To Reproduce Steps to reproduce the behavior:
RHOSP 17.1 "Configuring persistent storage" guide explains "cinder retype" command use cases telling that

If the retype operation requires moving the volume from one back end to another, the Block Storage service requires a specific flag

The thing is that "backend" definition is ambiguous if pools are involved: from Cinder scheduler perspective this is a host behind the volume. While Cinder generally uses "backend" definition differently (a good example is enabled_backends parameter).

Expected behavior
I'd rather change "backend" to "host".

Bug impact
It tricks some users into believing that they don't need to define migration_policy when migrating volumes between pools from the same backend.