Azure Database for PostgreSQL

At a glance

What is it?

Azure Database for PostgreSQL (the flexible server deployment) is a managed relational database service that runs the community edition of PostgreSQL, so existing applications usually move with little refactoring. The engine runs in a container on a Linux virtual machine, with data files on Azure storage kept as three synchronous copies. You choose Burstable, General Purpose or Memory Optimized compute, can stop the server to pause compute billing, and can add a warm standby in another availability zone. Elastic clusters, a managed form of the open source Citus extension, shard data across several nodes.¹³

What does it do?

The service patches the OS and engine in a system-managed or custom maintenance window, takes automatic backups to zone-redundant storage (7 days by default, up to 35), encrypts data at rest and in transit, supports private access through virtual networks and includes a built-in PgBouncer. For AI it supports pgvector for vector similarity search, DiskANN for large-scale vector indexing and the azure_ai extension, which lets SQL call Azure OpenAI in Microsoft Foundry models, Azure Language in Foundry Tools and Azure Machine Learning endpoints. Data can be mirrored into Microsoft Fabric.¹⁴⁵

Who needs it?

Teams that build on PostgreSQL and want a managed service instead of running their own servers, including those migrating from on-premises PostgreSQL, Amazon RDS or Aurora, Google Cloud SQL or AlloyDB, or EDB with the built-in migration service. It also suits developers who want vectors and LLM calls inside the database for retrieval-augmented generation.⁵

What does it need?⁴⁶

• An Azure subscription
• A supported PostgreSQL major version for your application
• A network choice: public access or private access through an Azure virtual network
• Extensions added to the server allowlist before running CREATE EXTENSION
• For azure_ai: an Azure OpenAI or Foundry endpoint and an API key or managed identity

What it is not

It is not a proprietary fork: it runs community PostgreSQL versions. It is not Azure HorizonDB, a separate cloud-native PostgreSQL service in preview with disaggregated compute and storage. It is not Azure DocumentDB, the MongoDB-compatible service that also builds on PostgreSQL. The azure_ai AI functions (embedding generation, generate, is_true, extract and rank) are marked preview, and the V6 SKU family is in public preview.¹²⁴

Availability and licensing. Generally available; some regions temporarily block new zone-redundant HA deployments (checked 26 Oct 2026). The V6 SKU family and the azure_ai AI functions are in preview. Microsoft's product page says the Azure free account includes up to 750 hours of Burstable compute and 32 GB of storage a month for 12 months. Azure HorizonDB, a separate cloud-native PostgreSQL service, was in private preview at Ignite 2025 and announced in preview at Build 2026; its Learn page lists nine regions and gaps such as no cross-region replicas.¹²⁷⁸

The problem it solves

How it works

Capabilities

• Managed operations and high availability¹

  Automated patching in a maintenance window, automatic backups with point-in-time restore (7 to 35 days), and zone-redundant high availability with a synchronous warm standby.

  Why it matters: Removes routine database operations and protects against an availability zone failure.

  Limits: Some regions temporarily block new zone-redundant HA deployments; patching may restart the server.

• Cost controls¹

  A Burstable tier for low-concurrency workloads and stop and start on demand, which stops compute billing immediately.

  Why it matters: Development, test and time-bound workloads cost less.

  Limits: A stopped server restarts automatically after seven days; V6 SKUs cannot scale to or from Burstable.

• Elastic clusters for sharding³

  A managed offering of the open source Citus extension that shards PostgreSQL across nodes with row-based or schema-based sharding and online rebalancing.

  Why it matters: Multitenant and large datasets can scale beyond a single server.

  Limits: All nodes must use the same compute and storage tier; DDL runs only through the coordinator node.

• Vector search with pgvector and DiskANN⁶⁹

  The pgvector extension adds vector types, distance operators and functions; DiskANN provides efficient large-scale vector indexing.

  Why it matters: RAG and semantic search can run inside the operational database.

  Limits: Extensions must be allowlisted per server and created in each database; the pgvector extension is named vector.

• azure_ai extension⁴

  Lets SQL call Azure OpenAI in Microsoft Foundry models, Azure Language in Foundry Tools and Azure Machine Learning endpoints, with managed identity support.

  Why it matters: Embeddings, classification and LLM calls can be part of SQL queries and pipelines.

  Limits: The AI functions (embedding generation, generate, is_true, extract and rank) are in preview, and model calls are billed by the AI service.

• Fabric mirroring and migration service⁵¹⁰

  Mirroring replicates data into Microsoft Fabric OneLake in near real time; the migration service moves databases online or offline from on-premises, Amazon RDS and Aurora, Google Cloud SQL and AlloyDB, and EDB.

  Why it matters: Simplifies both moving into Azure and analyzing the data afterwards.

  Limits: Mirroring requires a running Fabric capacity; online migration needs replication prerequisites at the source.

Practical use cases

Who uses it

• Air India

  In production

  Air India: AI.g customer service agent on Azure OpenAI

  about 40,000 (2026), up from an average of 30,000 (2024) (measured)

  Travel and hospitality, India

Works with

Relationship labels follow Microsoft's documentation. "Alternative approaches" does not mean one is better: each profile says when it fits.

Getting started

1. Create a server

   Create an Azure Database for PostgreSQL flexible server in the portal, CLI or Bicep, starting on the Burstable tier for development.

   Check: You can connect and run a query with the PostgreSQL extension for Visual Studio Code.

2. Set networking and HA

   Choose public or private (virtual network) access and enable zone-redundant high availability for production.

   Check: The portal shows the standby in a different availability zone.

3. Enable AI extensions

   Add vector and azure_ai to the azure.extensions allowlist, then run CREATE EXTENSION in each database.

   Check: SHOW azure.extensions lists both extensions.

4. Load or migrate data

   Use pg_dump and pg_restore for offline moves or the migration service for online migration.

   Check: Row counts match the source.

5. Connect analytics

   Set up Microsoft Fabric mirroring for the server.

   Check: Tables appear in a Fabric mirrored database.

Official resources

• Product pageazure.microsoft.com (opens in a new tab)
• Documentationlearn.microsoft.com (opens in a new tab)

Sources

Each statement above links to the source it comes from. Labels say who reported it.

1. What is Azure Database for PostgreSQL flexible server? (opens in a new tab) Microsoft · Vendor-reported
2. What is Azure HorizonDB? (opens in a new tab) Microsoft · Vendor-reported
3. Elastic clusters in Azure Database for PostgreSQL (opens in a new tab) Microsoft · Vendor-reported
4. azure_ai extension in Azure Database for PostgreSQL (opens in a new tab) Microsoft · Vendor-reported
5. Azure Database for PostgreSQL documentation (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
6. Vector search in Azure Database for PostgreSQL (opens in a new tab) Microsoft · Vendor-reported
7. Microsoft Ignite 2025 Book of News (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
8. Microsoft Build 2026 live blog (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
9. Azure Database for PostgreSQL product page (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
10. Mirroring in Microsoft Fabric (opens in a new tab) Microsoft · Vendor-reported