While the feature looks cool, something are not clear:
i) What is the criteria for moving the data from on-premise to cloud? Does Azure make its own decision or can we control which set of data goes in Azure?
iii) This "Uniqueness is not enforced via UNIQUE constraints and PRIMARY KEY constraints on a Stretch-enabled table" is a big bummer i think, sounds like this is great for a denormalized schema. If i have a table with PK's,lets say part of it goes to Azure , how would i even tick & tie the records where there is no PK in Azure ?