Skip to content

Speeding up a Lakehouse table with maintenance operations

Optimizing a Lakehouse table in Microsoft Fabric primarily involves compacting small Parquet files into larger ones and applying Delta Lake write optimizations to improve query performance and reduce storage/compute overhead. Fabric provides both automatic maintenance settings and manual commands (OPTIMIZE, VACUUM, Z-Order) to keep Delta tables efficient over time.

Must-know

  • V-Order is a Fabric-specific write-time optimization applied to Parquet files that improves read performance for Power BI, SQL, and Spark engines by sorting and encoding data for faster compression and retrieval, and it is enabled by default for Spark writes in Fabric.
  • The OPTIMIZE command compacts many small files into fewer, larger files to reduce metadata overhead and improve read throughput, and can be combined with ZORDER BY on frequently filtered columns to colocate related data.
  • VACUUM removes old, unreferenced data files left behind by updates/deletes/compaction after the table's retention period (default 7 days), reclaiming storage but permanently removing time-travel history beyond that window.
  • Table maintenance can be configured to run automatically (scheduled or triggered) in Fabric Lakehouse settings, or run manually via Spark notebooks/SQL, giving flexibility between hands-off and on-demand optimization.
  • Small file problems typically arise from frequent, small streaming or incremental writes; increasing batch size or scheduling periodic OPTIMIZE jobs mitigates this and reduces job planning/read latency.
  • Partitioning strategy matters: over-partitioning a table (e.g., by high-cardinality columns) can worsen performance by creating excessive small files, so partition columns should be chosen based on common filter predicates and reasonable cardinality.
Check this objectiveFree · always available

A data engineer ingests IoT telemetry into a Fabric Lakehouse Delta table through frequent small streaming micro-batches. After several months, query performance has degraded because the table now consists of thousands of very small Parquet files. Without changing the ingestion pattern, which action should the engineer take first to resolve this?

Your objective map0 tried · 0 right · 54 untouched

What you have tried across DP-700's objectives, not a readiness score.

Coverage checked against the published exam guide on Aug 12, 2026.

These are independent practice questions, written against this certification's published exam guide. They are not the certification vendor's own questions, and not the real exam.