The announcement itself is brief. The HorizonDB REST API documentation is where the practical details are, and those details should shape how you plan recovery.
What this feature is, and what it isn't
This is a restore-based disaster recovery path. It is not a live replica and it is not automatic failover. In a regional disaster you create a new cluster from an available backup in a different Azure location. The reviewed Microsoft material doesn't promise zero data loss, a recovery-time target, or continuity without restore work. It also doesn't say how long backups take to replicate between regions, which region pairs are supported, or what the feature costs. Treat all of those as open questions to confirm with Microsoft or in your own testing.
The earlier state of the docs shows how new this is. An earlier HorizonDB backup article listed geo-redundant backups as not currently supported. That page also described the service as being in preview at the time. Some Learn pages may still carry that older wording until they are refreshed.
How the cross-region restore works
Microsoft's HorizonDB API documents a specific example for restoring a cluster across Azure regions from the latest available backup. It uses the PointInTimeRestore creation mode. These are the behaviors that matter:
- It creates a new cluster. Point-in-time restore is built on WAL archiving plus periodic snapshots. The cross-region restore hydrates a new cluster's storage from an available source backup.
- A cross-region restore needs the source location. You supply
creationSourceClusterLocationalong with the source cluster's resource ID. The target location differs from the source. - Your timestamp is ignored across regions. The request must still include
pointInTimeUtc. For a cross-location restore, the service ignores it and restores the latest available backup. The response omits the field. A same-location restore honors the timestamp and uses the latest backup at or before it. - Omit the PostgreSQL version. The service reads it from backup metadata. The API version documented,
2026-10-01-preview, accepts PostgreSQL 17 and 18 for ordinary cluster creation. - The geo-redundancy setting is inherited. On restore, the service ignores the
backup.geoRedundancyvalue you request. It configures the new cluster with the same setting as the source cluster. - A source without a pool can still be restored. It only needs an available backup.
The API's example restores from eastus2 to westus2 with public access disabled. Those are sample values. They don't show that every region pair is supported.
What you won't get back automatically
A restore doesn't clone the surrounding configuration. According to the API documentation, it does not copy:
- pools
- pool-scoped endpoints
- firewall rules
- parameter-group associations
- other child resources
The new cluster gets a service-generated cluster-scoped endpoint named cluster. Any application that connected through a pool-scoped endpoint or relied on firewall rules will need those pieces rebuilt. Your runbook should include them. Public access defaults to disabled when no network object is supplied, so a restored cluster may be unreachable until networking is configured.
The encryption catch
This is the limit most likely to surprise teams. Geo-redundant backup is not supported when dataEncryption.type is AzureKeyVault, meaning customer-managed keys. In that case backup.geoRedundancy must be Disabled. The encryption type is chosen when the cluster is created and can't be switched later between system-managed and Key Vault. A cluster created with customer-managed keys can't be converted afterward to gain geo-redundancy.
WindowsForum has already covered this trade-off. Customer-managed keys in HorizonDB are in preview and exclude geo-redundant backups. If your compliance rules require customer-managed keys, you have to decide between key control and regional backup protection. The two can't be combined today.
Retention settings
The API documents retentionDays as 1 to 35 days, with a default of 7 when omitted. Microsoft's security overview likewise states that HorizonDB automated backups can be kept for 1-35 days. Retention sets how far back backups reach. It is not a recovery-point guarantee and it says nothing about replication delay.
A practical checklist
- Inventory your HorizonDB clusters and check which use Azure Key Vault encryption. Those can't use geo-redundant backup.
- Enable geo-redundancy at cluster creation where regional recovery matters. Check the
backup.geoRedundancysetting on existing clusters. - Record each cluster's resource ID and Azure location, and choose a target region.
- Write the restore as an API or template operation using
PointInTimeRestore. Include the source ID and location, and the requiredpointInTimeUtcfield. Plan around the latest backup, not a chosen time. - Script the post-restore work: pools, endpoints, firewall rules, parameter groups, and application connection strings.
- Run a drill. Measure how long recovery really takes and how much recent data you would lose, because Microsoft hasn't published either figure.
- Consider resource locks as well. Microsoft's security guidance recommends them on critical clusters because deleted servers can't currently be restored in HorizonDB, per the security overview's guidance. Locks protect against deletion. They don't replace cross-region recovery.
How this compares with other Azure databases
Other Azure databases have offered geo-restore for some time. Azure Database for MySQL's geo-restore uses the server's most recent backup. That is the same "latest backup" behavior HorizonDB documents. In practice it means a restore point can lag the primary, so a disaster recovery plan has to accept some data-loss exposure. For stricter targets, teams usually look at replication rather than backup restore. HorizonDB's documentation does describe in-region read nodes as failover targets. The reviewed sources don't describe a cross-region replica option for it.
Bottom line
For HorizonDB users, the GA is a useful safety net against a regional loss, as long as you treat it as a rebuild procedure and not as failover. Check your encryption choice first, because it decides whether you can use the feature at all. Then script the post-restore steps and test the timing yourself.
References
- (Launched) Generally Available: Geo-redundant backup and restore for Azure HorizonDB Azure Updates · 2026-11-17T23:47:55Z
- Clusters - Create Or Replace - REST API (Azure HorizonDB) | Microsoft Learn learn.microsoft.com
- Secure your Azure HorizonDB - Azure HorizonDB | Microsoft Learn learn.microsoft.com