An Azure service that runs native VMware workloads on Azure.
Assign a vSAN policy with site disaster tolerance set to Preferred or Secondary to the VM namespace and every VMDK; only after all relevant objects are compliant will that VM’s data remain in one site instead of being synchronously mirrored across both availability zones.
Configure a single-site policy
For the 1,000 VMs that do not require cross-zone protection:
- Create a custom vSAN storage policy with:
- Site disaster tolerance (
vSANSiteDisasterTolerance):-
Preferred— no site redundancy; keep data on the preferred site, or -
Secondary— no site redundancy; keep data on the secondary site.
-
- Local failures to tolerate (
vSANFailuresToTolerate): choose the required within-site protection, such asR1FTT1for RAID-1 protection against one host failure.
None/NoneStretchlocality for this purpose. The documented stretched-cluster settings specifically providePreferredandSecondaryfor retaining data on one site.Dualis the cross-site mirroring option.^1^ - Site disaster tolerance (
- Apply the policy to every existing VMDK of each selected VM using the vCenter storage-policy UI/API. On SDDCs running vCenter 8, AVS supports changing VM storage policies natively through vCenter UI/API.^1^
- Apply the policy to the VM namespace/home object as well:
- Azure portal → AVS private cloud → Run command → Packages →
Set-VMStoragePolicy - Enter the exact VMName and StoragePolicyName, then select Run.
- Monitor Notifications for completion.
Set-VMStoragePolicydoes not alter existing disk-level policies. Using this Run Command without step 2 can therefore leave the VMDKs dual-site mirrored.^1^ - Azure portal → AVS private cloud → Run command → Packages →
- In vCenter, inspect the VM namespace and each VMDK and confirm:
- The assigned policy is the intended
PreferredorSecondarypolicy. - Every object reports policy compliance.
- No VMDK remains assigned to the
Dual-Site Mirroringpolicy.
- The assigned policy is the intended
- Pilot the change on a small VM group, allow vSAN policy reconfiguration to complete, and then process the remaining VMs in controlled batches. Do not treat the operation as complete while objects are noncompliant or still being reconfigured.
Availability and SLA impact
This configuration intentionally removes availability-zone protection for those VMs. A stretched cluster normally uses synchronous vSAN replication so both zones confirm writes; with single-site locality, the affected data has no copy in the other site.^2^
The AVS stretched-cluster 99.99% infrastructure availability commitment requires Dual-Site Mirroring with local SFTT 1. VMs changed to Preferred or Secondary do not meet that dual-site policy requirement.^3^
Classify the 1,000 VMs by recovery objective before changing them:
- Keep critical, zero-RPO workloads on
Dual. - Use
PreferredorSecondaryonly where an AZ outage and recovery from backup or another DR mechanism is acceptable. - Retain an appropriate local SFTT setting if protection from host or disk failure within the selected site is still required.
Cost effect
Changing the storage policy does not directly reduce the AVS bill. AVS stretched clusters are charged by deployed nodes, not by how many VMs use dual-site mirroring; the witness and inter-AZ traffic are not separately charged.^3^
Savings occur only if reduced storage consumption permits host removal while preserving compute, memory, vSAN capacity, and operational headroom. Stretched clusters:
- Scale in and out only in host pairs.
- Require equal host counts across both sites.
- Have a minimum of six hosts.
- Cannot contain a mixture of stretched and standard clusters.
- Cannot be converted from stretched to standard; that requires a new cluster/private cloud and workload migration.^2^
Therefore, model the reclaimable vSAN capacity after policy conversion and confirm that it enables at least one valid two-host scale-in unit; otherwise, the policy change reduces replicated capacity but not the node charge.
References