Databricks

Overview

The "Update Serving Endpoint" operation in the Model Serving resource allows users to modify an existing model serving endpoint on Databricks. This node is useful for managing machine learning deployment endpoints by updating which models are served and how traffic is routed among them. Typical scenarios include rolling out new model versions, adjusting traffic splits between models for A/B testing, or scaling deployments by changing the served models dynamically.

For example, a data scientist might update the serving endpoint to shift 70% of traffic to a newly trained model while keeping 30% on the previous version to monitor performance before full rollout.

Properties

Output

The node outputs JSON data representing the updated state of the serving endpoint after applying the changes. This typically includes details such as the endpoint's name, the list of served models, their versions, and the current traffic distribution configuration.

If the node supports binary data output (not indicated here), it would generally represent artifacts related to the models or logs from the update operation.

Dependencies

• Requires an active connection to the Databricks API with appropriate authentication credentials (an API token).
• The node depends on the Databricks environment being accessible and the user having permissions to update model serving endpoints.
• No additional external services beyond Databricks are required.

Troubleshooting

• Common Issues:

  • Invalid JSON format in the "Served Models" or "Traffic Config" properties can cause request failures. Ensure these inputs are correctly formatted JSON strings.
  • Incorrect or non-existent endpoint names will result in errors indicating the endpoint cannot be found.
  • Insufficient permissions or expired API tokens may lead to authorization errors.

• Error Messages:

  • "Endpoint not found": Verify the endpoint name is correct and exists in your Databricks workspace.
  • "Invalid JSON format": Check the syntax of the JSON provided for served models and traffic config.
  • "Unauthorized" or "Authentication failed": Confirm that the API token credential is valid and has necessary scopes.

Resolving these usually involves validating input formats, confirming endpoint existence, and ensuring proper API credentials.

Links and References

• Databricks Model Serving Documentation
• Databricks REST API Reference