Azure Database for PostgreSQL flexible server: September 2026 release

The September 2026 version of Azure Database for PostgreSQL is now available. This version introduces new features and enhancements, resolves known issues, and includes essential security patches to ensure optimal performance and security.

Note

New servers created on or after September 11, 2026, automatically onboard to this latest version. Existing servers update during their next scheduled maintenance. Azure deployments roll out region by region, so availability might vary and the global rollout can take up to one month. Monitor your scheduled maintenance notifications to know when the update will be applied to your server.

Engine version changes

  • No PostgreSQL minor version change.

Features

  • Added the PgBouncer client connections metric (client_connections_total), which reports the total client connections tracked by PgBouncer across all states, including active/idle, waiting, and authenticating connections (Preview).
  • Fabric Mirroring: Support for secure PostgreSQL mirroring to OneLake through Microsoft Fabric workspace Private Link using the azure_cdc.onelake_use_workspace_dns server parameter.
  • Fabric Mirroring: Support for mirroring partitioned PostgreSQL tables to Microsoft Fabric as a single table, with automatic synchronization when partitions change.

Improvements

  • Improved error reporting for server provisioning failures due to unavailable compute SKUs or insufficient capacity, replacing generic internal server errors with specific error messages.
  • Multiple security fixes and hardening improvements to enhance platform security and reliability.

Known issue fixes

  • Fixed an issue where large PostgreSQL log entries could be silently dropped. Entries that exceed the supported size limit are now truncated with a clear marker.
  • Fixed an issue where audit-log forwarding could stop after a logging-component upgrade.
  • Improved server-log collection recovery following failed logging-configuration updates.
  • Resolved an issue where deletion of high availability-enabled servers could fail when underlying server instances were in a transitional state.