Dataflows would be really useful if you could connect to them via DirectQuery. This would mean that we could query enormous dataflows that are not practical to load to the Data model and can make agg...
This functionality is basically essential. All of the data intelligence tools need to work the following way..... 1) Incremental refresh to pull large raw / close to raw data into the service efficiently. Incremental refresh needs to be much more intelligent and robust than it is currently. 2) Transform / query against the raw data already in the service. DO NOT COPY THE OUTPUT SOMEPLACE ELSE!!!! 3) MUST!!!! NOT REQUIRE PREMIUM NODE. At a high level data ingest should be super efficient. NEVER ingest the same data twice. This means not only being able to detect new data based on a date column but also some mechanism to allow the data source owner to specify how to detect what data has been updated. Once data is in service DO NOT make a copy of it anyplace. Only make a copy if the owner of the transform specifically indicates that you can run the transform once and "cache" the output because he/she is certain that the results would never change or is ok with stale results. Default behavior should ALWAYS be something like a live connection to the data imported by the data owner.
