trino-dbt-query-performance

Your dbt incremental model scans the full table on Trino when it lacks bounded, partition-aware is_incremental() filters. You must design watermarks and partition-aware filtering to prevent accidental full scans and reduce query runtime.

To stop small file proliferation in Iceberg during dbt writes, choose the right incremental write strategy by selecting between MERGE and delete+insert patterns. This approach avoids heavy writes and prevents small-file growth.

The best way to tune Trino session properties for slow dbt runs is configuring thread settings, applying join hints like BROADCAST, and running targeted post-run ANALYZE commands to improve cost-based optimization planning.

You should use materialized views instead of incremental models in dbt based on specific cost and freshness needs. Selecting between view, ephemeral, table, incremental, and materialized_view materializations directly balances query expense against data update frequency.

To prevent risky full-refresh dbt runs on large Trino tables, implement safety safeguards against production regressions. This involves calling out common anti-patterns and blocking full-refresh usage to protect massive datasets from expensive recomputation.