Capacity for warehouse increased abnormal

Hi QuangThanh​,

The timing you are seeing is interesting because Microsoft's current Warehouse consumption documentation now describes a different consumption model based on allocated compute over time.

Fabric Warehouse now measures consumption using dynamically allocated vNodes, with consumption conceptually being Active vNodes × Active Time. Microsoft also documents a one-minute baseline allocation window, after which consumption is measured per second based on the number of allocated vNodes.

You may access the current guidance on Warehouse usage reporting here.

So I would not assume that your Warehouse queries suddenly became 2.5x more expensive computationally just because the CU(s) increased. Under the current model, workload shape matters as well. For example, concurrency can cause additional vNodes to be allocated, and Microsoft also includes system-generated Warehouse activity in consumption.

Before considering an F64 resize, I would compare one representative period before and after the increase:

1. Use Capacity Metrics to identify the highest Warehouse CU time windows.
2. Use queryinsights.exec_requests_history and queryinsights.frequently_run_queries during those windows.
3. Compare query count, concurrency, execution patterns and whether the activity is user-generated or system-generated.

Microsoft specifically recommends using Capacity Metrics together with Query Insights for investigating unexpected Warehouse consumption.

If you find many short queries running repeatedly, or bursts of concurrent queries, I would look at reducing unnecessary polling/chattiness or consolidating those workloads before scaling the capacity.

I also would not use the higher CU number alone to conclude that F64 is insufficient. Check the actual throttling/carryforward state in Capacity Metrics first. If the capacity remains healthy, the higher reported CU consumption may mainly reflect the changed metering model rather than an immediate capacity-sizing problem.

I cannot find a public Microsoft statement confirming that every tenant switched specifically on August 11, so I would avoid tying your exact cutover date to the change without Microsoft confirming that part.

AI-assisted drafting: AI was used to help structure and phrase this response. I reviewed and validated the technical content before posting.