MLflow Prometheus Exporter Docker Image

This Docker image contains the MLflow Prometheus Exporter, which collects additional metrics from MLflow server and exposes them in the Prometheus exposition format.

Usage

To use this Docker image, you can either pull it from a Docker registry or build it locally.

Pull from Docker Registry

You can pull the pre-built Docker image from a Docker registry using the following command:

docker pull charmedkubeflowy/mlflow-prometheus-exporter:tag

Build Locally Docker

If you prefer to build the Docker image locally, you can follow these steps:

1. Clone the repository:

   git clone https://github.com/your-username/mlflow-prometheus-exporter.git

2. Build the Docker image:

   docker build -t mlflow-prometheus-exporter .

3. Run a container using the built image:

   docker run -d -p 8000:8000 --name exporter mlflow-prometheus-exporter

   Adjust the port mapping (-p) and container name (--name) as needed.

Build Locally ROCK

This repository contains also rockcraft.yaml which can be used to build ROCK oci image. To build the ROCK follow these steps:

1. Install essential tools

   sudo snap install rockcraft --classic --edge
   sudo snap install skopeo --edge --devmode
   

2. Build the rock

   rockcraft clean && rockcraft pack --verbosity=trace
   

3. Copy the resulted rock to your local Docker registry

   sudo skopeo --insecure-policy copy oci-archive:mlflow-prometheus-exporter_v1.0.0_22.04_amd64.rock docker-daemon:<registry_user>/mlflow-prometheus-exporter:tag
   

4. Now you can locally run it using Docker daemon

   docker run -d -p 8000:8000 --name exporter <registry_user>/mlflow-prometheus-exporter:tag
   

5. You can also store it on Dockerhub

   docker push <registry_user>/mlflow-prometheus-exporter:tag
   

Configuration

The MLflow Prometheus Exporter can be configured using environment variables:

• PORT: The port on which the exporter server will run (default: 8000).
• MLFLOW_URL: The MLflow server URL for collecting data (default: http://localhost:5000/).
• TIMEOUT: The timeout for polling new metrics in seconds (default: 30).

Example ussage:

python mlflow_exporter.py -p 8999 -u http://localhost:31380/ -t 30
PORT=8000 MLFLOW_URL=http://localhost:31380/ TIMEOUT=20 python mlflow_exporter.py

Contributing

If you'd like to contribute to the MLflow Prometheus Exporter, feel free to fork this repository and submit a pull request. Contributions are always welcome!