PostgreSQL extensions in Microsoft Fabric HorizonDB

Fabric HorizonDB provides a curated PostgreSQL extension catalog. The service controls package versions, native libraries, startup preloading, and lifecycle compatibility. You can't install arbitrary operating-system packages or native libraries, and you don't receive PostgreSQL superuser access.

Important

Fabric HorizonDB is in public preview. Capabilities, limits, and behavior might change before general availability. For more information, see Microsoft Fabric preview features.

The available extension set can vary by preview environment. Inspect your database before designing an extension dependency; package visibility doesn't establish customer activation support.

Service-managed extensions

The preview can include these application and analytics extensions:

Extension Purpose
azure_ai Service integration point for approved AI operations. Its presence doesn't guarantee a model, provider, endpoint, region, quota, or billing option.
vector Vector data types and similarity operations in the released catalog.
pg_diskann DiskANN vector indexing in the released catalog.
pg_textsearch Enhanced text-search capabilities.
pg_fabric_fdw Read supported Fabric data from Fabric HorizonDB.

Don't create, drop, replace, update, or alter service-owned extension objects. If an expected extension is missing, treat the condition as a provisioning or service-health problem.

List installed extensions

Run this query on the PostgreSQL endpoint:

SELECT
    extension.extname AS extension_name,
    extension.extversion AS installed_version,
    namespace.nspname AS schema_name
FROM pg_catalog.pg_extension AS extension
JOIN pg_catalog.pg_namespace AS namespace
  ON namespace.oid = extension.extnamespace
ORDER BY extension.extname;

For Federated Analytics, verify that pg_fabric_fdw and the default fabric_dw foreign server are present. If either is missing, treat it as a provisioning or preview-enablement issue. Don't run CREATE EXTENSION or recreate the reserved server to repair it. Save sanitized query output and correlation information for your preview contact.

Discover packaged SQL extensions

PostgreSQL exposes extension control files through pg_available_extensions:

SELECT
    name AS extension_name,
    default_version,
    installed_version,
    comment
FROM pg_catalog.pg_available_extensions
ORDER BY name;

This result is discovery information, not an authorization list. Before you activate a customer-managed extension, confirm the released Fabric HorizonDB catalog for:

  • The exact extension and package version.
  • PostgreSQL compatibility.
  • Your database profile, region, and milestone.
  • The supported activation method.
  • Required privileges and dependencies.
  • Startup-preload behavior.
  • Networking, identity, and data-access restrictions.
  • Pause, resume, maintenance, and upgrade behavior.

Don't infer support from a package inventory or another managed PostgreSQL service's extension list.

Activate a customer-managed extension

Only use CREATE EXTENSION when the released catalog identifies the component as a customer-activatable SQL extension and your role has the required permission. For example, the general PostgreSQL form is:

CREATE EXTENSION IF NOT EXISTS extension_name;

Replace extension_name only with an approved catalog entry. Some packaged components are modules, output plugins, or background workers rather than SQL extensions, so CREATE EXTENSION isn't a valid activation method for them.

If activation is rejected, don't attempt to change shared_preload_libraries, load an arbitrary library, or work around the service policy. Record the complete PostgreSQL error and verify the released catalog.

Lifecycle behavior

Confirm supported activation and lifecycle behavior for each extension with your preview contact. Don't remove or replace service-managed extensions. You remain responsible for application compatibility, customer-owned objects, and query cost.

Security guidance

  • Connect as the database creator. Don't create custom login roles or memberships to bypass the current preview's identity restriction.
  • Review whether an extension sends data to another service or opens an external connection.
  • Don't place credentials in SQL definitions, query text, or support evidence.
  • Treat AI integration availability separately from the presence of azure_ai.
  • Treat foreign-data access separately from the presence of pg_fabric_fdw; the remote source still requires an approved identity and authorization.