aissemble-foundation-versioning-service

aiSSEMBLE™ Foundation Versioning

This module contains the implementation for the aiSSEMBLE versioning service. The service provides the ability to version artifacts produced by a Solution Baseline machine-learning pipeline, and uses Maven to package and deploy the artifacts to Nexus.

Model Versioning

The following GET endpoint is provided for model versioning: /version/model/{run_id}

run_id is the unique training run identifier assigned to an MLflow training experiment. The versioning service currently only supports models that have been trained using the MLflow implementation of a Solution Baseline machine-learning pipeline. It will package the training run artifacts produced by MLflow and the model artifacts produced when saving through MLflow.

The following environment variables are required for model versioning:

• MLFLOW_TRACKING_URI: the MLflow tracking URI path specified in the training pipeline. This is where MLflow saves the training run artifacts.
• MODEL_DIRECTORY: The model directory path specified in the training pipeline. This is where the training pipeline saves the trained models.
• NEXUS_SERVER: The Nexus server to use to push the versioned artifact to. The credentials for this server must be provided in the Maven settings.xml file.

Example

Below is a basic example of how to leverage the versioning service:

• Extending the baseline docker image to include Nexus credentials

  FROM boozallen/aissemble-versioning:${VERSION_AISSEMBLE}
  
  COPY /custom/maven/settings/containing/nexus/credentials.xml /root/.m2/settings.xml
  

• Docker-compose configurations for running the service

  example-versioning-service:
      image: my-extended-versioning-service-image:latest
      ports:
        - '80:80'
      environment:
          MLFLOW_TRACKING_URI: /path/to/mlflow/tracking/uri
          MODEL_DIRECTORY: /path/to/saved/models
          NEXUS_SERVER: http://nexus-server:port
  

• Note: By default Versioning requires authorization and therefore must have security services running (policy-decision-point).
  • Add the following to the above docker-compose file:

    policy-decision-point:
        image: ${CONTAINER_REGISTRY}boozallen/policy-decision-point-docker:latest
        hostname: policy-decision-point
        container_name: policy-decision-point
        ports:
          - "8780:8080"
          - "9700:9000"
    

  • Add an auth.properties file to your docker image with the following content.

    pdp_host_url=http://policy-decision-point:8080/api/pdp
    

  • Then add the following environment variable that points to your properties file.
    ENV KRAUSENING_BASE /path/to/auth.properties/file
  • To disable authorization you can add the following to the auth.properties file: is_authorization_enabled = False.