Configure catalog properties files for each data source (e.g., hive.properties, iceberg.properties, tpch.properties) pointing to the appropriate metastore URIs and object storage endpoints
Restart the Trino coordinator and verify that each catalog appears in SHOW CATALOGS and that tables within them are queryable with SHOW TABLES FROM catalog.schema
Write a cross-catalog JOIN query using fully qualified table names (catalog.schema.table) to federate data from two different sources in a single SQL statement
Use EXPLAIN or EXPLAIN ANALYZE to inspect the distributed query plan and verify that predicate pushdown is occurring in each catalog connector to limit data scanned
Monitor the Trino Web UI's query details page for stage-level data transfer volumes to identify cross-node shuffle bottlenecks in the federated query
Known gotchas
Trino federation pulls data from remote sources into the coordinator's worker memory for joins and aggregations; large cross-catalog joins that cannot be pushed down will materialize entire tables in Trino worker memory and may cause out-of-memory failures
Not all connectors support predicate pushdown for all data types; complex predicates on nested or semi-structured columns may not be pushed to the source, causing full table scans
Cross-catalog transactions are not supported; Trino executes read queries federally but write operations (INSERT INTO ... SELECT ... FROM other_catalog) are non-transactional and a failure midway leaves partial data in the target
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp