Cost allocation and optimization in MLOps - Commands & Configuration

Introduction

Managing cloud or infrastructure costs is important to avoid surprises in your bills. Cost allocation helps you track which projects or teams use resources. Optimization helps reduce waste and save money by using resources efficiently.

When you want to see how much each team spends on cloud resources to plan budgets better

When you notice your cloud bill is higher than expected and want to find which services cause it

When you want to automatically stop or scale down unused resources to save money

When you want to compare costs between different machine types or storage options before choosing

When you want to track cost trends over time to catch unexpected spikes early

Commands

Create a new MLflow experiment to track runs and their resource usage for cost allocation.

Terminal

mlflow experiments create --experiment-name cost-tracking

Expected Output

Experiment 'cost-tracking' created with ID 1

→

--experiment-name - Sets the name of the experiment to organize runs

Run your ML project under the cost-tracking experiment to log metrics and parameters for cost analysis.

Terminal

mlflow run . --experiment-name cost-tracking

Expected Output

2024/06/01 12:00:00 INFO mlflow.projects: === Run (ID 123abc) started ===

2024/06/01 12:00:10 INFO mlflow.projects: === Run (ID 123abc) succeeded ===

→

--experiment-name - Associates this run with the cost-tracking experiment

Log a custom metric named 'resource_cost' to record the cost of resources used in this run.

Terminal

mlflow metrics log --run-id 123abc --key resource_cost --value 12.50

Expected Output

Metric 'resource_cost' with value 12.5 logged for run 123abc

Start the MLflow UI to visualize experiments, runs, and cost metrics for optimization decisions.

Terminal

mlflow ui

Expected Output

2024/06/01 12:01:00 INFO mlflow.ui: Running MLflow UI at http://127.0.0.1:5000

Key Concept

If you remember nothing else from this pattern, remember: tracking resource usage and costs per run helps you find where to save money.

Common Mistakes

Not associating runs with an experiment name

Without grouping runs, it's hard to compare costs across projects or teams.

Always use --experiment-name to organize runs for clear cost allocation.

Forgetting to log cost metrics explicitly

MLflow does not track cost automatically; missing metrics means no cost data to analyze.

Use mlflow metrics log to record resource costs for each run.

Not using the MLflow UI to review cost data

Without visualization, spotting expensive runs or trends is difficult.

Run mlflow ui to see cost metrics and optimize resource use.

Summary

Create an MLflow experiment to group runs for cost tracking.

Run your ML project under this experiment to log resource usage.

Log custom cost metrics for each run to measure spending.

Use the MLflow UI to visualize and analyze costs for optimization.