Third-party online stores
This article describes how to work with third-party online stores for real-time serving of feature values. You can also use Databricks online tables for real-time feature serving with much less setup required. See Databricks Online Tables.
With real-time serving, you publish feature tables to a low-latency database and deploy the model or feature spec to a REST endpoint.
Databricks Feature Store also supports automatic feature lookup. In this case, the input values provided by the client include values that are only available at the time of inference. The model incorporates logic to automatically fetch the feature values it needs from the provided input values.
The diagram illustrates the relationship between MLflow and Feature Store components for real-time serving.
Databricks Feature Store supports these online stores:
Online store provider |
Publish with Feature Engineering in Unity Catalog |
Publish with Workspace Feature Store |
Feature lookup in Legacy MLflow Model Serving |
Feature lookup in Model Serving |
---|---|---|---|---|
Amazon DynamoDB |
X |
X (Feature Store client v0.3.8 and above) |
X |
X |
Amazon Aurora (MySQL-compatible) |
X |
X |
||
Amazon RDS MySQL |
X |
X |
Start using online stores
See the following articles to get started with online stores:
Automatic feature lookup with Databricks Model Serving (includes example notebook)