Skip to content
DP-700 Fabric Engineer Hub

Manage Capacity and Performance in Microsoft Fabric

Manage Capacity and Performance in Microsoft Fabric

Source

Summary

This module focuses on Fabric Capacity Units (CU), compute sizing, bursting, smoothing, throttling stages, and using the Microsoft Fabric Capacity Metrics App to diagnose over-consumption.

Capacity Management Mechanics

1. Capacity Units (CU) & Sizing

  • Capacities range from F2 (2 CUs) up to F2048 (2048 CUs).
  • 1 Capacity Unit equals approximately the compute power of 1 virtual core.

2. Bursting & Smoothing

  • Bursting: Fabric allows workloads (such as heavy Spark jobs or Warehouse queries) to temporarily consume more CUs than the base SKU without immediately throttling.
  • Smoothing:
    • Interactive Operations (e.g., live Power BI queries, notebook cells): Smoothed over a 5-minute moving average window.
    • Background Operations (e.g., scheduled Data Factory pipelines, Dataflow refreshes, Scheduled Spark batches): Smoothed over a 24-hour moving window.

3. Throttling Stages & Burndown

[CU Usage Exceeds 100%]
       │
       ▼ (Over-consumption accumulates in Burndown Debt)
┌────────────────────────────────────────────────────────┐
│  Phase 1: Interactive Delay (10 mins over limit)       │
│  - Interactive queries experience 2-5s artificial delay│
└──────────────────────────┬─────────────────────────────┘
                           │
                           ▼ (Burndown debt continues to accumulate)
┌────────────────────────────────────────────────────────┐
│  Phase 2: Interactive Rejection (60 mins over limit)   │
│  - New interactive queries rejected with HTTP 429      │
└──────────────────────────┬─────────────────────────────┘
                           │
                           ▼ (Persistent overload)
┌────────────────────────────────────────────────────────┐
│  Phase 3: Background Rejection (24 hours over limit)   │
│  - Scheduled pipelines and background batch jobs fail  │
└────────────────────────────────────────────────────────┘

4. Fabric Capacity Metrics App

  • Timepoint Drill-down: Pinpoints the exact minute and specific workspace item responsible for high CU spikes.
  • Top Consumers by Item Type: Breaks down compute burn across Lakehouses, Warehouses, Spark sessions, and Dataflows.

Exam Traps & Gotchas

[!CAUTION]

  • Throttling Hierarchy: Interactive jobs are throttled first (delay, then rejection), while background scheduled jobs are protected until sustained overload persists across the 24-hour smoothing window.
  • Autoscale Option: Azure Fabric capacities support autoscale to temporarily add burst CUs during predictable monthly ETL peaks.