Airflow at Scale — Week 1 CheckPlatform mechanics: daemons, governance layers, sensors, and multi-cluster moves. Every question is answerable from the Week 1 brief.

A single-node Airflow install keeps crashing during heavy ETL windows. What is the FIRST structural change?

An AST-based DAG linter passes a file, but the cluster policy hook rejects the DAG at parse time. Why can both be correct?

Downstream systems get hammered at exactly midnight every night by thousands of DAG runs starting simultaneously. What's the platform-level fix?

Why do sensors dominate capacity on large Airflow clusters?

A platform caps metadata retention at 28 days to keep the UI and upgrades fast. What capability does this deliberately sacrifice?

During a planned DAG migration between clusters, why does coordination state (the switch list) live in ConfigMaps rather than Airflow Variables?