What are the infrastructure requirements for deploying Feast in a multi-cloud environment?

All answers to this question.

Managing Feast across different clouds is technically possible because it is a "pluggable" framework. You can definitely point your feature_store.yaml to BigQuery for your offline historical data and a Redis cluster on AWS for your online serving. However, you need to be mindful of egress costs when materializing data from BigQuery to an external Redis instance. We found that the latency for fetching online features remains low (sub-10ms) as long as your inference service is in the same region as your Redis store. The "materialization" job—the process that moves data from offline to online—is what will take the hit on cross-cloud bandwidth, so schedule those jobs during off-peak hours.

   Answered 2025-05-24 by Danielle O'Connor

Have you considered using a unified data lake like Snowflake as your offline store to simplify the permissioning across your AWS and GCP accounts?

   Answered 2025-05-25 by Bradley Vance

• We did look into Snowflake, Bradley, but our data is already deeply nested in BigQuery. We decided to stick with a "best of breed" approach. To handle permissions, we’re using Workload Identity Federation, which allows our AWS-based Feast workers to securely authenticate with GCP without managing long-lived service account keys. It’s a bit more setup initially, but it makes the multi-cloud architecture much more secure and compliant with our internal SOC 2 requirements for data handling across environments.

     Commented 2025-05-26 by Sean Maxwell

Feast's provider-agnostic design is its biggest strength for enterprises that don't want to be locked into a single cloud vendor's ecosystem

   Answered 2025-05-27 by Gregory Hudson

• I agree with Gregory. Being able to swap out storage backends without rewriting our feature logic has been a lifesaver during our recent cloud migration

     Commented 2025-05-28 by Vanessa Jenkins