A cloud database infographic shows daily snapshots and continuous logs enabling PostgreSQL recovery to any point in time.
Azure HorizonDB, Microsoft's PostgreSQL-compatible database service, now lets administrators set automated backup retention anywhere from 1 to 35 days. Microsoft's Azure Updates entry for the change was published on November 17, 2026. Its title says "Generally Available" but also carries an "[In preview]" tag. That mixed label turns out to match the rest of the documentation.

A cloud database infographic shows daily snapshots and continuous logs enabling PostgreSQL recovery to any point in time. What changed​

Until now, HorizonDB clusters kept automated backups for a fixed seven days. One Microsoft Learn overview page still lists configurable backup retention as "Not yet available", with retention currently seven days and work under way to enable 1-35 days. That page is stale, and more on that below.

The new range is now documented elsewhere. Microsoft's HorizonDB security guidance, updated on the announcement date, says the service performs daily backups of database files and continuously backs up transaction logs. Microsoft's REST reference for HorizonDB clusters exposes the setting as retentionDays, with a minimum of 1, a maximum of 35 and a default of 7. Nothing in the reference describes a zero-day option, so don't assume you can switch automated backups off.

The documentation doesn't agree with itself​

If you go looking for confirmation, you may hit contradictions. Three Learn pages disagree, and some of them predate this launch:

  • Backup concepts page: it says the default retention period is currently seven days. It also lists maximum backup retention currently supported as seven days as a limitation.
  • Business continuity page: it states that backups are currently retained for seven days.
  • Updated security guidance: it now says the service allows keeping automated backups for 1-35 days. The same page also tells readers to validate recovery requirements against "preview backup limits".

Microsoft's own overview page shows the same split. Its "what you manage at each level" table lists backup retention (1 to 35 days) as a cluster-level setting. Its limitations table still says retention is capped at 35 days and that long-term retention is unavailable. The most likely explanation is that the pages were updated at different times. I haven't seen Microsoft confirm that, so check your own environment before relying on any single page.

How the setting works​

Microsoft's REST documentation describes the configuration like this:

  1. retentionDays sits in the cluster's backup properties.
  2. If you omit it, the service uses 7 days. A cluster that was running before the announcement should therefore be checked, because you can't assume it moved to a longer window.
  3. Updates are an Azure Resource Manager PATCH to the cluster resource, with the backup settings nested under properties. Microsoft's example sets 21 days.
  4. The example response is 202 Accepted with a Location header and a Retry-After of 30 seconds, which means the update is asynchronous. Track the operation to completion rather than treating the 202 as proof the change is live.
  5. The reference's API version is 2026-10-01-preview. A preview API label doesn't tell you which management surfaces expose the feature as generally available.

I could not confirm the exact Azure portal steps, a CLI command, the regional rollout or any pricing change from the available material. I'm not going to guess at menu labels.

What the retention window does and doesn't do​

The retention period controls how far back you can recover. Per Microsoft's overview, HorizonDB archives write-ahead log data to Azure blob storage and keeps it for the configured short-term retention time. The backup concepts page adds that the configured retention policy determines how long the system retains WAL files. A longer window therefore keeps older recovery points available. It doesn't change how often backups are taken or how fast restores run. Microsoft says restores are quick because they use storage snapshots, regardless of data size.

Recovery works by creating a new cluster. The security guidance says point-in-time restore creates a new cluster from a restore point inside the current retention window. It isn't an in-place rollback. The same guidance warns that deleted clusters can't currently be restored, and recommends Azure resource locks on critical clusters.

Restored clusters also keep the retention setting. According to Microsoft, a restored cluster inherits the backup retention period from the source snapshot.

Cost​

Longer retention means more stored WAL and snapshot data, and Microsoft bills for it. The overview lists used backup storage for the short-term retention period among HorizonDB's charges. Backup costs also continue after deletion. For a deleted database, Microsoft says billable backup storage is scaled by the remaining retention period over the configured one.

I can't tell you what 35 days costs for a given workload. The material doesn't include figures. A busy, write-heavy cluster will generate far more WAL than a quiet one, so test before you set 35 days across a fleet.

Limits to keep in mind​

  • Not long-term retention. Microsoft still lists long-term retention (LTR) as unavailable. Thirty-five days is a short-term recovery window, not an archive or a compliance answer for multi-year retention.
  • Not disaster recovery. Retention length is separate from geo-redundant backup. Microsoft's REST reference says geo-redundant backup isn't supported when a cluster uses Azure Key Vault encryption. That was already the case and isn't a consequence of this launch.
  • HorizonDB only. Azure SQL Database already offers 7 days by default, configurable between 1 and 35 days, with different tier rules. Don't carry HorizonDB assumptions over to those services, or the reverse.

Practical checklist​

  1. Read the current retentionDays value on each HorizonDB cluster rather than assuming it changed.
  2. Choose a value based on how long it takes your team to notice bad data, for example a faulty deployment or an accidental delete. If problems often surface weeks later, 7 days may be too short.
  3. Check earliestRestorePointUtc in the cluster properties to see how far back you can actually restore.
  4. Watch backup storage consumption after a change, especially on large or write-heavy clusters.
  5. Practice a point-in-time restore into a new cluster, so you know how long it takes and how your connection strings and access settings carry over.
  6. Keep a separate retention plan for anything that needs more than 35 days.

Bottom line​

This is a useful, practical change for teams running HorizonDB, because a fixed seven-day window was tight for many recovery scenarios. Microsoft's service is still labeled preview across much of its documentation. Several pages still show the old seven-day limits. Treat the 1 to 35 day range as real, confirm it in your own subscription and region, and don't treat longer retention as a substitute for a complete backup strategy.

 

References

  1. (In preview) Generally Available: Customize backup retention from 1 to 35 days in Azure HorizonDB Azure Updates 2026-11-17T23:46:26Z
  2. Secure Your Azure HorizonDB - Azure HorizonDB learn.microsoft.com
  3. What Is Azure HorizonDB? - Azure HorizonDB | Microsoft Learn learn.microsoft.com