We don't refetch documents from source because of failures of oplog application. Instead, we make sure the insert/update/delete and other index related operations never fail due to index constraints in SERVER-26202. If unexpected failures happen, e.g. DocumentTooLarge(17419), we restart the initial sync. This design is because idempotency issues may also come from rollback and unclean shutdown, in the latter cases a sync source may not be available.