Many users who try ADX in direct query mode encounter errors right away.
The errors complain about lack of memory.
If the tables are small enough, it may work but still performance will not be as advertised on TV.
The reason in most cases is the behavior of joins in ADX as they are created by PBI.
In this article I’ll show different approaches to joining tables as used by PBI for related tables or as can be expressed in KQL in general.
I created a special table in the help cluster with 31 million rows that is big enough to demonstrate the performance differences between the variations.
Read the whole thing. This one’s a little surprising to me.