Explore Fabric capacity overview events in Fabric Real-Time hub

Fabric Capacity Overview Events provide summary level information related to your capacity. These events can be used to create alerts related to your capacity health via Data Activator or can be stored in an Eventhouse for granular or historical analysis.

View Fabric capacity overview events detail page

1. In Real-Time hub, select Fabric events.
2. Select Fabric capacity overview events from the list.
3. You should see the detailed view for Fabric capacity overview events.

Actions

At the top of the detail page, you see the following two actions.

• Create eventstream - lets you create an eventstream based on events from the selected Fabric capacity.
• Set alert - lets you set an alert when an operation is done for a Fabric capacity, such as a state change.

See what's using this category

This section shows the artifacts using Fabric capacity overview events. Here are the columns and their descriptions shown in the list.

Event types

Here are the supported Capacity overview events:

Schemas

An event has the following top-level data:

The data object has the following properties for Summary events:

Summary Events Schema

The data object has the following properties for State events:

State Events Schema

Considerations and Patterns

• When a capacity is paused, all smoothed usage is pushed into the next available window. As a result, large spikes in capacity utilization can appear for the window in which the pause occurs.

• Utilization percentage can't be calculated for P SKUs when autoscale is enabled because autoscale adds extra capacity units that aren't included in the states table.

• This system is a real-time system and doesn't backfill historical data. Even if you're using Real-Time Hub for alerting, start pushing data to Eventhouse or OneLake early, to ensure you have enough data for analysis such as graphing.

• Guidance on scenarios you might encounter:

  In order to ensure low latency and high performance, Capacity Events in Real-Time Hub are based on a best effort delivery mode. For the Summary table this behavior can mean that, while rare, events can either fail to be sent, or duplicates might be received.

  As follows are some patterns that can be employed for each of the scenarios mentioned.

  • Duplicate events

  Duplicates can be corrected through filtering, as there should only be one event per 30-second window per capacity. This step allows easy identification of duplicate values.

  In Kusto Query Language, you can use the following clause to remove duplicate data from the Summary table:

  | summarize take_any(*) by windowStartTime, windowEndTime, capacityId
  

  take_any is preferred to distinct because there's a possibility there might be small differences between duplicate data. In this case, a distinct operation would keep both. Here, take_any ensures exactly one row is taken.

  • Missing events

  Missing data is far rarer than duplicate data but might still occur. This condition doesn't have a noticeable impact on alerting or most capacity analysis, but might impact graphing/ analysis at the granularity of window start time. A useful pattern to handle dropped data is sampling the maximum value of a column over a fixed time window, such as five minutes. This approach helps ignore dropped data while still analyzing your data at a detailed level.

  • Paused capacities

  When a capacity is paused, all smoothed usage is pushed into the next available window. This behavior can cause large spikes in CU appearing for the window in which the pause occurs (up to 288,000% of normal if a capacity is busy when paused).

  For some graphing you might want to exclude paused windows, which impacts your X axis. The best way to do this is to calculate average utilization and exclude any values greater than ~500%. This approach should allow you to retain data related to normal overages, but exclude pause events. In Kusto Query Language, one way to achieve this approach is as follows:

  ['_summaryTable']
  | extend capacityUnitMsBudget = baseCapacityUnits * 1000 * 30
  | extend UtilizationPct = todecimal(capacityUnitMs)/ capacityUnitMsBudget * 100
  | project windowStartTime, UtilizationPct | where UtilizationPct < 500
  | render timechart
  

Subscribe permission

For more information, see subscribe permission for Fabric events