Go backend for MLflow

In order to increase the performance of the tracking server and the various stores, we propose to rewrite the server and store implementation in Go.

Getting started

Installation

pip install mlflow-go-backend

CLI Usage

mlflow-go-backend is meant to be a drop-in-replacement for mlflow.

You can update your existing you used mlflow command with mlflow-go:

- mlflow server --backend-store-uri postgresql://postgres:postgres@localhost:5432/postgres
+ mlflow-go server --backend-store-uri postgresql://postgres:postgres@localhost:5432/postgres

Every existing setting of mlflow server can be passed to mlflow-go.

Caution

The Go implementation currently does not support file storage as a backend store. You must provide the --backend-store-uri argument pointing to a database.

Python Usage

import mlflow
import mlflow_go

# Enable the Go client implementation (disabled by default)
mlflow_go.enable_go()

# Set the tracking URI (you can also set it via the environment variable MLFLOW_TRACKING_URI)
# Currently only database URIs are supported
mlflow.set_tracking_uri("sqlite:///mlflow.db")

# Use MLflow as usual
mlflow.set_experiment("my-experiment")

with mlflow.start_run():
    mlflow.log_param("param", 1)
    mlflow.log_metric("metric", 2)

Python Store Usage

import logging
import mlflow
import mlflow_go

# Enable debug logging
logging.basicConfig()
logging.getLogger('mlflow_go').setLevel(logging.DEBUG)

# Enable the Go client implementation (disabled by default)
mlflow_go.enable_go()

# Instantiate the tracking store with a database URI
tracking_store = mlflow.tracking._tracking_service.utils._get_store('sqlite:///mlflow.db')

# Call any tracking store method
tracking_store.get_experiment(0)

# Instantiate the model registry store with a database URI
model_registry_store = mlflow.tracking._model_registry.utils._get_store('sqlite:///mlflow.db')

# Call any model registry store method
model_registry_store.get_latest_versions("model")

Why bother?

For that sweet performance my friend!

Initial benchmarks show us that critical API calls are faster and Go can handle more concurrent requests.

Note

These were initial results and can perhaps still be more optimized.

See benchmarks for more information.

Contribution

Try it out!

We advocate this project as a drop-in replacement for mlflow. Please give it a try, and let us know if anything isn't working for you! We look forward to your feedback!

Missing Endpoints

Hey there! We need your help! This project is not yet a full port of the current tracking server. Not every endpoint in the REST API is implemented in Go. Those missing are redirected to the existing Python implementation. Please consider helping us out and implement a missing endpoint.

Not yet a Go wizard? We've collected some resources to learn Go. Trust us, it will be an enriching and fun experience!

Community

There is a dedicated Slack channel called #mlflow-go on the official mlflow Slack. Shoot us a message if you have any questions!