​
Implementation

​
Primary Sources

​
Tables

Table sources are used directly via a view. Featureform will never write to a primary source.

​
Transformation Sources

SQL transformations are used to create a view. By default, those views are materialized and updated according to the schedule parameter. Deprecated transformations are converted to un-materialized views to save storage space.

​
Offline to Inference Store Materialization

When a feature is registered, Featureform creates an internal transformation to get the newest value of every feature and its associated entity using a ClickHouse ASOF JOIN. A Kubernetes job is then kicked off to sync this up with the Inference store.

​
Training Set Generation

Every registered feature and label is associated with a view table. That view contains three columns, the entity, value, and timestamp. When a training set is registered, it is created as a materialized view via a JOIN on the corresponding label and feature views.

​
Configuration

First we have to add a declarative ClickHouse configuration in Python.
clickhouse_config.py
import featureform as ff

ff.register_clickhouse(
    name = "clickhouse_docs",
    description = "Example offline store",
    team = "Featureform",
    host = "0.0.0.0",
    port = "9000",
    user = "default",
    password = "password",
    database = "default",
    ssl=False
)
Note: SSL connections typically use port 9440. Once our config file is complete, we can apply it to our Featureform deployment 
featureform apply clickhouse_config.py --host $FEATUREFORM_HOST
We can re-verify that the provider is created by checking the Providers tab of the Feature Registry.

​
Mutable Configuration Fields

  • description
  • username
  • password
  • port
  • ssl