Performance Optimization Flashcards

Which command re-runs only the nodes that failed or were skipped in the previous invocation?

Answer: dbt retry

dbt retry reads run_results.json from the last run and executes only failed or skipped nodes.

What is the main purpose of the is_incremental() macro in an incremental model?

Answer: To return true only when the model already exists as a table and is not run with --full-refresh

is_incremental() lets you filter source rows only on subsequent runs, when the target table exists and no full refresh is requested.

Which incremental strategy deletes matching rows by unique_key and then inserts the new rows, which can be faster than merge on some warehouses?

Answer: delete+insert

The delete+insert strategy removes rows matching unique_key and inserts fresh rows, avoiding the cost of a MERGE statement.

How can you reduce the amount of data processed in development without changing production logic?

Answer: Use a target.name conditional to filter to recent data only in dev

Wrapping a date filter in an if target.name == 'dev' Jinja block limits dev data while prod runs the full dataset.

Which artifact contains per-node execution timing that helps identify the slowest models?

Answer: run_results.json

run_results.json records execution_time and timing details for every node in the last invocation.

What does the --empty flag do when used with dbt run?

Answer: Builds models with zero-row refs and sources to validate SQL cheaply

--empty limits refs and sources to zero rows so you can check that models compile and execute without processing real data.

Which materialization stores no object in the warehouse and is interpolated as a CTE into downstream models?

Answer: ephemeral

Ephemeral models are inlined as CTEs, so nothing is created in the database.