The Airflow database stores lots of logs and information about dag_runs, tasks, sensors, SLA misses, etc.
This data will eventually fill Airflow's database enough that it becomes unusable.
We need to put some mechanism in place that removes unnecessary data from the DB, probably data older than a given number of months.
Here's an example of how we could do it:
https://cloud.google.com/composer/docs/cleanup-airflow-database