Query Federation

Query federation enables a single query to access data across many sources without copying data: a federated query can join PostgreSQL tables with S3 files and Snowflake data in a single query. The federated query engine coordinates execution: determine which data lives where, push predicates to each source (filter at source when possible), fetch required data, combine results. Execution plan might push down SELECT and WHERE to database, handle JOIN at query level, combine with S3 data. The goal is minimizing data movement: execute as much computation as possible at each source.

Query federation became practical with query engines like Trino and Presto that support multiple connectors. Traditional federated databases required explicit setup; modern approaches dynamically connect to sources. Trade-offs include: queries may be slower than querying single system (network overhead, lack of cross-source optimization), dependency on all sources being available (one slow source slows entire query).

In practice, federation is useful for specific scenarios: accessing rarely-used data in specialized systems, connecting to third-party data without replicating, or joining operational and analytical data. For frequently-accessed data, replication is usually more efficient than federation.