In this section, we provide guides and references to use the MLflow connector.
Configure and schedule MLflow metadata and profiler workflows from the OpenMetadata UI:
To run the Ingestion via the UI you'll need to use the OpenMetadata Ingestion Container, which comes shipped with custom Airflow plugins to handle the workflow deployment. If you want to install it manually in an already existing Airflow host, you can follow this guide.
If you don't want to use the OpenMetadata Ingestion container to configure the workflows via the UI, then you can check the following docs to run the Ingestion Framework in any orchestrator externally.
Run Connectors from the OpenMetadata UILearn how to manage your deployment to run connectors from the UI
Run the Connector ExternallyGet the YAML to run the ingestion externally
External SchedulersGet more information about running the Ingestion Framework Externally
To extract metadata, OpenMetadata needs two elements:
- Tracking URI: Address of local or remote tracking server. More information on the MLflow documentation here
- Registry URI: Address of local or remote model registry server.
The first step is to ingest the metadata from your sources. To do that, you first need to create a Service connection first.
This Service will be the bridge between OpenMetadata and your source system.
Once a Service is created, it can be used to configure your ingestion workflows.
Select your Service Type and Add a New Service
Add a new Service from the Services page
Select your Service from the list
Provide a name and description for your Service.
OpenMetadata uniquely identifies Services by their Service Name. Provide a name that distinguishes your deployment from other Services, including the other Mlflow Services that you might be ingesting metadata from.
Note that when the name is set, it cannot be changed.
Provide a Name and description for your Service
In this step, we will configure the connection settings required for Mlflow.
Please follow the instructions below to properly configure the Service to read from your sources. You will also find helper documentation on the right-hand side panel in the UI.
Configure the Service connection by filling the form
- trackingUri: Mlflow Experiment tracking URI. E.g., http://localhost:5000
- registryUri: Mlflow Model registry backend. E.g., mysql+pymysql://mlflow:password@localhost:3307/experiments
Once the credentials have been added, click on Test Connection and Save the changes.
Test the connection and save the Service
In this step we will configure the metadata ingestion pipeline, Please follow the instructions below
Configure Metadata Ingestion Page
- Name: This field refers to the name of ingestion pipeline, you can customize the name or use the generated name.
- Mark Deleted Ml Models (toggle):: Set the Mark Deleted Ml Models toggle to flag ml models as soft-deleted if they are not present anymore in the source system.
- ML Model Filter Pattern (Optional): To control whether to include an ML Model as part of metadata ingestion.
- Include: Explicitly include ML Models by adding a list of comma-separated regular expressions to the Include field. OpenMetadata will include all ML Models with names matching one or more of the supplied regular expressions. All other ML Models will be excluded.
- Exclude: Explicitly exclude ML Models by adding a list of comma-separated regular expressions to the Exclude field. OpenMetadata will exclude all ML Models with names matching one or more of the supplied regular expressions. All other ML Models will be included.
- Enable Debug Log (toggle): Set the Enable Debug Log toggle to set the default log level to debug.
Scheduling can be set up at an hourly, daily, weekly, or manual cadence. The timezone is in UTC. Select a Start Date to schedule for ingestion. It is optional to add an End Date.
Review your configuration settings. If they match what you intended, click Deploy to create the service and schedule metadata ingestion.
If something doesn't look right, click the Back button to return to the appropriate step and change the settings as needed.
After configuring the workflow, you can click on Deploy to create the pipeline.
Schedule the Ingestion Pipeline and Deploy
If there were any errors during the workflow deployment process, the Ingestion Pipeline Entity will still be created, but no workflow will be present in the Ingestion container.
- You can then Edit the Ingestion Pipeline and Deploy it again.
- From the Connection tab, you can also Edit the Service if needed.