Share your thoughts on DirectQuery for Power BI datasets and Azure Analysis Services (preview)

Hi, here's what I found:

Important considerations when using DirectQuery
The following three points should be taken into consideration when using DirectQuery:

Performance and load - All DirectQuery requests are sent to the source database, so the time required to refresh a visual is dependent on how long that back-end source takes to respond with the results from the query (or queries). The recommended response time (with requested data being returned) for using DirectQuery for visuals is five seconds or less, with a maximum recommended results response time of 30 seconds. Any longer, and the experience of a user consuming the report becomes unacceptably poor. In addition, once a report is published to the Power BI service, any query that takes longer than a few minutes will timeout, and the user will receive an error.

Load on the source database should also be considered, based on the number of Power BI users who will consume the published report. Using Row Level Security (RLS) can have a significant impact as well; a non-RLS dashboard tile shared by multiple users results in a single query to the database, but using RLS on a dashboard tile usually means the refresh of a tile requires one query per user, thus significantly increasing load on the source database and potentially impacting performance.

Power BI creates queries that are as efficient as possible. Under certain situations however, the generated query may not be efficient enough to avoid refresh that would fail. One example of this situation is when a generated query would retrieve an excessively large number of rows (more than 1 million) from the back-end data source, in which case the following error occurs:

The resultset of a query to external data source has exceeded
the maximum allowed size of '1000000' rows.
This situation can occur with a simple chart that includes a very high cardinality column, with the aggregation option set to Don’t Summarize. The visual needs to only have columns with a cardinality below 1 million, or must have appropriate filters applied.

Security - All users who consume a published report connect to the back-end data source using the credentials entered after publication to the Power BI service. This is the same situation as data that is imported: all users see the same data, irrespective of any security rules defined in the backend source.

Supported features - Not all features in Power BI Desktop are supported in DirectQuery mode, or have some limitations. In addition, there are some capabilities in the Power BI service (such as Quick Insights) that are not available for datasets using DirectQuery. As such, the limitation of such features when using DirectQuery should be taken into consideration when determining whether to use DirectQuery.