Create and serve online features¶

Data freshness¶

A feature view with online feature serving automatically synchronizes data from its offline feature views.

Use the target_lag parameter to configure how often data is synchronized to your online feature table. You can set this value from a minimum of 10 seconds to a maximum of 8 days.

Refresh modes¶

Snowflake uses the following refresh modes to update the data:

• Incremental Refresh: This is the preferred and most efficient mode. Snowflake tracks changes in the sources and merges only the new or updated rows into the online store. This minimizes compute and I/O costs.

• Full Refresh: This mode drops all existing data in the table and reloads everything from the source. It is more resource-intensive and is used when an incremental refresh is not possible.

You can explicitly set the refresh mode to INCREMENTAL or FULL, or set it to AUTO to let Snowflake determine the most efficient available refresh mode.

Time series data handling¶

To ensure data consistency, you can specify a timestamp_col. When multiple rows with the same primary key are found in the source, Snowflake only ingests the version with the most recent timestamp. If you don’t specify a timestamp column, the most recently processed row takes precedence.

GRANT CREATE ONLINE FEATURE TABLE ON SCHEMA IDENTIFIER($SCHEMA_FQN) TO ROLE IDENTIFIER($FS_ROLE_PRODUCER);

GRANT SELECT, MONITOR ON FUTURE ONLINE FEATURE TABLES IN SCHEMA IDENTIFIER($SCHEMA_FQN) TO ROLE IDENTIFIER($FS_ROLE_CONSUMER);

GRANT SELECT, MONITOR ON ALL ONLINE FEATURE TABLES IN SCHEMA IDENTIFIER($SCHEMA_FQN) TO ROLE IDENTIFIER($FS_ROLE_CONSUMER);

from snowflake.ml.feature_store import FeatureView
from snowflake.ml.feature_store.feature_view import OnlineConfig

online_config = OnlineConfig(enable=True, target_lag="30 seconds")

fv = FeatureView(
    name="MY_FV",
    entities=[entity],
    feature_df=my_df, # Snowpark DataFrame containing feature transformations
    timestamp_col="ts", # optional timestamp column name in the dataframe
    refresh_freq="5 minutes",
    refresh_mode="AUTO", # refresh mode of the feature data
    desc="my feature view", # optional description
    online_config=online_config
)

fv = fs.register_feature_view(feature_view=fv, version="v1")

Update a feature view to enable/disable online feature serving¶

For existing feature views, you can update the online feature serving configuration using the update_feature_view method. You can use this method to enable online feature serving for existing feature views.

Use the following code to enable online feature serving.

# Enable online feature serving

fs.update_feature_view(
    name="<name>",
    version="<version>",
    online_config=OnlineConfig(enable=True, target_lag="5 minutes")
)

Use the following code to disable online feature serving.

# Disable online feature serving

fs.update_feature_view(
    name="<name>",
    version="<version>",
    online_config=OnlineConfig(enable=False)
)

Retrieve features from online storage¶

To retrieve feature values from online storage for a given sample, use the read_feature_view method and pass the list of feature names as well as the join keys of the sample:

fs.read_feature_view(
    feature_view=fv,
    keys=[["<k_1>", "<k_2>"]],
    feature_names=["<feature1>", "<feature2>", "<feature3>"],
    store_type=StoreType.ONLINE
)

Suspend/resume online data refresh¶

Use the following code to temporarily suspend data refresh.

fs.suspend_feature_view(feature_view=fv)

Use the following code to resume data refresh.

fs.resume_feature_view(feature_view=fv)

These operations suspend/resume both the offline feature view (dynamic table and associated task) and the online feature table (if it exists) to ensure consistent state across all storage types.

Manually refresh feature view¶

fs.refresh_feature_view(
    feature_view=fv,
    store_type=<store_type>
)

The store_type argument specifies whether to refresh offline (StoreType.OFFLINE) or online (StoreType.ONLINE) feature data.

View refresh history¶

fs.get_refresh_history(
    feature_view=fv,
    store_type=store_type
)

The store_type argument specifies whether to return the offline (StoreType.OFFLINE) or online (StoreType.ONLINE) store refresh history.