You stream-insert 10k social media messages/min into BigQuery and run aggregations immediately, but queries miss in-flight rows due to eventual consistency. How should you adjust the application design?

The correct answer addresses the eventual consistency of BigQuery streaming inserts. While data is immediately available for querying, it may not be fully consistent for a short period, meaning some recently inserted rows might be temporarily absent from query results. Estimating the average latency for data availability and waiting twice that long before querying ensures a higher probability of consistent results. Rewriting the application to load accumulated data every 2 minutes or converting to batch loading for individual messages would introduce significant delays in data availability, contradicting the need for immediate aggregations. Loading to Cloud SQL and then exporting hourly to BigQuery would also introduce substantial latency, making real-time aggregations impossible.