The new Fabric Capacity Metrics app preview: a 30-minute weekly capacity review

The Microsoft Fabric Capacity Metrics app now ships with a second, redesigned report in preview. It reorganises capacity health, compute, storage and item-level detail into a cleaner layout, and it answers a question the old report made hard: which of my capacities is in trouble right now?

The default report stays exactly as it was, so there is no migration to plan. That makes this an easy preview to try, and a good excuse to build a habit most Fabric teams do not have: a short, regular capacity review.

What changed

With version 3.0 of the app, the navigation pane shows two reports after you install or upgrade. The default report is still generally available and still opens first. The preview report sits beside it, and you can switch at any time. Microsoft documents the preview report as having four main pages: Health, Compute, Item details (reached by drill-through) and Storage, plus a page for Autoscale for Spark and a Help page with explanations of visuals, icons and thresholds.

Three things stand out to me.

The Health page gives you an estate view. You pick a region, and it shows the capacities you can access there. Summary cards count how many were active, how many never throttled, how many spent time in interactive delay, interactive rejection or background rejection, and how many had surge protection active. Below that, a table offers views for the latest 30-second window, the last hour, the last 24 hours and the last 14 days.

It shows time spent in each state. In the utilization views you see the percentage of time each capacity spent paused, not throttled, in interactive delay, in interactive rejection, in background rejection, or with surge protection active. That is a far better health signal than a single utilization peak.

It marks mitigations. Icons show where surge protection or capacity overage was active. If overage quietly prevented throttling, you can now see it instead of finding it on the invoice.

How to read it without fooling yourself

Microsoft's own documentation spends time on two points that trip people up, and they are worth repeating.

The first is smoothing. A smoothed view spreads a background operation's usage over the following 24 hours, which reflects how it affects the capacity. An unsmoothed view shows the usage when the operation reported it. If you are investigating throttling, use the smoothed views. If you want to know which items report the most compute, use the unsmoothed one. Mixing them up is an easy way to blame the wrong job.

The second is that utilization and throttling are not the same thing. According to the documentation, throttling is based on future usage reaching 100% over a window: 10 minutes for interactive delay, 60 minutes for interactive rejection and 24 hours for background rejection. A capacity can run above 100% without throttling, and it can throttle below 100% while it burns down carried-forward overage.

Also keep the data delay in mind. The Latest view has a latency of about 10 to 15 minutes, so it is a near-real-time view, not a live one.

A 30-minute weekly review

This is the routine I would try, based on the scenarios Microsoft lays out in its monitoring guide.

1. Open the Health page and the 14-day Utilization view. Sort by time spent in an unhealthy state.
2. Open the At risk view. It lists only capacities that spent time above 90% of the interactive delay threshold, so these are the ones that have not failed yet.
3. Note where surge protection or capacity overage was active. Those are mitigations doing their job, and each one is a conversation about sizing or scheduling.
4. For the worst capacity, drill through to the Compute page and open the smoothed 14-day view. Use the capacity health heat map to find the repeating pattern, such as a Monday morning refresh pile-up.
5. Select the problem period. Approx CU Consumption in Window shows which items contributed, and Impacted Items During Time Window shows which items were affected.
6. Drill through on the top item to the Item details page. You get up to 14 days of history, a breakdown by operation, status and user, and operation IDs.
7. Write down one action per capacity: optimise an item, move a workspace, adjust surge protection, change a schedule, or resize. Check the result next week.

What to keep in mind

• It is a preview. Microsoft marks it as such, so do not build your only monitoring on it yet. Keep the default report as the reference.
• The Health page covers one region at a time and only the capacities you can access. If you manage capacities in several regions, you will switch between them.
• Deleted capacities do not appear, and the percentages for states can add up to more than 100% because some states overlap.
• The Item details page keeps the time filters from the Compute page, so it may show less than 14 days at first. Clear the filters if you need the full period.
• The app does not send alerts. For notifications, Microsoft points to Fabric capacity overview events and capacity operation events instead.

What this means for you

If you run Fabric in production, the useful change is not a new chart. It is that the app now makes it quick to answer three questions: which capacities are unhealthy, what was happening when they were, and which item was responsible.

After two decades in data, I have seen that capacity problems rarely arrive as a surprise. They show up as a slow creep in the numbers weeks before someone complains. A weekly look is cheap compared with explaining a throttled board report.

My suggestion: install or upgrade the app, open the preview report next to the default one, and run the routine above once on your busiest capacity. If it tells you something you did not already know, make it a calendar item.

How often does someone on your team actually open the Capacity Metrics app?

Sources

• What is the Microsoft Fabric Capacity Metrics app? (Microsoft Learn)
• Monitor capacity and throttling in the Fabric Capacity Metrics app preview (Microsoft Learn)
• Understand the Health page in the preview report (Microsoft Learn)
• Understand the Compute page in the preview report (Microsoft Learn)
• Understand the Item details page in the preview report (Microsoft Learn)