SQL platforms & tooling

One engine, three platforms

SQL Server, Azure SQL and SQL database in Fabric share one code base — the SQL Database Engine. The T-SQL is the same; what differs is who runs the engine, how new features reach it and which features are switched off by the platform. Every T-SQL reference page carries an Applies to banner listing the platforms; these notes copy that habit with a strip like the one below.

flowchart TB
  subgraph ENGINE["One SQL Database Engine"]
    SS["SQL Server 2025 · 17.x<br/>you run it: Windows · Linux · containers · Azure VM · Arc"]
    subgraph AZ["Azure SQL · PaaS"]
      DB["Azure SQL Database<br/>versionless · database-scoped"] ~~~ MI["Azure SQL Managed Instance<br/>instance-scoped · update policy"]
    end
    FAB["SQL database in Fabric<br/>SaaS · auto-mirrored to OneLake"]
  end

Managed Instance update policies

An Azure SQL Managed Instance is pinned to an update policy that decides which engine features it receives. This is the single most common reason a DP-800 answer says "MI: yes, but…".

SQL database in Fabric

SQL database in Microsoft Fabric is an operational (OLTP) database created as an item in a Fabric workspace. It shares Azure SQL Database's engine and adds Fabric's integration: every eligible table is automatically mirrored to OneLake in Delta format, a read-only SQL analytics endpoint is created alongside it, and the workspace's Git integration and deployment pipelines cover its schema.

• Identity: Microsoft Entra only. Workspace roles grant access, and SQL permissions refine it.
• Not supported today: Always Encrypted, ledger tables, in-memory (memory-optimized) tables, change data capture, application roles, server-level roles, EXECUTE AS, SQL Agent, partition switching.
• Supported with differences: full-text search (preview), external tables (Parquet and CSV only), auditing (written to the item's OneLake folder and read with sys.fn_get_audit_file_v2), clustered columnstore (create it with the table, or stop mirroring first).
• Limits: up to 32 vCores and 4 TB; a trial capacity is limited to three databases.

Which feature exists where

The matrix below is the exam's platform line for every feature the skills outline names. GA is generally available, Preview is in preview (and on SQL Server 2025 needs PREVIEW_FEATURES), and No is not available. "MI" assumes an update policy that carries the feature, noted where it matters.

AUTD is the Always-up-to-date update policy; "2025" is the SQL Server 2025 update policy. Statuses checked against the live documentation in November 2026 — previews graduate in cumulative updates and service releases, so check the banner on the reference page before relying on one.

Switching features on

Three switches explain most "the function does not exist" and "the feature is disabled" errors in exam scenarios. Each applies to some platforms only.

-- SQL Server 2025 only: opt a database in to preview features
-- (vector indexes, VECTOR_SEARCH, fuzzy matching, change event streaming,
-- half-precision vectors). Not needed in Azure SQL Database or Fabric.
ALTER DATABASE SCOPED CONFIGURATION SET PREVIEW_FEATURES = ON;

-- Several new functions are only visible at compatibility level 170
-- (AI_GENERATE_CHUNKS, for example). 170 is already the default in Azure SQL
-- Database and SQL database in Fabric; on SQL Server it follows the database.
ALTER DATABASE CURRENT SET COMPATIBILITY_LEVEL = 170;

-- SQL Server 2025 and Azure SQL MI: allow outbound REST calls
-- (sp_invoke_external_rest_endpoint, external models). On by default in
-- Azure SQL Database and SQL database in Fabric.
EXECUTE sp_configure 'external rest endpoint enabled', 1;
RECONFIGURE WITH OVERRIDE;

PREVIEW_FEATURES (database scoped configuration)

SQL Server 2025 only. Enabling it does not make the database "non-GA"; a preview that later goes GA in a cumulative update simply stops needing it. Persisted preview objects (a vector index) can fail if you switch it off again.

Compatibility level 170

Required by some new functions — the engine cannot even find AI_GENERATE_CHUNKS below 170.

external rest endpoint enabled (sp_configure)

Gates sp_invoke_external_rest_endpoint and therefore every external-model call. Must be enabled on SQL Server 2025 and Managed Instance; already on in Azure SQL Database and SQL database in Fabric.

The developer tooling

The exam names its tools. Know what each one is for, and which skill uses it.

Names to keep straight

Microsoft Foundry (formerly Azure AI Foundry)

The platform for deploying models such as Azure OpenAI embeddings and chat models. The study guide uses the new name; many SQL pages still say "Azure OpenAI in Azure AI Foundry Models", and the old portal is "Foundry (classic)".

SQL database in Microsoft Fabric

Sometimes shortened to "Fabric SQL database" in the docs. Not the Warehouse and not a SQL analytics endpoint.

SQL MCP Server

Data API builder's MCP surface, available from DAB 1.7 and fully featured in DAB 2.0 and later.

SQL Server 2025 = 17.x

SQL Server 2022 is 16.x. The compatibility level that unlocks the newest functions is 170.

ANN and ENN (also kNN)

Approximate nearest neighbours (a vector index, VECTOR_SEARCH) versus exact nearest neighbours (VECTOR_DISTANCE over every row). Skill 3 covers when to choose each.