Treat per-disk resiliency as application-visible I/O failure, not uninterrupted disk access

Source facts

With the per-disk resiliency preview enabled, Azure can detach a delayed data disk while leaving its VM running, then reattach it after the platform issue resolves. I/O fails while the guest cannot access the disk. The VM reports Degraded, while the portal can still show the disk attached. The feature excludes OS disks, Write Accelerator VMs, and VMs created before August 15, 2026; it is region-limited. Restore points, Azure Backup, and Site Recovery do not preserve its setting. Microsoft Learn.

Applicability

Confirm preview eligibility and authorization first. Enablement must occur while the VM is deallocated or before disk attachment, not during VM or scale-set creation. Identify the affected application’s behavior when a required disk temporarily disappears.

DSE recommendation

Require an application-specific response to disk disappearance before evaluating this preview. Ask the workload owner to distinguish the benefit of keeping unrelated work running from the recovery needs of the disk-dependent component. Document error handling, operator escalation, and the decision to resume disk-dependent work. Add the setting to post-restore configuration checks rather than assuming recovery preserves it.

Verification

Use an approved nonproduction validation plan to compare platform health, guest disk visibility, and application behavior. Preserve observed I/O errors and recovery results without claiming the platform threshold can be chosen or that reattachment proves application consistency. After any restore rehearsal, inspect the actual disk setting before treating the original resiliency design as restored.

Official references

Microsoft Learn: Improve workload availability with per-disk resiliency preview. Source reviewed September 9, 2026.