datarobot-bosun 9.2.18

datarobot-bosun module providing MLOps Management framework and plug-ins

Bosun - DataRobot's MLOps event manager

This is the Python Bosun package, which complements the Bosun distribution. In this package you will find two important components of the Bosun system: - The Bosun Python plugin framework - The Bosun model connector framework

Installation

The Bosun wheel is provided as part of the MLOps Agents tarball. To install the Bosun Python module:

pip install datarobot-bosun

Additional Plugin Dependencies

The basic install will provide support for the filesystem plugin; however, there are several other reference plugins that are provided with this package but which require additional dependencies that are not installed by default. For convenience, we provide the following optional dependencies that will install the packages necessary for the relevant plugin:

• docker - Installs dependencies for the docker plugin.
• kubernetes - Installs dependencies for the kubernetes plugin.
• s3 - Installs dependencies for the s3 plugin. It also enables support for fetching external model artifacts from S3 (regardless of plugin used).

You can install these extra dependencies using pip, for example:

pip install datarobot-bosun[docker]

You can also install multiple sets of dependencies at once, for example:

pip install datarobot-bosun[kubernetes,s3]

The Bosun Python Plugin Framework

This framework provides a simple way to develop a Bosun plugin in Python. A Bosun Python plugin is a Python class that implements several API methods, one method for each Bosun action. The plugin framework provides a runner scrip that is used to run the plugin as an external program. This way the plugin writer does not need to worry about handling command line arguments.

Example

plugin-runner --plugin test --config my_config.yaml --action pe_status --status-file /tmp/status.json

The example above will invoke the plugin-runner program which in turn will load and run the test plugin. The my_config.yaml file will be loaded and the configuration will be provided to the plugin. The action to run is pe_status which is supposed to return the status of the prediction environment. The --status-file argument is the path to the status file that should contain the plugin result of running the action.

The Bosun Model Connector Framework

This framework provides a simple way to develop and test model connectors. A model connector is a program that fetches the model artifact given the model metadata. This framework provides some built-in support for models stored on S3 and local file systems.