Diagram showing a manually triggered failover that promotes the standby scheduler in Azure Region B and redirects clients from Region A.
Azure Functions Durable Task Scheduler (DTS) now has cross-region disaster recovery, but it is a preview feature, and failover is something you trigger yourself. Microsoft's announcement on November 17, 2026 puts it in public preview. The feature is aimed at teams whose durable workflows can't simply restart from scratch after a regional outage.

Diagram showing a manually triggered failover that promotes the standby scheduler in Azure Region B and redirects clients from Region A. What Microsoft announced​

Disaster recovery pairs two Durable Task Scheduler resources, a primary and a secondary, that must run in two different Azure regions. The service replicates orchestration and entity state to the secondary. If you need to move, you promote the secondary and work resumes from that replicated state instead of starting over.

DTS is a managed durable-execution backend. It works with Durable Functions, the Durable Task SDKs and supported agent-framework integrations. Your application code can run in Azure Functions, Azure Container Apps or AKS. The disaster recovery feature covers the scheduler and its workload state only. It does not replicate your compute or reroute your traffic.

Microsoft's documentation also carries the usual preview caveat: disaster recovery for Durable Task Scheduler is in public preview, and details may change before general availability. Treat it as a feature to evaluate now, not yet one to build a production recovery plan on.

This was a long-promised capability. An early public-preview roadmap post listed failover and other business continuity and disaster recovery capabilities as planned work.

How the pairing works​

  • Primary: It serves all active data-plane traffic, including clients, workers and the Durable Task SDKs, and processes task hub work.
  • Secondary: It receives replicated state but does not serve business traffic before promotion. Microsoft's Learn documentation says every data-plane business call to it returns HTTP 503 Service Unavailable. You can't use it for read-only queries or smoke tests.
  • Pairing rules: The pairing is one-to-one. A scheduler can have only one partner, and a secondary can't serve multiple primaries.

Because the standby rejects business calls, Microsoft recommends validating the full workload through a planned failover.

Requirements before you enable it​

  • Both schedulers must use the Dedicated SKU.
  • They must be in different Azure regions.
  • The scheduler you plan to use as the secondary must not already have task hubs.
  • Your deployment identity needs Contributor, or another Azure RBAC role with the required permissions, on both schedulers.
  • Client and worker identities need a DTS data role on both schedulers. Most applications need Durable Task Data Contributor. Durable Task Worker suits apps that only process orchestrations, activities and entities. Durable Task Data Reader suits read-only apps. The Contributor role alone doesn't grant access to orchestration data.
  • Create your task hubs on the primary first. When you enable disaster recovery, the service copies your existing task hub records to the secondary as part of setting up the pairing. Task hub names stay the same after failover.
  • Microsoft suggests keeping sku.capacity aligned on both schedulers. This isn't enforced. A smaller secondary can throttle after promotion.

Region exclusions. Disaster recovery is available in Azure public cloud regions where DTS is available, except Austria East, Belgium Central, Central US EUAP, Chile Central, East US 2 EUAP, France South, Italy North, Jio India West, Northeast US 5 and Southeast US 5. Check this list before you pick a pair.

The cost​

You need two Dedicated schedulers. Microsoft's documentation says you are billed for both at the same time, including while the secondary sits on standby. Each is billed for its provisioned capacity units at its own region's price. The standby is not free, and sizing it too small risks throttling when you most need it.

Planned and forced failover​

Planned failover is for runbook testing or moving ahead of scheduled maintenance. The sequence is:

  1. The current primary stops accepting new data-plane traffic.
  2. The service performs a planned transition of the underlying database.
  3. The secondary becomes primary, and the old primary becomes the secondary.

It does not drain or wait for running orchestrations. Replicated state resumes once clients and workers point at the new endpoint.

Forced failover is for when the primary is unreachable and a planned failover isn't possible. It runs asynchronously. Any data not yet replicated before the outage is unavailable afterward, so you should expect possible data loss. Updates to the old primary are best-effort while its region is down. Before failing back, confirm it is healthy and its replicationRole is Secondary.

Failback is manual. Disaster recovery doesn't automatically fail back, so you decide when and whether to do so. If both schedulers are healthy, a planned failover returns traffic to the original region.

Stopping replication. To keep both schedulers but end replication, disable disaster recovery on the current primary. This cleans up the relationship on both sides. If the partner region is down, cleanup there may have to wait until it recovers.

What this means for your recovery plan​

This is not automatic regional failover. Each scheduler keeps its own endpoint, and there is no shared or floating endpoint, so you must redirect your clients and workers to the new primary's endpoint after a failover. Several things remain your job:

  • Making your endpoint configuration switchable, whether through app settings, configuration services or deployment automation.
  • Making sure your app identities already hold roles and have network access on the partner scheduler.
  • Having compute running, or ready to start, in the surviving region.
  • Deciding who has the authority to call a forced failover, and how much data loss is acceptable.

This analysis is my own reading of Microsoft's documented behavior, not a Microsoft recommendation.

Where it sits in the wider DTS update​

Microsoft's blog post bundles this with several other DTS announcements. Task queues are also in public preview. Private endpoints and customer-managed keys for Dedicated schedulers are now generally available. The Durable Task Go SDK is in preview. For regulated or business-critical workloads, customer-managed keys and private endpoints plus a cross-region standby make a more complete enterprise story. Only the disaster recovery piece is still in preview.

Bottom line​

For teams running critical durable workflows, such as AI agent orchestration or long-running business processes, this fills a real gap. It doesn't remove the need to rehearse. Do the following:

  1. Check your regions against the exclusion list.
  2. Budget for a second Dedicated scheduler.
  3. Set up RBAC on both schedulers.
  4. Build endpoint switching into your clients and workers.
  5. Run a planned failover and fail back before you need either one.

A recovery plan that has never been run is a hypothesis, not a plan.

 

References

  1. (In preview) Public Preview: Native disaster recovery for Azure Functions durable task scheduler Azure Updates 2026-11-17T20:00:32Z
  2. What's new in Azure Functions durable task scheduler: Durable agents, disaster recovery, and more | Microsoft Community Hub techcommunity.microsoft.com
  3. Disaster recovery in Durable Task Scheduler - Durable Task learn.microsoft.com