Investigate ‘Failure’ actions

[vivien57]

Hello everyone,

Our F64 capacity is sometimes heavily used due to user actions that are in status failure (a lot of resources are used until they reach a time-out, I think).

How do you go about investigating these failed actions to find out what the root cause is?

Thanks in advance for your feedback,

Vivien

[Anonymous]

Hi @vivien57 

Here are some directions and ideas for troubleshooting the failures, I hope it will be helpful.

• As the failures happened on query activities on some datasets, identify whether these activities are interactive operations or background operations on those datasets, and try to find more detailed error messages. If they are background operations like refreshing a dataset, go to its Refresh history to check the detailed error messages. If they are interactive operations like user viewing/interacting with a report, ask the users if they received any error message when their operations failed. 
• Go to the Capacity Utilization and Throttling section in the metrics app. Find out which periods of capacity usage are low, and then move some background operation activities to those hours of low usage. For example, schedule your refreshes for less busy times, especially if your semantic models are on Power BI Premium. If you distribute the refresh cycles for your semantic models across a broader time window, you can help avoid peaks that might otherwise overtax available resources.
• Let the users perform the same operations on the same datasets at a less busy time and check whether the operations would fail. If the operations succeed, it indicates that this may be a problem with a lack of resources. You can try rearranging background operations as the previous point mentions or try upgrading the capacity. 
• Check whether the datasets are in Import mode or DirectQuery mode. In Import mode, the data is pre-loaded into memory, so the concurrency limits are generally higher. DirectQuery/LiveConnect mode has several limitations, such as a one million-row limit for returning data and a 225-seconds response time limit for running queries. There are some concurrent connections and parallelism limitations for DirectQuery/LiveConnect in different SKUs. For detailed information, you can refer to What is Power BI Premium? - Power BI | Microsoft Learn
• If all of above don't help, you might consider adjusting/optimising the datasets to improve the performance.

Best Regards,
Jing
If this post helps, please Accept it as Solution to help other members find it. Appreciate your Kudos!

[frithjof_v]

The screenshot from the metrics app shows that this is an interactive operation called "Query" on a Dataset. The title of the visual says "Interactive operations for timerange". 

https://learn.microsoft.com/en-us/fabric/enterprise/fabric-operations

I couldn't find the word "dataset" in the list of operations in the link above. Perhaps the operation shown in the screenshot is the same as "Interactive query" on a semantic model?

Is this a Direct Lake or Import Mode (or DirectQuery) semantic model?

Here are some ideas for finding out which queries failed:

• Find out which user triggered the failed queries, and ask them if they got any error message.
• You could try using the report yourself and see if any visuals fail.
• You can also open the report in Power BI Desktop and run performance analyzer on each report page. Then you will find out which queries are slow to load (performance issues).
• Perhaps it's possible to use Azure Log Analytics to find information about failed queries (although I haven't tried it personally).

-----------

On a side note: rescheduling background operations would probably not help much even if the issue was related to background operations. As background operations are smoothed over a 24 hour period, it shouldn't matter much about the specific time the operation gets triggered. Instead, reducing the number of refreshes within a 24-hour period would help to reduce the utilization (if background operations were the main issue here).

https://learn.microsoft.com/en-us/fabric/enterprise/throttling#balance-between-performance-and-relia...

However, in this case we are looking at interactive operations (ref. the screenshot), not background operations. So background operations are not directly relevant for this case anyway.

[Anonymous]

Hi @frithjof_v 

Thank you for highlighting the areas I missed and correcting my mistakes. 

The Metrics app still uses the old term "dataset" rather than the new term "semantic model". I think you are correct that the operation is the same as "Interactive query" on a semantic model.

As the status is "Failure" not "Rejected"/"Delayed", and their throttling time is 0s, does that mean it's more likely to be an issue with query timeouts or query data volume over limits? Maybe the cause is more related to the semantic model and the Power BI report? Of course, in the case of DirectQuery/LiveConnect, may also need to consider the impact of data sources and networks.

Best Regards,
Jing