Share your thoughts on DirectQuery for Power BI datasets and Azure Analysis Services (preview)

Outside of the feedback below the measures not flowing through and being able to delete tables/columns in the remote datasets are the biggest need for me. There's a need to have measures in the parent model which supports our enterprise reporting and governance around having a single source of truth/calculation for different measures across the organization. Right now they all pull in as errors so you'd have to remove them from the parent or leave them with errors in the remote dataset. There would then be rework to replicate measures in all remote datasets and maintenance to make sure everything is aligned going forward. Second thing would be to allow users to delete tables/columns in the remote dataset. The use case is for self-service users we want to provide a dataset that might have some sensitive tables/columns that exist in the parent, this would be easier/cleaner than having to come up with some sort of RLS/CLS/workspace access combination to achieve. 

•  Performance
  • Minimal testing on small to medium sized datasets seems fine. I'm sure it's slower than the parent dataset but that's to be expected
• Query editor experience--the remote model query doesn't show up in the query editor and only in the data source settings dialog. What are your thoughts?
  • Most of the query editor steps should probably be pushed back to the parent dataset rather than only existing in the remote dataset, though there's probably plenty of use cases for being able to use PQ editor in both. 
• Navigator experience
  • Works fine
• Thoughts around governance and permissions for models that leverage this feature
  • There's a use case for having the parent dataset permissions push to the remote datasets as well as the remote datasets having their own set of RLS/permissions. We'll use this to have the parent dataset support enterprise reporting with remote datasets filtered down to different subsets of business groups that have their own security needs controlled by admins
• Nesting models, i.e. building a composite model on top of a composite model
  • Haven't come up with a use case where this is really needed or wouldn't be able to be handled in another way, though I'm sure the need exists