The comments for ReplicationRecovery::_truncateOplogTo indicate that it should truncate entries after and including 'truncateTimestamp'.

The implementation, however, finds the first oplog entry less than the 'truncateTimestamp' then tells the storage engine to truncate the oplog starting at that entry, and passes true for the 'inclusive' flag. This means it will actually truncate the entry one before the 'truncateTimestamp' passed it.

We need to audit exactly how and when the oplogTruncateAfterPoint is set and make sure we're truncating at the right points.