This happened in an admin training class.

After the backup section, student used mongodump to create a backup of a secondary (from a replica set we had previously set up on everyone's laptops; for data, imagine importing the training data set and making a few changes to it).

Then, the student used mongorestore --drop to restore the dump onto their primary. At this point, their secondaries went into RECOVERING state, and never came back. Their last optimes (via rs.status()) were shown to be about the time the restore was performed.