Direct Lake Fallback Checker

Check a Direct Lake table against the guardrails of each Fabric SKU and the common causes of DirectQuery fallback, before they slow down — or break — your reports.

Runs entirely in your browser. Nothing you enter is uploaded or stored.

REAL-WORLD FABRIC SCENARIOS:

Model & Capacity Parameters

Fact Table Row Count35.0 Million Rows
1MGuardrail: 300M450M
Direct Lake Fallback Triggers
DirectQuery Fallback Active
Framing Rejected (Falling back to SQL Endpoint)

Bypassed In-Memory Framing — DirectQuery Mode Triggered

Critical: Direct Lake acceleration is deactivated. All visual queries hit the SQL Analytics Endpoint via DirectQuery, causing high visual latency and rapid Fabric CU exhaustion.

EST. LATENCY
~6400 ms
⚠️ User perceptible lag
CU CONSUMPTION
12.5x
🔥 Severe capacity burn
SKU GUARDRAIL
12%
Limit: 300M rows (F16)

Architectural Findings & Remediation1 trigger identified

HARD FALLBACK
DAX Calculated Column in Direct Lake Table
DL_001_CALC_COL

Direct Lake maps Parquet columns directly into VertiPaq memory without processing. Any DAX calculated column forces the entire table or query to fall back to the SQL Analytics endpoint via DirectQuery.

Recommended Remediation:

Materialize the calculation upstream in Lakehouse Delta tables using PySpark or a Fabric Dataflow Gen2 / SQL pipeline.

# PySpark Lakehouse remediation (Delta Table)
df = df.withColumn("MarginPct", (df["Revenue"] - df["Cost"]) / df["Revenue"])
df.write.format("delta").mode("overwrite").saveAsTable("fact_sales")
Fabric Direct Lake Advisory

Migrate from Import to High-Velocity Direct Lake Without Capacity Throttling

I work with BI teams to re-engineer Lakehouse schemas, optimize Delta Parquet V-Order indexing, eliminate calculated columns upstream in Medallion architectures, and guarantee in-memory framing.

Request Architecture Audit

Privacy: the analysis runs in your browser and what you enter never leaves it. I only count that the tool was used — no inputs, results or personal data are recorded.

Get the next Engineering Note

Practical DAX, semantic model and Fabric notes. 1–2 a month, unsubscribe any time.

1–2 useful notes a month. Unsubscribe at any time.

Understanding Direct Lake vs DirectQuery Fallback

Direct Lake In-Memory Framing

Direct Lake loads Delta table columns from OneLake into memory when queries need them. A lightweight refresh (framing) points the model at the latest table versions instead of re-importing the data.

The DirectQuery Fallback Penalty

With Direct Lake on SQL endpoints, queries the engine cannot serve from memory — for example over SQL views or tables beyond the guardrails — fall back to DirectQuery if fallback is enabled. Results still return, but more slowly.

Fabric Capacity Guardrails

Each Fabric SKU caps rows and Parquet files per table (300 million rows on F2–F32, 1.5 billion on F64). Beyond them, SQL-endpoint models fall back to DirectQuery and OneLake models fail to refresh.

Frequently Asked Questions About Fabric Direct Lake

What causes a Power BI Direct Lake model to fall back to DirectQuery?

Fallback applies to Direct Lake on SQL endpoints. Common causes are tables exceeding the SKU's guardrails (rows, files or row groups per table), tables based on SQL views rather than Delta tables, and security defined on the SQL endpoint that the model must respect. Direct Lake on OneLake has no fallback: exceeding guardrails makes the refresh fail instead. The model's Direct Lake behavior setting (Automatic, DirectLakeOnly, DirectQueryOnly) controls whether fallback is allowed.

Can I use calculated columns in a Direct Lake model?

Calculated columns aren't supported on Direct Lake tables, because the data comes straight from the Delta files rather than from an import the engine could extend. Compute the column upstream in the Lakehouse or Warehouse instead, so it exists in the Delta table.

How do I fix a calculated column in Direct Lake?

Move the calculation upstream into your Fabric Lakehouse or Warehouse Medallion layer. Compute the column in PySpark (`df.withColumn(...)`) or SQL (`CREATE TABLE ... AS SELECT ...`) so it is physically written into the Delta Parquet storage.

What is V-Order and why does it matter for Direct Lake?

V-Order is a write-time optimisation that sorts and compresses Parquet files in a way the VertiPaq engine reads efficiently, which improves Direct Lake load and query performance. Regular OPTIMIZE (with V-Order) also compacts small files, keeping tables under the per-table file and row-group guardrails.

Evaluate Direct Lake readiness with an Engineering Health Check

My fixed-scope Health Check reviews OneLake delta table schemas, relationship designs, and capacity consumption to prevent unexpected DirectQuery fallbacks in production.