mongod opens every collection and index upon startup to clear plan cache and ensure settings are correct for the current version. When there are tens or hundreds of thousands of collections and index (1 file per collection/index) this can take minutes to complete, even on very fast storage. Consider an option to do this in parallel to improve startup time.
When starting the mongod service in a replicaset with ~50GB of data on extremely high-performance disk, it's taking approximately 2 minutes before the mongo database becomes available. This seems like a large regression from the 2.8 and 3.0 days.