Vertex AI Feature Store: set up online feature serving (current Feature Store, BigQuery-based) with Bigtable online serving

Steps

1. Prepare your feature data as a BigQuery table or view — the current Vertex AI Feature Store reads features directly from BigQuery rather than requiring a separate offline ingestion step.
2. Register the data source by creating a feature group (a Feature Registry resource associated with that BigQuery table/view), then create individual features within the group, each mapped to a specific column in the BigQuery source.
3. Create an online store instance configured for Bigtable online serving — this is the current supported serving backend (the alternative, 'Optimized online serving', is deprecated: no new features after May 17 2026 and full sunset/API removal on Feb 17 2027, so new setups should use Bigtable online serving).
4. Create a feature view inside that online store, referencing the feature group(s)/features you registered; configure a data sync — Bigtable online serving supports both scheduled sync (cron-based) and continuous (near real-time) sync from the BigQuery source.
5. Start/monitor the data sync operation so feature values are populated into the online store; once synced, fetch feature values for low-latency online serving using the Bigtable-online-serving fetch API/SDK call against the feature view.
6. If you need to preserve exact null values from the source, use a feature group created with dense=true, Bigtable online serving, and scheduled (cron) sync — other combinations may not represent nulls correctly.