If ML pipeline runs for many days and suddenly system crashes or server restarts, then how Metaflow handles this situation so that information about the pipeline does not get lost?

This section covers failure scenarios https://docs.metaflow.org/scaling/failures