Power BI Concurrency: Build a Representative Test
Distinguish licensed viewers, simultaneous activity, query complexity, and observed capacity pressure.
Справочная статья на английском
User count is not query concurrency
A tenant with 500 licensed users does not necessarily have 500 queries running at once. Conversely, one report interaction can trigger several visual queries. Describe how readers actually use the report: opening a page, changing a filter, drilling through, or exporting. Record the model and report version so later comparisons have a stable reference.
An illustrative morning test
Suppose 20 readers open a sales report around the same time while a refresh is running. Define the actions each test user performs and the expected totals after each filter. Compare this scenario with the same report when the refresh is not overlapping. This helps distinguish an interaction problem from resource competition. Do not infer a production limit from one successful test or assume that doubling CUs doubles the number of supported readers.
Licensing is a separate check
For organizational Power BI consumption, F64 or higher can support free-license viewers with the appropriate workspace or app access. Smaller F capacities do not remove the need for qualifying user licenses. Authors and administrators have their own licensing and permission requirements. Verify the current Microsoft licensing table before estimating the total cost of a rollout.
Evidence to retain
Record response times, failed actions, refresh outcomes, query behavior, and the relevant capacity metrics during the same observation window. Include the test client, network conditions, data size, and cache state. If a report is slow even with little concurrent activity, investigate its model and queries before treating the issue as a capacity shortage. Use the capacity calculator to explore planning inputs, then set a measurable acceptance target and repeat the representative workload before release.