This post is a part of the DP-600: Implementing Analytics Solutions Using Microsoft Fabric Exam Prep Hub; and this topic falls under these sections:
Implement and manage semantic models (25-30%)
--> Optimize enterprise-scale semantic models
--> Configure Direct Lake, including default fallback and refresh behavior
Overview
Direct Lake is a storage and connectivity mode in Microsoft Fabric semantic models that enables Power BI to query data directly from OneLake without importing data into VertiPaq or sending queries back to the data source (as in DirectQuery). It is designed to deliver near–Import performance with DirectQuery-like freshness, making it a key feature for enterprise-scale analytics.
For the DP-600 exam, you are expected to understand:
- How Direct Lake works
- When and why fallback occurs
- How default fallback behavior is configured
- How refresh behaves in Direct Lake models
- Common performance and design considerations
How Direct Lake Works
In Direct Lake mode:
- Data resides in Delta tables stored in OneLake (typically from a Lakehouse or Warehouse).
- The semantic model reads Parquet/Delta files directly, bypassing data import.
- Metadata and file statistics are cached to optimize query performance.
- Queries are executed without duplicating data into VertiPaq storage.
This architecture reduces data duplication while still enabling fast, interactive analytics.
Default Fallback Behavior
What Is Direct Lake Fallback?
Fallback occurs when a query or operation cannot be executed using Direct Lake. In these cases, the semantic model automatically falls back to another mode to ensure the query still returns results.
Depending on configuration, fallback may occur to:
- DirectQuery, or
- Import (VertiPaq), if data is available
Fallback is automatic and transparent to report users unless explicitly restricted.
Common Causes of Fallback
Direct Lake fallback can be triggered by:
- Unsupported DAX functions or expressions
- Unsupported data types in Delta tables
- Complex model features (certain calculation patterns, security scenarios)
- Queries that cannot be resolved efficiently using file-based access
- Temporary unavailability of OneLake files
Understanding these triggers is important for diagnosing performance issues.
Configuring Default Fallback Behavior
In Fabric semantic model settings, you can configure:
- Allow fallback (default) – Ensures queries continue to work even when Direct Lake is not supported.
- Disable fallback – Queries fail instead of falling back, which is useful for enforcing performance expectations or testing Direct Lake compatibility.
From an exam perspective:
- Allowing fallback prioritizes reliability
- Disabling fallback prioritizes predictability and performance validation
Refresh Behavior in Direct Lake Models
Do Direct Lake Models Require Refresh?
Unlike Import mode:
- Direct Lake does not require scheduled data refresh to reflect new data in OneLake.
- New or updated Delta files are automatically visible to the semantic model.
However, metadata refreshes are still relevant.
Types of Refresh in Direct Lake
- Metadata Refresh
- Updates table schemas, partitions, and statistics
- Required when:
- Columns are added or removed
- Table structures change
- Lightweight compared to Import refresh
- Hybrid Scenarios
- If fallback to Import is enabled and used, those imported parts do require refresh
- Mixed behavior may exist in composite or fallback-heavy models
Impact of Refresh on Performance
- No large-scale data movement during refresh
- Faster model readiness after schema changes
- Reduced refresh windows compared to Import models
- Lower memory pressure in capacity
This makes Direct Lake especially suitable for large, frequently updated datasets.
Performance and Design Considerations
To optimize Direct Lake usage:
- Use supported Delta table features and data types
- Keep models simple and star-schema based
- Avoid unnecessary bidirectional relationships
- Monitor fallback behavior using performance tools
- Test critical DAX measures for Direct Lake compatibility
From an exam standpoint, expect scenario-based questions asking you to choose Direct Lake and configure fallback appropriately for scale, freshness, and reliability.
When to Use Direct Lake
Direct Lake is best suited for:
- Large datasets stored in OneLake
- Near-real-time analytics
- Enterprise models that need both performance and freshness
- Organizations standardizing on Fabric Lakehouse or Warehouse architectures
Key DP-600 Takeaways
- Direct Lake queries Delta tables directly in OneLake
- Default fallback ensures query continuity when Direct Lake isn’t supported
- Fallback behavior can be enabled or disabled
- Data refresh is not required, but metadata refresh still matters
- Understanding fallback and refresh behavior is critical for enterprise-scale optimization
DP-600 Exam Tip 💡
Expect scenario-based questions where you must decide:
- Whether to enable or disable fallback
- How refresh behaves after schema changes
- Why a query is falling back unexpectedly
Practice Questions:
Here are 10 questions to test and help solidify your learning and knowledge. As you review these and other questions in your preparation, make sure to …
- Identifying and understand why an option is correct (or incorrect) — not just which one
- Look for and understand the usage scenario of keywords in exam questions to guide you
- Expect scenario-based questions rather than direct definitions
1. What is the primary benefit of using Direct Lake mode in a Fabric semantic model?
A. It fully imports data into VertiPaq for maximum compression
B. It queries Delta tables in OneLake directly without data import
C. It sends all queries back to the source system
D. It eliminates the need for semantic models
Correct Answer: B
Explanation:
Direct Lake reads Delta/Parquet files directly from OneLake, avoiding both data import (Import mode) and source query execution (DirectQuery), enabling near-Import performance with fresher data.
2. When does a Direct Lake semantic model fall back to another query mode?
A. When scheduled refresh fails
B. When unsupported features or queries are encountered
C. When the dataset exceeds 1 GB
D. When row-level security is enabled
Correct Answer: B
Explanation:
Fallback occurs when a query or model feature is not supported by Direct Lake, such as certain DAX expressions or unsupported data types.
3. What is the default behavior of Direct Lake when a query cannot be executed in Direct Lake mode?
A. The query fails immediately
B. The query retries using Import mode only
C. The query automatically falls back to another supported mode
D. The semantic model is disabled
Correct Answer: C
Explanation:
By default, Direct Lake allows fallback to ensure query reliability. This allows reports to continue functioning even if Direct Lake cannot handle a specific request.
4. Why might an organization choose to disable fallback in a Direct Lake semantic model?
A. To reduce OneLake storage costs
B. To enforce consistent Direct Lake performance and detect incompatibilities
C. To allow automatic data imports
D. To improve data refresh frequency
Correct Answer: B
Explanation:
Disabling fallback ensures queries only run in Direct Lake mode. This is useful for performance validation and preventing unexpected query behavior.
5. Which action typically requires a metadata refresh in a Direct Lake semantic model?
A. Adding new rows to a Delta table
B. Updating existing fact table values
C. Adding a new column to a Delta table
D. Running a Power BI report
Correct Answer: C
Explanation:
Schema changes such as adding or removing columns require a metadata refresh so the semantic model can recognize structural changes.
6. How does Direct Lake handle new data written to Delta tables in OneLake?
A. Data is visible only after a scheduled refresh
B. Data is visible automatically without data refresh
C. Data is visible only after manual import
D. Data is cached permanently
Correct Answer: B
Explanation:
Direct Lake reads data directly from OneLake, so new or updated data becomes available without needing a traditional Import refresh.
7. Which scenario is MOST likely to cause Direct Lake fallback?
A. Simple SUM aggregation on a fact table
B. Querying a supported Delta table
C. Using unsupported DAX functions in a measure
D. Filtering data using slicers
Correct Answer: C
Explanation:
Certain complex or unsupported DAX functions can force fallback because Direct Lake cannot execute them efficiently using file-based access.
8. What happens if fallback is disabled and a query cannot be executed in Direct Lake mode?
A. The query automatically switches to DirectQuery
B. The query fails and returns an error
C. The semantic model imports the data
D. The model switches to Import mode permanently
Correct Answer: B
Explanation:
When fallback is disabled, unsupported queries fail instead of switching modes, making incompatibilities more visible during testing.
9. Which statement about refresh behavior in Direct Lake models is TRUE?
A. Full data refresh is always required
B. Direct Lake models do not support refresh
C. Only metadata refresh may be required
D. Refresh behaves the same as Import mode
Correct Answer: C
Explanation:
Direct Lake does not require full data refreshes because it reads data directly from OneLake. Metadata refresh is needed only for structural changes.
10. Why is Direct Lake well suited for enterprise-scale semantic models?
A. It eliminates the need for Delta tables
B. It supports unlimited bidirectional relationships
C. It combines near-Import performance with fresh data access
D. It forces all data into memory
Correct Answer: C
Explanation:
Direct Lake offers high performance without importing data, making it ideal for large datasets that require frequent updates and scalable analytics.
The Data Community 
One thought on “Configure Direct Lake, including default fallback and refresh behavior”