Hello, I also had this issue. It was because I was trying to read a DLT table with a Machine Learning Runtime. At time of writing, Machine Learning Runtimes are not compatible with shared access mode, so I ended up setting up two clusters, one MLR assigned access mode, then a normal runtime shared access mode, and then played a fun game of dependency jenga. 

I was only setting up a demo, so no big deal. If you're looking at something in production, I'm not sure I could recommend this approach unless it's a last resort.