A while back, we implemented support for non _id clusterKeys on non-replicated collections. Since then (to my knowledge), we've decided to only support _id clusterKeys.

There are tests we use to validate clusteredIndex behavior which use a non _id clusterKey on a local database.

All tests that use a non _id clusterKey should be re-evaluated / changed to use _id until we decide to add full support for non _id clustered indexes.