Quick Answer
The Fabric Capacity Metrics app is the single most important tool for operating a Fabric capacity. It tells you who is consuming CU, when throttling occurs, whether smoothing is saving you, and which operations need optimization. Install it day one on every capacity. Review weekly. Act on the insights monthly.
1. App Pages Overview
- Summary: high-level KPIs including CU consumption, throttling events, and top operations.
- Compute: detailed operation-level CU consumption. Sort by CU to find top consumers.
- Storage: data stored in OneLake, semantic models, and other workloads.
- Throttling: any throttling events, their duration, and affected operations.
- Items: per-item (dataset, notebook, pipeline) resource consumption.
- Timepoint explore: drill-down into a specific 30-second window showing exactly what ran during a high-CU period.
2. Understanding CU Smoothing
A capacity at F64 provides 64 CU per second, which equals 5.5 million CU per 24-hour day. Smoothing averages consumption over this 24-hour window. If you consume 10 million CU in one hour but 0 CU in other hours and your 24-hour total stays below 5.5 million, no throttling occurs.
In practice, this lets you run heavy scheduled refreshes and batch jobs during off-hours and absorb burst loads during peak hours without a larger capacity. To optimize, identify underutilized hours (typically 10pm to 6am) and reschedule heavy operations into those windows.
The Metrics app shows smoothed vs instantaneous CU so you can see how much headroom smoothing is giving you. If smoothing is saving you from throttling on 30 percent of days, your capacity is right-sized. If smoothing is barely keeping you within limits, plan for a capacity upgrade.
3. Interactive vs Background Operations
The Metrics app separates operations into Interactive and Background categories. This distinction matters because Fabric throttles them differently.
- Interactive (user-facing queries): protected from throttling longest. Users see slowdowns before outright rejection.
- Background (refreshes, pipelines, notebooks): throttled first. Scheduled refreshes fail quickly under pressure.
If background operations are being throttled regularly, move them to off-peak windows or split them into smaller chunks. If interactive operations are being throttled, users are experiencing degraded performance and a capacity upgrade is overdue.
4. Diagnosing Throttling
When the Throttling page shows events, take these steps in order:
- Identify the time window of throttling.
- Use Timepoint Explore to view the top operations running in that window.
- Identify whether throttling was triggered by a single heavy operation or by cumulative load.
- For single-heavy-operation: optimize the DAX, partition the dataset, or move the operation off-peak.
- For cumulative load: reschedule background operations or consider capacity scale-up.
5. Optimization Playbook
Week 1: Install and baseline
Install the app. Let it collect 7 days of telemetry. Record baseline CU consumption, peak utilization, and smoothing utilization.
Week 2: Identify top 10 consumers
List the 10 operations consuming most CU. Classify them: optimize-worthwhile, reschedule-worthwhile, or acceptable-as-is.
Week 3-4: Implement fixes
For optimize-worthwhile operations, open the dataset in DAX Studio and profile the query. Implement the fix (index, measure rewrite, column removal) and redeploy.
Month 2: Measure impact
Compare post-fix CU consumption to baseline. Most organizations reduce overall CU consumption 20 to 40 percent in the first month of systematic optimization. This often enables downsizing to a smaller SKU and significant cost savings.
For the full cost optimization framework, see the Fabric capacity cost optimization playbook.
Frequently Asked Questions
What is the Fabric Capacity Metrics app?
The Fabric Capacity Metrics app is a free Power BI app published by Microsoft that provides detailed visibility into capacity usage, performance, and cost. It shows CU consumption by operation, by user, and by workload, as well as smoothing and throttling behavior. Every organization running Fabric or Power BI Premium capacity should install it and review it weekly. It is the primary tool for capacity planning and performance diagnostics.
How do I install the Capacity Metrics app?
Search for "Microsoft Fabric Capacity Metrics" in the Power BI app store, click Get it now, and connect it to your capacity. You will need to provide the capacity ID or name during setup. Only capacity administrators can install and view the app. The app refreshes automatically every few hours with telemetry from the capacity.
What is CU smoothing?
Smoothing is a Fabric feature that averages CU consumption over a 24-hour window. Burst workloads that briefly exceed 100 percent utilization are smoothed against underutilized hours. As long as 24-hour average stays under 100 percent, no throttling occurs. Smoothing effectively lets you handle short bursts up to 10x your capacity ceiling without paying for larger capacity, as long as you have downtime elsewhere in the day. This is one of the most important Fabric features for cost optimization.
What is throttling and how do I avoid it?
Throttling occurs when sustained CU consumption exceeds capacity limits even after smoothing. Three throttling levels exist: delay (interactive operations slow down), reject (new operations return 429 errors), and background reject (scheduled refreshes are rejected). Avoid throttling by right-sizing capacity, scheduling heavy operations during off-peak hours, and removing CU-hungry queries. The Capacity Metrics app shows throttling events in the Throttling page.
What is the difference between interactive and background operations?
Interactive operations are user-facing queries that execute in response to visual rendering, DAX queries from XMLA, or Copilot requests. They have priority and are throttled last. Background operations are scheduled refreshes, deployment pipeline executions, and Fabric pipeline runs. They are throttled first under capacity pressure. The Metrics app separates these categories, letting you identify whether throttling affects user experience (interactive) or data freshness (background).
How do I identify the top CU-consuming operations?
The Compute page of the Capacity Metrics app lists operations by CU consumption. Sort descending to find the biggest consumers. Look for outliers: a single operation consuming 20 percent of a day's CU is almost always a DAX query that can be optimized or a scheduled refresh processing too much data. Capture the operation text, open DAX Studio, and optimize the query or restructure the refresh to reduce CU consumption.
How often should I review capacity metrics?
Weekly at minimum for active operations, monthly for strategic planning. Weekly reviews catch emerging problems: a new dataset consuming unexpected CU, a user running ad-hoc queries that throttle the capacity. Monthly reviews identify trends: capacity utilization creeping up as data volumes grow, reservation renewal timing, or right-sizing opportunities. Many mature organizations build a simple scorecard from the Metrics app and share it with the data team.
Can the Capacity Metrics app trigger alerts?
Yes, via Power BI data alerts or Microsoft Purview audit events. Configure data alerts on the Throttling KPI to notify administrators via email when throttling occurs. For more sophisticated alerting, export capacity metrics data to Azure Log Analytics or a SIEM, and build custom rules for threshold crossings, sustained high utilization, or operation-level anomalies.
Fabric Capacity Running Hot?
Our consultants diagnose capacity utilization, identify CU hogs, and optimize workloads to reduce throttling and cost. Contact us for a capacity review.