In an ETL pipeline, what steps run first and which steps must wait for others to complete?

Typically, in an ETL pipeline, extraction runs first because you need to pull raw data from source systems. Once extraction completes, transformation can begin, which may involve cleaning, enriching, and aggregating the data. Loading into the target warehouse or data lake is the final step and must wait until transformation finishes to ensure only clean, consistent data is stored. In a DAG-based orchestrator like Airflow, you would set dependencies so that the load task depends on the transform task, which in turn depends on the extract task. If you have parallel extraction jobs, they can run concurrently, but downstream steps must wait for all upstream tasks to finish. Additionally, any downstream analytics or dashboards should be scheduled after the load step to guarantee they consume up-to-date data.