 4. Cleanup
terraform destroy

---

Perfect for: ML experimentation, model training, automated workflows, cost-effective compute 🤖