Dataflows and the enhanced compute engine

• When exact are the entities loaded into the backed SQL server database of the enhanced compute engine?
   - After the ingest flow completes?
   - Before used in another flow as a computed entity?
   - ? something else

Data is loaded to the compute engine's sql db as part of the ingest flow. All data loaded to entities in a workspace backed by a capacity with the engine enabled are loaded to the sql db

•  Can we control which entities will load into the enhanced compute engine database and which don't? Do ALL entities load into the database? What happens if we allocate 1 GB to enhanced compute engine and I have 5 GB worth of linked entities? Or 500 GB? Or a TB?

You can only control this by which workspaces are backed by capacities with the compute engine enabled. The allocation setting in the capacity admin page is about memory allocation. This is memory that is set aside for the engine and not able to be used by the tabular models on the capacity. because the sql db's are not in memory objects, the size of the data loaded to dataflows is not related to the memory allocation. instead the DB size consumes the 1TB of disk storage associated with each capacity.

• Why are merges (even using left outer join) still SUPER slow in the GUI? When I create the dataflow "final" table, using the linked entities for billing and patient, it can take 5 - 10 minutes before results show up in the web GUI.

I don't know. Its hard to make definite or even likely statements about performance without more detail.

• Is the size of the underlying entity csv file, which we can see in Power BI desktop when loading a dataflow entity, the same size used in the back end enhanced compute engine database? aka does a 1 GB text file mean it takes 1 GB in the database? If not - how do we have any idea how much memory something will take, so we can allocate appropriately?

See first answer. No there are no easy metrics here. That would be nice.

• What happens if trying to join dataflow entities that are too large to fit in the enhanced compute engine memory?

see first answer. since the DB is not an in memory object, this only happens as part of query execution. both power query and, i assume, the compute engine sql server have memory paging systems.