We have been getting continuous failures in our refreshed (scheduled and manual) and I can't figure out why. This has been occuring for about 2 weeks although this weekend was worse than normal. To make things more strange, the message is not specifying the dataset:
Everywhere I can find this error referenced on a dataset using incremental refresh, it never specifies a data source in that first sentence...not once. The reports that have failed for the same reason but are not on incremental refresh will specify a dataset name.
Refreshing in PBI Desktop works fine.
Gateway appears to be working fine since other refreshes work although I am currently working on sifting through whatever I can find there to try and find something.
Datasource for all datasets is a SQL Server connected via the gateway.
While watching operations on the SQL Server, when the datasets make it past the initial wait (generally 0-2 minutes when it gets this far) I can see the queries from the Power BI datasets get passed down and run/execute on the databases. They will run for what appears the expected amount of time then the query will no longer appear as a running function on the server (assuming this means the query is done and data has been pulled). Once the queries are off the server, the dataset still has the spinning circle in PBI Service, I would assume this is when it's building the cube or whatever else it does to finish prepping the dataset in the cloud - then the refresh will fail. I can't say this happens 100% of the time because I don't watch it consistently, but I'd say this is the pattern for 80-90% of the failures I watch from start to finish on the main problem datasets.
For this weekend (4/6 & 4/7), my failure counts:
15% of all refreshes failed. (80% of those failures are on incremental refresh datasets). We are on a P1 premium capacity; here is a screenshot of our refresh data:
We have run hot on our CPU and Memory consumption, but the failures came at all times including when we are not even close to peak consumption of CPU or Memory.
100% failure rate on one dataset (has incremental refresnh this dataset has never worked since these issues started popping up):