On-demand is not supported for cube data sources (neither for imported cubes nor for in-database cubes).
In contrast to data connections from relational data sources, selections for cube data connections always result in a single data table.
You cannot switch from an imported configuration to an in-database configuration in a cube data connection, the way you can with relational data sources, or vice versa. However, you can always replace a data table based on a cube data connection with another data table.
When cube hierarchies have been added to a connection, the hierarchies in Spotfire will not be updated when the data connection is edited. However, you can always recreate new hierarchies manually.
Import of large amounts of data may not work. If you try to import and it fails, edit the connection and limit the selected dimensions and/or categories some more.
The estimation of the expected import size is only an approximation and cannot give any definitive information regarding whether or not an import will work. If the estimation takes a long time it often, but not always, means that the result of the selection is too big to import.
If multiple instances of one leaf value exist you may get different behavior between imported data and in-db data, since the imported data may have lost hierarchical information required in order to distinguish the leaf values from each other. For example, there is more than one city called Berlin in the world and after an import, all of those cities will be clumped together into one group in the flattened data table, unless you make sure to always include full hierarchies on all axes.
You will always get the leaf nodes for those branches in the hierarchies you select. This means that you can only select full hierarchies, not single columns from a hierarchy.