We recently tried to copy our DB to Atlas using mongomirror.

mongomirror has three phases:

1. copy the collection snapshots
2. build indexes
3. tail the oplog for updates since step 1 completed

In our case, it took 48h for step 2 (index build) to complete, and our oplog only had 22h worth of data. When we reached step 3 (after waiting 2 days!) the tool could not proceed.

It would be better if the tool could inter-mix steps 2 and 3, i.e. tail the oplog each 1 hour between index builds to stay current. This would potentially save a lot of heartbreak for users in the future.