Don't cancel refresh for all tables if one refresh fails
It is very irritating to wait for over an hour for a report to refresh, only to have every single table's upload cancelled out because one table failed.
I frequently have to work with Excel files on a shared source, as well as a variety of Oracle SQL and SQL Server databases. This causes problems when some tables could upload fine but get cancelled because of another user being in the excel file or an SQL connection timeout.
Power BI does a good job of letting me know which tables could not refresh and for what reason, but it would be ideal if it had the notification for the failed table without cancelling the rest.
Daniel Davis commented
At minimum I'd say any query dependency "tree" should be isolated. Leveraging Data Flows in Power BI Service might be an appropriate area to implement this. But I can't for the life of me understand why one problem table should drag down the entire dataset when it affects nothing else. Never before in BI has this been an issue. Think about SSIS for example? If one table fails, it fails, the rest still run depending on precedence constraints.
EXAMPLE: My company tracks calls (customer service), Service tickets, Sales, and then we have built in tables such as Date dimensions. Sales, Service Tickets, and Call stats all pull from THREE different databases. I own none of them. Monday morning I come in and the callcenter is having an intermittent issue with their database. This causes the master dataset to fail in it's ENTIRETY. Sales execs are now mad they can't get their stats.
DISCLAIMER: now, before you say, why not use separate datasets? Consider that a single report cannot use multiple datasets. Some of my reports require overlaps such as Service Tickets and Call SLAs in the same report or sales and service items by product.
Alex Dupler commented
Totally agree. Some concern around making it hard to know if data is in sync, making this a togglable option or making PBI service and PBI desktop different in this regard could be a great nuance to the feature.
Matt Allington commented
I agree with Anonymous - this is a must fix. I realise there may be query dependencies, but there has to be some middle ground here. It would be much better to complete a full refresh and then send a message "query XYZ did not refresh - this may affect dependent queries" rather that "query XYZ failed, so you can't do anything until that is fixed".