How would you monitor the health of a data pipeline and identify issues?