Forum Discussion

Sanyukti_Jain's avatar
Sanyukti_Jain
Icon for Advocate II rankAdvocate II
3 months ago
Solved

Fabric Capacity Metrics

I'm trying to wrap my head around the Fabric Capacity Metrics app for monitoring performance. What are the key metrics you check first when troubleshooting a slow report or Dataset refresh?
data engineering
  • BalajiL's avatar
    BalajiL
    3 months ago

    Start by analyzing CU utilization in the Capacity Metrics app, focusing on the Timepoint Summary or Detail views. Identify which workspace, item, and operation are contributing to the spike. This will help you quickly narrow down the source and understand the reason behind increased CU consumption. Given below sample reference that indicates one of the data movement causing CU utilization. 

     

    For performance-related issues, you can also use the Monitoring (Monitor) app to check the execution duration of Fabric items such as pipelines, notebooks, or dataflows. This provides additional insights into where time is being spent during execution and helps pinpoint bottlenecks.

     

     

  • Murtaza_Ghafoor's avatar
    Murtaza_Ghafoor
    2 months ago

    Sanyukti_Jain 
    Hi,
    In fabric capacity metrics app, you can check the following KPI's

    1-CU(CapacityUnit) Consumption:
    2-Thortlling and delays
    3-Identify Top Consumers
    4-Refresh Duration and resource Usage

    You may export or download the report and analyze these metrics for perfromance related issues.
    Screen shot attached below for your reference or for complete detail, please refer to microsoft documentation link.
    https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app

     

    If this helps, ✓ Mark as Kudos | Help Others

GilbertQ's avatar
GilbertQ
Icon for Super User rankSuper User
3 months ago

Hi Sanyukti_Jain 

 

I always look at the interactive and background processes first to make sure that the combination of these are not over 100%, because if they're over 100% there would be some throttling happening. If it is over 100% you can then drill through to the time point detail to see what has caused that to go over 100%, which then allow you to troubleshoot further.