Airflow consists of a scheduler that decides which tasks to run, a web server, a metadata database, a DAG folder, and an executor that runs the tasks: locally, on Celery workers, or on Kubernetes. The triggerer handles deferred tasks. The choice of executor determines scalability.