In MMAPv1 we need to scan the _id index to ensure we see documents once since records can move in the collection, allowing us to see duplicates or miss documents. In other storage engines this is not allowed, so we should be able to do table scans rather than index scans which can be orders of magnitude slower.
This would result in a user-facing change since we would no longer be returning documents in _id order, but that is not part of the documented behavior of $snapshot.
- is depended on by
-
SERVER-19781 QueryPlanner::plan() can sometimes build collscans for geonear and text queries
- Closed
- is related to
-
SERVER-22214 CloneCollection with query doing COLLSCAN although there is an index on field
- Closed
-
TOOLS-845 Mongodump very slow with custom _id
- Closed