We have an interesting problem. To defragment a large analytics cluster we deleted a large amount of data from it and then started to rebuild one of its secondaries. The deletion job was interrupted with db.killop() when it became clear the process to rebuild the secondaries would impinge on the upcoming Christmas holiday. After the job was killed, the DB was operating normally so we began the rebuild of one of the secondaries. After that, the primary went into rollback mode, and then into a restart loop because there is too much data to roll back. The primary moved to another secondary. The secondary we are rebuilding is also stuck in a loop. It gets to about 10% full, and then fails and restarts.