⚡ Quick Reference Cheatsheet

  • Based on: Microsoft Fabric documentation (Microsoft Learn)
  • 📁 ← Back to Home

Last-minute review sheet for the DP-600 Microsoft Fabric Analytics Engineer Associate exam. Print it, bookmark it, or skim it in the parking lot.

Table of contents

🔢 1 — Key Numbers & Limits

Item Value / Threshold
Max rows per table (Direct Lake, F2–F32) 300 million
Max rows per table (Direct Lake, F64 / P1) 1.5 billion
Max Parquet files per table (Direct Lake, F2 / F64) 1,000 / 5,000
Max row groups per table (Direct Lake, F2 / F64) 1,000 / 5,000
Max model size on OneLake (Direct Lake, F2 / F16 / F64) 10 GB / 20 GB / Unlimited
XMLA read/write endpoint requirement Premium / Fabric F64+ capacity (read available at lower SKUs)
Deployment pipeline stages (max) 10 (Dev, Test, Prod, plus up to 7 custom stages)
Workspace roles 4: Admin, Member, Contributor, Viewer
Sensitivity labels — who can apply Users with Information Protection license + label publishing policy
Incremental refresh — max partitions (Import) 10,000 per table
Max measures in a single DAX query Depends on memory; no hard cap but keep below 100 for performance
OneLake file format for Direct Lake Delta / Parquet (V-Order optimized)
Max scheduled refreshes per day (Pro) 8
Max scheduled refreshes per day (Premium) 48

Exam Tip: SKU-specific guardrails (row limits, table counts) are the most common “trick number” questions. Know at least the F2 and F64 tiers.

💾 2 — Storage Mode Decision Matrix

Criteria Import DirectQuery Direct Lake Composite
Data freshness Scheduled / on-demand refresh Real-time Near real-time (framing) Mixed
Query performance Fastest (in-memory) Depends on source Fast (columnar from OneLake) Varies by table
Data volume Limited by capacity memory No size limit Guardrails per SKU Mixed
Data source Any supported Any supported Delta tables in OneLake only Any mix
DAX support Full Full (slower) Full (with fallback) Full
Use case Small-to-mid datasets, max perf Enforce source RLS, large data Fabric-native analytics Migration, mixed sources
Refresh needed? Yes No Framing only Partial 
flowchart TD
    A[Choose Storage Mode] --> B{Data in OneLake<br/>as Delta tables?}
    B -- Yes --> C{Need real-time<br/>to the second?}
    C -- No --> D["**Direct Lake**<br/>Best of both worlds"]
    C -- Yes --> E["**DirectQuery**<br/>to Lakehouse SQL endpoint"]
    B -- No --> F{Data source<br/>supports DirectQuery?}
    F -- No --> G["**Import**<br/>Schedule refresh"]
    F -- Yes --> H{Need max<br/>performance?}
    H -- Yes --> G
    H -- No --> I["**DirectQuery**<br/>or Composite"]

Exam Caveat: Direct Lake is only available for Delta tables stored in OneLake (Lakehouse or Warehouse). If data is in Azure SQL or Synapse, you cannot use Direct Lake.

📐 3 — DAX Quick Reference

🔧 CALCULATE & Filter Modifiers

Pattern What It Does Example
CALCULATE(expr, filter) Evaluates expression in modified filter context CALCULATE(SUM(Sales[Amount]), Product[Color]="Red")
ALL(table/col) Removes all filters from table or column CALCULATE(SUM(Sales[Amount]), ALL(Product))
ALLEXCEPT(table, col) Removes all filters except specified columns CALCULATE(…, ALLEXCEPT(Product, Product[Category]))
REMOVEFILTERS(col) Alias for ALL — clearer intent CALCULATE(…, REMOVEFILTERS(Date[Year]))
KEEPFILTERS(filter) Adds filter without overriding existing context CALCULATE(…, KEEPFILTERS(Product[Color]="Red"))

🔄 Iterators

Function Purpose
SUMX(table, expr) Row-by-row sum
AVERAGEX(table, expr) Row-by-row average
COUNTX(table, expr) Count rows where expr is not blank
RANKX(table, expr) Rank each row by expression
MAXX / MINX Row-by-row max/min

📊 Table Functions

Function Purpose
FILTER(table, condition) Returns filtered table (iterator — slower)
ADDCOLUMNS(table, name, expr) Adds calculated columns to a table expression
SUMMARIZE(table, groupCol) Groups by column(s) — avoid adding measures here
SUMMARIZECOLUMNS(col, filter, name, expr) Preferred for grouping + measures (optimized)
SELECTCOLUMNS(table, name, expr) Projects specific columns
DISTINCT(col) Unique values of a column
VALUES(col) Unique values including blank row from broken relationships

⏰ Time Intelligence

Function Purpose
TOTALYTD(expr, dateCol) Year-to-date total
TOTALQTD / TOTALMTD Quarter / month-to-date
SAMEPERIODLASTYEAR(dateCol) Shifts dates back one year
DATEADD(dateCol, -1, YEAR) Shifts dates by interval
DATESYTD(dateCol) Returns dates from start of year to current date
PARALLELPERIOD(dateCol, -1, QUARTER) Full parallel period

Exam Caveat: Time intelligence functions require a contiguous Date table marked as a date table. Gaps in dates break these functions.

🪟 Windowing Functions (New in Fabric)

Function Purpose
OFFSET(n, relation, orderBy) Access row N positions away
WINDOW(from, to, relation, orderBy) Define a sliding window
INDEX(n, relation, orderBy) Access the Nth row

📌 Variables Pattern & USERELATIONSHIP 

-- VAR / RETURN pattern (always prefer for readability & performance)
Profit Margin =
VAR TotalRevenue = SUM(Sales[Revenue])
VAR TotalCost   = SUM(Sales[Cost])
RETURN
    DIVIDE(TotalRevenue - TotalCost, TotalRevenue, 0)

-- USERELATIONSHIP for role-playing dimensions
Ship Date Sales =
CALCULATE(
    SUM(Sales[Amount]),
    USERELATIONSHIP(Sales[ShipDateKey], Date[DateKey])
)

Exam Tip: USERELATIONSHIP activates an inactive relationship for that measure only. The relationship must already exist in the model as inactive.

🔒 4 — Security Layers Cheatsheet

Layer RLS (Row-Level Security) CLS (Column-Level Security) OLS (Object-Level Security)
What it restricts Rows visible to user Columns visible to user Tables or columns hidden from user
Defined in Semantic model (DAX filter) Semantic model (role config) Semantic model (role config)
Implementation DAX expression on table Column permissions in role None or Read metadata permission
Tooling Power BI Desktop / XMLA / Tabular Editor XMLA / Tabular Editor only XMLA / Tabular Editor only
Works with Direct Lake? ✅ ✅ ✅
Test method “View as Role” in service XMLA query as role XMLA query as role 
flowchart LR
    subgraph "Security Enforcement Stack"
        A["Workspace Roles<br/>(Admin, Member,<br/>Contributor, Viewer)"] --> B["Semantic Model<br/>RLS / CLS / OLS"]
        B --> C["Sensitivity Labels<br/>(Microsoft Purview)"]
    end
    style A fill:#1a3a5c,stroke:#4a9eff
    style B fill:#1a3a5c,stroke:#4a9eff
    style C fill:#1a3a5c,stroke:#4a9eff

Exam Caveat: RLS is defined in the semantic model, NOT in the Lakehouse or Warehouse. Users with Admin, Member, or Contributor roles (all have model edit access) bypass RLS when browsing data in the workspace — only Viewers (and read/build-only consumers) are subject to it.

🗄️ 5 — Data Store Selection

Criteria Lakehouse Warehouse KQL Database Eventhouse
Query language Spark SQL, PySpark T-SQL KQL (Kusto) KQL
File format Delta (Parquet) in OneLake Delta (Parquet) in OneLake (T-SQL–managed) Columnar time-series Columnar time-series
Best for Data engineering, ML, ELT Enterprise BI, complex joins, stored procs Log/telemetry analytics Streaming + real-time analytics
Schema Schema-on-read & schema-on-write Schema-on-write (strict) Semi-structured, schemaless Semi-structured
Transactions ACID via Delta Full T-SQL transactions Append-optimized Append-optimized
SQL endpoint? ✅ (read-only auto-generated) Native ❌ (KQL only) ❌ (KQL only)
Direct Lake support ✅ ✅ ❌ ❌

Exam Tip: If the question mentions T-SQL stored procedures or views with complex joins, the answer is Warehouse. If it mentions streaming or IoT telemetry, think Eventhouse / KQL Database.

📥 6 — Ingestion Method Selection

Criteria Dataflow Gen2 Notebook (Spark) Copy Activity (Pipeline) Shortcut
Skill level Low-code (Power Query) Pro-code (Python/Scala) Low-code (config) No-code
Transformation ✅ (M / Power Query) ✅ (full Spark) Minimal (mapping only) None (pass-through)
Best for Simple transforms, small data Complex ETL, ML prep, large data Bulk copy between stores Virtualized access, zero-copy
Scheduling Pipeline or standalone Pipeline Pipeline Always live
Compute Fabric Dataflow compute Spark pool Pipeline IR None
Writes Delta? ✅ (to Lakehouse) ✅ ✅ ❌ (reads in-place)

Exam Caveat: Shortcuts do not copy data — they create a virtualized reference. Data stays in the source (ADLS, S3, or another OneLake location). This is critical for questions about data residency and latency.

🔷 7 — Direct Lake Deep Dive

📌 Key Facts

Fact Detail
Required format Delta tables with V-Order optimization
Where data lives OneLake (Lakehouse or Warehouse)
Framing Snapshot of Delta log metadata; determines which Parquet files to read
Refresh type Framing (lightweight metadata update, not data copy)
Automatic updates Auto-reframe when Fabric detects Delta data/schema changes → near-real-time with no schedule
On-demand reframe Via XMLA endpoint, Fabric refresh API, or pipeline (when auto-updates is off)
Fallback behavior Direct Lake on SQL only — falls back to DirectQuery via SQL endpoint. Direct Lake on OneLake never falls back (query/refresh errors instead)
Fallback triggers (SQL flavour) SQL view, SQL-based RLS/CLS, guardrails exceeded, unsupported scenario
Disable fallback DirectLakeBehavior = DirectLakeOnly (query errors instead of going to DQ)
V-Order Special sort order applied at write time; improves read performance dramatically 
flowchart TD
    Q["DAX Query Arrives"] --> FLV{Which Direct<br/>Lake flavour?}
    FLV -- "on OneLake" --> G1{Within SKU<br/>guardrails?}
    G1 -- Yes --> DL["Direct Lake<br/>(read Parquet via memory)"]
    G1 -- No --> ERR["Query / refresh ERROR<br/>(no fallback)"]
    FLV -- "on SQL" --> G2{Servable<br/>in-memory?}
    G2 -- Yes --> DL
    G2 -- No --> FB{Fallback<br/>enabled?}
    FB -- Yes --> DQ["DirectQuery<br/>(SQL endpoint)"]
    FB -- No --> ERR
    DL --> R["Return Results"]
    DQ --> R

🔀 On OneLake vs On SQL Analytics Endpoint

Aspect Direct Lake on OneLake Direct Lake on SQL
Connector Azure Data Lake Storage (OneLake) SQL Server / OneLake.SqlAnalytics
Data sources per model Multiple Fabric items (many lakehouses/warehouses) Single lakehouse/warehouse
DirectQuery fallback ❌ Never (errors instead) ✅ Yes (unless disabled)
Composite with Import/DQ ✅ Yes ❌ No (only via a new composite in Desktop)
SQL-endpoint RLS/CLS/OLS ❌ Not applied (uses OneLake security) ✅ Enforced (delegated identity)
SQL views as tables ❌ (use materialized view / Import) ✅ but falls back to DirectQuery
Created from Service (OneLake catalog), Desktop, SQL endpoint page SQL analytics endpoint page only
Guardrail exceeded Refresh fails; not queryable Refresh warns; queries fall back to DQ

Exam Caveat: Direct Lake does not import data into the model — it reads Parquet from OneLake into memory on demand, and a “refresh” is just framing (metadata pointer). The on OneLake vs on SQL distinction (fallback, multi-source, composite, security) is the new July 2026 exam objective.

⚠️ 8 — Common Exam Traps

  1. RLS is defined in the semantic model, not in the Lakehouse, Warehouse, or data source.
  2. Direct Lake requires Delta tables in OneLake — no CSV, no Azure SQL, no external Parquet.
  3. Viewers cannot build reports on shared semantic models without Build permission.
  4. Workspace Admin, Member, and Contributor roles bypass RLS when browsing data directly in the workspace (they have model edit access); only Viewers and read/build-only consumers are filtered by RLS.
  5. XMLA read/write requires at minimum Power BI Premium Per User (PPU) or Fabric F64+ for write; read is available at lower SKUs.
  6. Deployment pipelines compare content between stages — they do not version-control code (use Git integration for that).
  7. Incremental refresh in Import mode creates partitions automatically — you configure the RangeStart and RangeEnd parameters, not partition logic.
  8. Sensitivity labels require Microsoft Purview Information Protection; they propagate downstream when data is exported.
  9. SUMMARIZE should not be used to add new measure columns — use SUMMARIZECOLUMNS or ADDCOLUMNS(SUMMARIZE(...)) instead.
  10. Composite models can mix Import and DirectQuery tables. Direct Lake on OneLake tables can now be mixed with Import tables in one model (web modeling / Desktop live edit); Direct Lake on SQL cannot — you’d build a new composite on top of it in Desktop.
  11. Shortcuts are read-only virtual references — they do not move, copy, or transform data.
  12. CLS and OLS cannot be configured in Power BI Desktop — they require XMLA endpoint or Tabular Editor.
  13. Time intelligence functions require a Date table marked as a date table with no gaps.
  14. V-Order is a write-time optimization — it must be applied when data is written to Delta, not at query time.
  15. KEEPFILTERS does not remove context; it intersects with existing filters. Contrast with CALCULATE which overrides.
  16. Direct Lake fallback to DirectQuery happens only with Direct Lake on SQL (via the SQL analytics endpoint). Direct Lake on OneLake never falls back — over-guardrail/unsupported queries error and refresh fails instead.
  17. Git integration in Fabric works at the workspace level, not at individual item level.
  18. Dataflow Gen2 outputs to Lakehouse/Warehouse; it does not load directly into a semantic model like legacy Power BI dataflows could.
  19. USERELATIONSHIP only works with inactive relationships — you cannot reference a relationship that does not exist.
  20. Large format datasets (over 10 GB) require Premium or Fabric capacity — Pro licenses are capped.

Exam Tip: When two answers both seem correct, ask yourself: “Where is this configured?” The exam loves testing whether something is set in the semantic model, the workspace, the data source, or the admin portal.

✅ 9 — Pre-Exam Checklist

🔒 Domain 1 — Maintain a Data Analytics Solution (25-30%)

  • I can explain the four workspace roles and their permissions
  • I know how deployment pipelines compare content across stages
  • I understand Git integration scope (workspace-level, supported item types)
  • I can configure sensitivity labels and explain downstream inheritance
  • I know when to use XMLA endpoints and which SKUs support read vs. read/write
  • I can describe the difference between RLS, CLS, and OLS and where each is configured
  • I understand how workspace Admin/Member roles interact with RLS

🔄 Domain 2 — Prepare Data (45-50%)

  • I can choose between Lakehouse, Warehouse, KQL Database, and Eventhouse
  • I know when to use Dataflow Gen2 vs. Notebook vs. Pipeline Copy Activity vs. Shortcut
  • I can design a star schema with fact and dimension tables
  • I understand how Shortcuts work and their limitations (read-only, no transformation)
  • I can write basic Spark SQL, T-SQL, and KQL queries
  • I know the incremental refresh parameter pattern (RangeStart / RangeEnd)
  • I can describe Delta Lake format, V-Order, and their role in Direct Lake

📐 Domain 3 — Implement and Manage Semantic Models (25-30%)

  • I can choose between Import, DirectQuery, Direct Lake, and Composite storage modes
  • I can choose between Direct Lake on OneLake and Direct Lake on SQL analytics endpoint (multi-source, composite, fallback, SQL-endpoint security)
  • I know the Direct Lake guardrails, framing/refresh (auto-updates) process, and fallback behavior (SQL flavour only)
  • I can write CALCULATE with ALL, ALLEXCEPT, REMOVEFILTERS, and KEEPFILTERS
  • I understand iterator functions (SUMX, AVERAGEX, COUNTX, RANKX)
  • I know when to use SUMMARIZECOLUMNS vs. SUMMARIZE
  • I can implement time intelligence with a proper Date table
  • I understand USERELATIONSHIP for role-playing dimensions
  • I can configure RLS with DAX and test it with “View as Role”
  • I know how to optimize a model (remove unused columns, avoid bi-directional filters, use variables)

Exam Tip: The DP-600 is heavily weighted toward Domain 2 (Prepare Data) at 45-50%. Spend the most study time on data ingestion patterns, star schema design, and query languages.

← 03 — Implement and Manage Semantic Models Appendix A — KQL Reference →