Requirements Folder

DAGs are just Python, so if a DAG imports pandas, boto3, or any other third-party package, that package needs to be installed into the Airflow environment itself — declared in a requirements.txt and installed at image build time (Docker) or environment setup time (MWAA).

Why it matters in practice:

• Not automatic — Airflow doesn’t install DAG-level imports on the fly; missing packages fail at DAG parse time, not just at task run time
• Version pinning — Airflow itself has known-compatible constraint files per version; unpinned installs can break the Scheduler, not just your DAG
• Rebuild required — changing requirements.txt means rebuilding the image/ environment, not just editing a DAG file

See also: Plugins