V-Order tuning

What V-Order is

V-Order is a write-time optimization applied to Parquet files: sorting, row-group reordering, dictionary encoding, and compression tuned for the VertiPaq engine behind Power BI and the SQL analytics endpoint. It stays 100% Parquet-compliant — any Delta reader can still open the files.

CU impact

Write: +10–25% CU on the writing Spark job (extra sort + encode).
Read: −15–50% on Direct Lake and SQL-endpoint scans of the same data.

The break-even is read frequency. A table read hundreds of times a day by Power BI should keep V-Order on. A staging table that is written once and consumed by one nightly Spark job should have it off.

Control it at three levels

# 1. Session default (Fabric enables this by default)
spark.conf.set("spark.sql.parquet.vorder.default", "true")

# 2. Per write
(df.write
   .option("parquet.vorder.enabled", "true")
   .mode("overwrite")
   .saveAsTable("mart.sales_by_day"))

# 3. Per table — sticks for every future write
spark.sql("""
  ALTER TABLE staging.raw_events
  SET TBLPROPERTIES ('delta.parquet.vorder.enabled' = 'false')
""")

Recommended defaults

Table role	V-Order
Direct Lake semantic model source	On
SQL endpoint / warehouse-facing marts	On
Bronze / landing / staging	Off
Intermediate Spark-only silver tables	Off unless also queried by SQL
Large fact tables rebuilt nightly, read all day	On, and pair with `OPTIMIZE`

Retrofitting existing tables

Setting the table property does not rewrite existing files. Force it:

spark.sql("ALTER TABLE mart.sales_by_day SET TBLPROPERTIES ('delta.parquet.vorder.enabled' = 'true')")
spark.sql("OPTIMIZE mart.sales_by_day")   # rewrites files with V-Order applied

Runtime

V-Order is available on all current Fabric Runtimes. New workspaces created after the 2024 change ship with the session default disabled for generic Spark workloads — check spark.sql.parquet.vorder.default before assuming it is on.