Pyomo + Nextmv: From Local Model to Cloud-Native Optimization

This repository demonstrates how to evolve a basic Pyomo optimization model into a fully-featured, cloud-native application using Nextmv. We'll take you through three stages of development, showing how minimal changes unlock powerful capabilities like observability, collaboration, instance management, version control, and experimentation.

The Journey: Three Stages of Evolution

🏠 Stage 1: Basic Local Pyomo Model (starting-state/)

A standard Pyomo diet optimization model that runs locally:

cd starting-state/
python main.py

What you get:

• ✅ Working optimization model
• ✅ Local execution
• ❌ No version control
• ❌ No collaboration features
• ❌ No observability
• ❌ No experimentation

☁️ Stage 2: Run in Nextmv (run_in_nextmv_state/)

Changes made:

1. Added app.yaml manifest - Tells Nextmv how to run your model

cd run_in_nextmv_state/
tar czf - diet.dat | nextmv app run -a pyomo-example 
nextmv app output -a pyomo-example -r <insert runID here>

What you unlock:

• ✅ Observability: Detailed execution logs and metrics
• ✅ Collaboration: Share apps with team members
• ✅ Instance Management: Deploy and manage app versions
• ✅ Version Control: Track model changes over time
• ✅ Cloud Execution: Run on Nextmv's optimized infrastructure

🧪 Stage 3: Full Experimentation Platform (final_state/)

Additional changes:

1. Added statistics output - For experiment tracking and comparison
2. Exposed options in app.yaml - For model configuration

cd final_state/
tar czf - diet.dat | nextmv app run -a pyomo-example -o "solver=highs,limit_dairy=true"
nextmv app output -a pyomo-example -r <insert runID here>

What you unlock:

• ✅ Configuration Testing: Compare different solver configurations
• ✅ Parameter Sweeps: Test multiple scenarios automatically
• ✅ Performance Analytics: Track solution quality and runtime
• ✅ Business Metrics: Custom statistics for your domain

The Model: Diet Optimization

This is a classic optimization problem that minimizes the cost of a diet while meeting nutritional requirements:

• Objective: Minimize total cost of food
• Constraints:
  • Meet minimum nutritional requirements (calories, protein, vitamins, etc.)
  • Stay within maximum volume limit
  • Optional: Limit dairy products (experimentation feature)

Foods available: Cheeseburger, Ham Sandwich, Hamburger, Fish Sandwich, Chicken Sandwich, Fries, Sausage Biscuit, Lowfat Milk, Orange Juice

Key Changes Explained

The Minimal App Manifest (app.yaml)

type: python
runtime: ghcr.io/nextmv-io/runtime/pyomo:latest
python:
  pip-requirements: requirements.txt
files:
  - main.py
  - diet.py
configuration:
  content:
    format: "multi-file"
    multi-file:
      input:
        path: "."
      output:
        solutions: "."
        statistics: "statistics.json"

This tells Nextmv:

• Use the Pyomo-optimized runtime
• Include your Python files
• How to handle input/output data

Adding Experimentation Capabilities

1. Statistics Output (main.py):

import nextmv

statistics = nextmv.Statistics(
    result=nextmv.ResultStatistics(
        value=value(instance.cost),
        custom={
            "nvars": len(instance.x),
            "nconstraints": len(instance.nutrient_limit) + 1,
        },
    ),
)

2. Configurable Options (app.yaml):

configuration:
  options:
    items:
      - name: solver
        option_type: string
        ui:
          control_type: select
        additional_attributes:
          values: [cbc, glpk, highs]
      - name: limit_dairy
        option_type: bool
        default: false
        ui:
          control_type: toggle

Running the Examples

Prerequisites

Local Execution

pip install -r requirements.txt
python main.py

Deploy to Nextmv Cloud

python push.py

The Power of This Approach

With just two simple changes (adding app.yaml and renaming to main.py), you transform a local script into a cloud-native optimization application with:

• Zero infrastructure management
• Built-in experiment tracking
• Team collaboration features
• Automatic scaling
• Version control and rollback
• Performance monitoring

This is the power of "Nextmvifying" your optimization models - minimal changes, maximum impact.