How To

Set Up a Windows Server 2025 Workgroup Cluster Without a Domain

A failover cluster doesn’t need Active Directory. Since Windows Server 2016 you could build one from workgroup servers, and the Windows Server 2025 workgroup cluster is the first one Microsoft supports for Hyper-V virtual machines, with live migration that works without Kerberos. What it still needs is DNS, one local account that exists on every node with the same password, and a quorum witness the nodes can reach without a domain account. Miss any of those and you’ll hit errors that read like a domain problem, because most cluster documentation assumes there is one.

Original content from computingforgeeks.com - post 180521

This guide builds a two-node failover cluster without a domain from scratch. You get matching local accounts, a cluster created with New-Cluster -AdministrativeAccessPoint Dns, iSCSI storage, both witness types, and a clustered Hyper-V VM moved by a timed live migration. It ends with what breaks without AD, marked tested or documented. Everything below was run on Windows Server 2025 Datacenter in October 2026, on three Proxmox VMs that had never seen a domain controller.

A Workgroup Cluster Runs Some Roles, Not All

The cluster service itself does not care whether its nodes are domain members. What changes is how the nodes prove who they are to each other and to anything that connects to the cluster. A domain cluster gets a computer object (the cluster name object) and uses Kerberos for everything, while a workgroup cluster has neither, so any role that expects clients to authenticate to the cluster name with Kerberos is out. Microsoft’s workgroup cluster documentation gives the supported and unsupported workloads in the middle column; the reasons in the last column are ours, except “authentication issues” for file servers, which is Microsoft’s own wording:

WorkloadWorkgroup cluster on Windows Server 2025Why
Hyper-V virtual machinesSupported (new in 2025)Live migration authenticates with self-signed PKU2U certificates instead of Kerberos
SQL Server Availability GroupsSupportedUses domain-independent availability groups with certificates
File ServerNot supported“Authentication issues”; SMB clients would authenticate to a cluster name that has no account
SQL Server Failover Cluster InstanceNot supportedMicrosoft gives no reason

Read the table as a rule about who logs in. A Hyper-V VM role never asks a client to log in to the cluster, and node-to-node live migration now has its own certificate path, so it works. A clustered file share needs every client to authenticate to a name that has no computer account behind it, so it doesn’t. If your plan is two Hyper-V hosts at a branch office or an edge site with no domain controller on site, this is the build. If you need a highly available file share, join the nodes to a domain and follow the domain-joined failover cluster setup instead.

The Lab Has No Domain Controller

Three VMs sit on an isolated VLAN, which is why the 10.0.1.0/24 addresses below are the real lab addresses. The two cluster nodes run Failover Clustering and Hyper-V. The third box, stor01, plays every supporting part a domain would normally play or that a branch site would have anyway: DNS for the lab.local zone, an iSCSI target with the shared disks, and an SMB share for the witness.

Windows Server 2025 workgroup cluster architecture diagram: node01 10.0.1.21 and node02 10.0.1.22 with heartbeats on UDP 3343 and live migration, vm01 on the CSV, cluster name wgclus 10.0.1.20 registered in stor01's lab.local DNS zone, iSCSI LUNs and an SMB witness share on stor01 10.0.1.30

Each node got 2 vCPUs, 6 GB of RAM and nested virtualization (CPU type host on Proxmox), because the Hyper-V role refuses to install without hardware virtualization. What ran where:

  • node01 10.0.1.21: Failover Clustering, Hyper-V, iSCSI initiator
  • node02 10.0.1.22: the same roles
  • stor01 10.0.1.30: DNS Server, iSCSI Target Server, the witness SMB share
  • wgclus 10.0.1.20: the cluster name, created by New-Cluster

All three started as clones of one Windows Server 2025 template, and that is the first trap. Clones of a template that wasn’t generalized share a machine SID, and Microsoft only supports cloned Windows installs that were generalized with Sysprep, domain or not. A sibling lab on the same template saw a domain join fail with “The user name or password is incorrect” for exactly this reason, so we didn’t build a cluster on top of it. Our template had a broken sysprep state (GeneralizationState 3), so every clone came up with the same SID. Check yours before anything else:

(Get-LocalUser | Where-Object SID -like '*-500').SID.Value

The built-in Administrator’s SID is the machine SID plus -500, so the part before -500 must differ on every node. After the fix below, our three VMs each reported their own:

node01  S-1-5-21-1395323420-1762120319-2583289541-500
node02  S-1-5-21-2713880376-3435347698-2684583517-500
stor01  S-1-5-21-4119681338-1446146690-2191393814-500

If two nodes print the same value, generalize the source and clone again. Our template’s sysprep refused to run until we reset its recorded state, because its GeneralizationState was stuck at 3 and the Panther log showed hr = 0x8007001f. Setting the value to 7 let sysprep proceed. The first line copies the template’s cloudbase-init answer file to where the sysprep command expects it:

New-Item -ItemType Directory -Force C:\lab | Out-Null
Copy-Item 'C:\Program Files\Cloudbase Solutions\Cloudbase-Init\conf\Unattend.xml' C:\lab\unattend.xml -Force
Set-ItemProperty HKLM:\SYSTEM\Setup\Status\SysprepStatus -Name GeneralizationState -Value 7
Start-Process -FilePath C:\Windows\System32\Sysprep\sysprep.exe -ArgumentList '/generalize','/oobe','/shutdown','/quiet','/unattend:C:\lab\unattend.xml'

The unattend file we passed was the template’s own cloudbase-init Unattend.xml, which completes setup on first boot without anyone at the console; a template without cloudbase-init needs its own answer file there, or no /unattend switch and a pass through the setup screens on each clone. Generalize took under 10 minutes and powered the VM off, and every clone made from it afterwards got a fresh SID during its first boot. Our Windows Server template guide for Proxmox ends with a Sysprep step, so templates built that way don’t need this repair.

Every Server Needs the Same Local Account

Without a domain, the nodes trust each other through a local account that has the same name and the same password on every node, plus membership in the local Administrators group. The cluster service uses it to reach the other nodes over WinRM and SMB. stor01 needs the same base setup too: it’s the DNS server everyone points at, and its own copy of the account is what the witness share will authenticate with later. These settings go on all three servers, stor01 first:

  1. a static IP address, with stor01 (10.0.1.30) as the DNS server
  2. the matching local administrator account
  3. LocalAccountTokenFilterPolicy set to 1
  4. the same primary DNS suffix, lab.local
  5. the other servers in the WinRM TrustedHosts list

Item 3 is the one people skip. When a local account other than the built-in Administrator connects over the network, UAC hands it a filtered token without admin rights, and the cluster’s remote calls fail with access denied. Setting LocalAccountTokenFilterPolicy to 1 gives remote local admins their full token, which also means a stolen local admin hash works remotely on these servers, so keep them on a network you control. You can avoid the registry change by using the built-in Administrator account with the same password everywhere, but a named account is easier to audit and to rotate. We used lab.local because the lab network is isolated; on a real network pick a subdomain you own, since .local is also used by mDNS.

Run this block in an elevated PowerShell at each server’s console, not over RDP or WinRM, because its first lines remove the DHCP address and would cut a remote session halfway through. Change the IP per server (node01 10.0.1.21, node02 10.0.1.22, stor01 10.0.1.30), set the gateway to your own, and pick your own password:

New-Item -ItemType Directory -Force C:\lab | Out-Null
$nic = Get-NetAdapter | Where-Object Status -eq 'Up' | Select-Object -First 1
Set-NetIPInterface -InterfaceIndex $nic.ifIndex -Dhcp Disabled
Get-NetIPAddress -InterfaceIndex $nic.ifIndex -AddressFamily IPv4 -ErrorAction SilentlyContinue | Remove-NetIPAddress -Confirm:$false -ErrorAction SilentlyContinue
New-NetIPAddress -InterfaceIndex $nic.ifIndex -IPAddress 10.0.1.21 -PrefixLength 24 -DefaultGateway 10.0.1.1 | Out-Null
Set-DnsClientServerAddress -InterfaceIndex $nic.ifIndex -ServerAddresses 10.0.1.30
Set-DnsClient -InterfaceIndex $nic.ifIndex -RegisterThisConnectionsAddress $true

$pw = ConvertTo-SecureString 'Str0ng!Passw0rd' -AsPlainText -Force
New-LocalUser -Name kenneth -Password $pw -PasswordNeverExpires -FullName 'Cluster admin'
Add-LocalGroupMember -Group Administrators -Member kenneth

New-ItemProperty -Path HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\Policies\System -Name LocalAccountTokenFilterPolicy -Value 1 -PropertyType DWord -Force
Set-ItemProperty -Path HKLM:\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters -Name 'NV Domain' -Value 'lab.local'
Set-Item WSMan:\localhost\Client\TrustedHosts -Value 'node01,node02,stor01,wgclus,*.lab.local' -Force
Enable-NetFirewallRule -DisplayGroup 'File and Printer Sharing'

The NV Domain value is what the GUI’s “Primary DNS suffix of this computer” box writes, and it takes effect after the restart below. PasswordNeverExpires isn’t laziness: if the password expires on one node and gets changed only there, the nodes stop authenticating to each other, so plan password changes for all nodes in the same maintenance window.

On node01 and node02, install Failover Clustering and Hyper-V together, rename the node (node02 on the second one), and restart once for both:

Install-WindowsFeature -Name Failover-Clustering, Hyper-V -IncludeManagementTools
Rename-Computer -NewName node01 -Force
Restart-Computer

Hyper-V is what asks for the restart; Failover Clustering on its own does not need one. The install reported success on both nodes:

Success       : True
RestartNeeded : Yes
FeatureResult : {Failover Clustering, Hyper-V, Hyper-V Module for Windows PowerShell, Hyper-V GUI Management Tools...}
ExitCode      : SuccessRestartRequired
WARNING: You must restart this server to finish the installation process.

Expect the first boot after this to be slow. Hyper-V finishes its setup across two boot phases (“Customizing features for you”), and on our busy Proxmox host one node sat at 0% CPU on that restart until we powered it off and on again. stor01 skips the Hyper-V and clustering install: rename it with Rename-Computer -NewName stor01 -Force and restart it, and its own roles come in the next section.

DNS Replaces the Domain

A workgroup cluster still registers its name in DNS, and so do the nodes, so you need a DNS zone that accepts registrations from machines that have no domain account. On a domain, the zone is AD-integrated and allows only secure dynamic updates, which a workgroup machine cannot perform. That is why stor01 hosts a plain primary zone with nonsecure updates allowed. If your branch site already has a DNS server you control, create the zone there; if you’d rather not allow dynamic updates at all, add static A records for each node and for the cluster name before you run New-Cluster.

On stor01, install the DNS Server and iSCSI Target Server roles in one go. Neither needs a restart:

Install-WindowsFeature -Name DNS, FS-iSCSITarget-Server -IncludeManagementTools

Then create the zone the servers use as their primary DNS suffix, and confirm it is not AD-integrated:

Add-DnsServerPrimaryZone -Name lab.local -ZoneFile lab.local.dns -DynamicUpdate NonsecureAndSecure
Get-DnsServerZone lab.local | Format-Table ZoneName, ZoneType, DynamicUpdate, IsDsIntegrated

The two columns that matter are DynamicUpdate and IsDsIntegrated. NonsecureAndSecure lets the nodes and the cluster register themselves, and False confirms there’s no directory behind the zone:

ZoneName  ZoneType DynamicUpdate      IsDsIntegrated
--------  -------- -------------      --------------
lab.local Primary  NonsecureAndSecure          False

Nonsecure updates mean any machine that can reach stor01 on port 53 can register a name in this zone, so keep the zone on the cluster’s own network and nowhere else.

Shared Disks Come From iSCSI

A Hyper-V cluster needs storage that both nodes can reach, and the workgroup rules do not change that. Microsoft lists Storage Spaces Direct, SAN storage and SMB 3 NAS shares as the options. We used the iSCSI Target Server role on stor01 because it’s built into Windows and behaves like a small SAN: one 12 GB LUN for the VM data and one 1 GB LUN for a disk witness. The iSCSI target and initiator guide covers the role in more depth; these are the commands this lab ran on stor01:

New-Item -ItemType Directory -Force C:\iSCSIVirtualDisks, C:\witness | Out-Null
New-IscsiVirtualDisk -Path C:\iSCSIVirtualDisks\clusdata.vhdx -SizeBytes 12GB | Out-Null
New-IscsiServerTarget -TargetName wgcluster -InitiatorIds 'IQN:iqn.1991-05.com.microsoft:node01.lab.local','IQN:iqn.1991-05.com.microsoft:node02.lab.local' | Out-Null
Add-IscsiVirtualDiskTargetMapping -TargetName wgcluster -Path C:\iSCSIVirtualDisks\clusdata.vhdx
New-IscsiVirtualDisk -Path C:\iSCSIVirtualDisks\witness.vhdx -SizeBytes 1GB | Out-Null
Add-IscsiVirtualDiskTargetMapping -TargetName wgcluster -Path C:\iSCSIVirtualDisks\witness.vhdx
Get-IscsiServerTarget | Select-Object TargetName, Status, InitiatorIds, LunMappings | Format-List

Notice what the initiator IDs contain: node01.lab.local, not just node01. A Windows initiator builds its IQN from the full computer name, primary DNS suffix included, which is why the suffix had to be set before the nodes restarted. NotConnected is expected at this point:

TargetName   : wgcluster
Status       : NotConnected
InitiatorIds : {Iqn:iqn.1991-05.com.microsoft:node01.lab.local, Iqn:iqn.1991-05.com.microsoft:node02.lab.local}
LunMappings  : {TargetName:wgcluster;VHD:"C:\iSCSIVirtualDisks\clusdata.vhdx";LUN:0,
               TargetName:wgcluster;VHD:"C:\iSCSIVirtualDisks\witness.vhdx";LUN:1}

The last job on stor01 is the share for a file share witness, which comes up again in the quorum section. The kenneth account here is stor01’s own local account, created by the same block you ran on the nodes:

New-SmbShare -Name witness -Path C:\witness -FullAccess 'stor01\kenneth'
icacls C:\witness /grant 'kenneth:(OI)(CI)F'

stor01 is done. Nothing on it is clustered, which also makes it the single point the cluster leans on for DNS, disks and the witness, so in production split those jobs across boxes you already trust.

Connect Both Nodes to the Shared Disks

First make sure both nodes registered themselves in stor01’s lab.local zone, because validation, the iSCSI portal and the cluster service all address servers by their full names:

ipconfig /registerdns
Resolve-DnsName node01.lab.local -Server 10.0.1.30 -DnsOnly | Format-Table Name, IPAddress

node02 and stor01 registered on their own; node01 only appeared after ipconfig /registerdns, so run it on any server that is missing from the zone instead of waiting for the next registration cycle:

Name             IPAddress
----             ---------
node01.lab.local 10.0.1.21

Repeat the lookup for node02 from both nodes before you go further.

Each node logs in to the iSCSI target with its own initiator, and the IQN it presents is the one stor01 was told to expect. Start the initiator service, add stor01 as a portal and connect persistently so the session survives a reboot:

Set-Service msiscsi -StartupType Automatic
Start-Service msiscsi
New-IscsiTargetPortal -TargetPortalAddress stor01.lab.local | Out-Null
Get-IscsiTarget | Connect-IscsiTarget -IsPersistent $true | Select-Object TargetNodeAddress, IsConnected

Then check the name the initiator presented and the state of the session:

(Get-InitiatorPort).NodeAddress
Get-IscsiSession | Format-Table TargetNodeAddress, IsConnected, IsPersistent

On node01 the initiator name carried the suffix, which is exactly what the target’s allow list contains, and the session came up persistent. The row of zeros is a second initiator port entry with no IQN, and you can ignore it:

0000000000000000
iqn.1991-05.com.microsoft:node01.lab.local

TargetNodeAddress                                 IsConnected IsPersistent
-----------------                                 ----------- ------------
iqn.1991-05.com.microsoft:stor01-wgcluster-target        True         True

Bring the disks online, partition and format them on one node only. Formatting the same LUN from both nodes at once is the quickest way to corrupt it, and the cluster takes ownership of the disks as soon as it exists anyway. Neither volume gets a drive letter: the cluster manages both disks, and a disk witness must not be used for anything else. The loop labels the 1 GB disk Witness and the larger one ClusData:

foreach ($d in Get-Disk | Where-Object BusType -eq iSCSI) {
  Set-Disk -Number $d.Number -IsOffline $false
  Initialize-Disk -Number $d.Number -PartitionStyle GPT
  if ($d.Size -lt 2GB) {
    New-Partition -DiskNumber $d.Number -UseMaximumSize | Format-Volume -FileSystem NTFS -NewFileSystemLabel Witness | Out-Null
  } else {
    New-Partition -DiskNumber $d.Number -UseMaximumSize | Format-Volume -FileSystem NTFS -NewFileSystemLabel ClusData | Out-Null
  }
}

Run only the connection commands on node02, then list its iSCSI disks:

Get-Disk | Where-Object BusType -eq iSCSI | Format-Table Number, Size, OperationalStatus

Its view of the same two disks reads Offline, which is correct: node01 owns them until the cluster takes over.

Number        Size OperationalStatus
------        ---- -----------------
     1 12884901888 Offline
     2  1073741824 Offline

Leave them offline. Bringing them online on node02 by hand while node01 has them mounted is the same corruption risk as formatting from both sides.

Run Validation as the Shared Account

Every cluster cmdlet from here on has to run as the matching local account (node01\kenneth in this lab), not as SYSTEM and not as an account that exists on only one node. That is the whole authentication model of a workgroup cluster: when node01 calls node02, it presents kenneth with the same password, and node02 accepts it because its own kenneth matches. Sign in to node01 as that account, open an elevated PowerShell and validate:

Test-Cluster -Node node01, node02 -ReportName C:\lab\validate

Validation took 23 minutes on our lab, about 10 of them in the storage tests (disk arbitration and failover across both nodes), on a host that was already I/O-bound. The result was approval with warnings:

System Configuration - Validate Active Directory Configuration: The test reported some warnings..
System Configuration - Validate Software Update Levels: The test reported some warnings..
Network - Validate Network Communication: The test reported some warnings..

Test Result:
HadUnselectedTests, ClusterConditionallyApproved
Testing has completed for the tests you selected. You should review the warnings in the Report.
A cluster solution is supported by Microsoft only if you run all cluster validation tests, and
all tests succeed (with or without warnings).

All three warnings are expected here, and each one tells you something. The Active Directory warning lists both nodes as “Not in a domain” and “Standalone Server”, then explains what that costs you: “Multi-domain and Workgroup clusters (clusters with non-domain joined nodes) introduce higher risk of configuration drift.” It’s the validator admitting that nothing central keeps the nodes alike, so patch levels and settings are your job. The network warning says the nodes reach each other “by only one pair of network interfaces”, checked with UDP on port 3343, which is a real single point of failure in production and fine in a lab. The software update check found “All servers have the same software updates” but flagged both nodes as having updates pending a reboot, which a restart clears. Open validate.htm and read each one rather than trusting the summary line.

New-Cluster Picks the DNS Access Point

Most guides tell you that a workgroup cluster needs -AdministrativeAccessPoint Dns. The New-Cluster reference lists four values for the parameter: ActiveDirectoryAndDns, Dns, ActiveDirectory and None. We first ran New-Cluster without the parameter, expecting it to fail on the missing domain, and it didn’t. It built the cluster, and Get-Cluster afterwards reported AdministrativeAccessPoint : Dns, so on workgroup nodes the cmdlet chooses the DNS access point by itself. We still pass it explicitly, because a script that states the access point says what it intends instead of depending on a default we only observed in one lab. After removing that first cluster, this is the command we kept:

New-Cluster -Name wgclus -Node node01, node02 -AdministrativeAccessPoint Dns -StaticAddress 10.0.1.20

It took 372 seconds and printed only the cluster name. Check what you got:

Get-Cluster | Format-List Name, Domain, AdministrativeAccessPoint
Get-ClusterNode | Format-Table Name, State, Id
Get-ClusterResource | Format-Table Name, State, OwnerGroup, ResourceType
Get-ClusterQuorum | Format-List Cluster, QuorumResource, QuorumType

The Domain field shows lab.local, which is only the primary DNS suffix and not a domain membership. Both nodes are up, every core resource is online, and the cluster already chose a disk witness on its own:

Name                      : wgclus
Domain                    : lab.local
AdministrativeAccessPoint : Dns

Name   State Id
----   ----- --
node01    Up 1
node02    Up 2

Name                         State OwnerGroup         ResourceType
----                         ----- ----------         ------------
Cluster Disk 1              Online Cluster Group      Physical Disk
Cluster Disk 2              Online Available Storage  Physical Disk
Cluster IP Address          Online Cluster Group      IP Address
Cluster Name                Online Cluster Group      Network Name
User Manager                Online User Manager Group User Manager
Virtual Machine Cluster WMI Online Cluster Group      Virtual Machine Cluster WMI

Cluster        : wgclus
QuorumResource : Cluster Disk 1
QuorumType     : Majority

The cluster name resource is where the missing domain shows. Read its private properties and resolve the name against stor01:

Get-ClusterResource 'Cluster Name' | Get-ClusterParameter | Where-Object Name -in 'DnsName','DnsSuffix','ObjectGUID','CreatingDC','StatusKerberos','LastDNSUpdateTime' | Format-Table Name, Value
Resolve-DnsName wgclus.lab.local -Server 10.0.1.30 | Format-Table Name, Type, IPAddress

ObjectGUID and CreatingDC are empty because there is no computer object anywhere, and StatusKerberos is 0. What the name does have is a DNS registration, which landed in stor01’s zone through the nonsecure dynamic update you allowed earlier:

Name              Value
----              -----
DnsName           wgclus
StatusKerberos    0
CreatingDC
LastDNSUpdateTime 10/30/2026 6:23:04 PM
ObjectGUID
DnsSuffix         lab.local

Name             Type IPAddress
----             ---- ---------
wgclus.lab.local    A 10.0.1.20

Run the same three checks from the second node as well. A cluster that answers identically from node02, signed in as node02’s own kenneth, proves the shared-account trust works in both directions:

PowerShell on node02 showing Windows Server 2025 workgroup cluster wgclus with AdministrativeAccessPoint Dns, both nodes Up and the wgclus.lab.local A record resolving to 10.0.1.20

If node02 instead returns access denied or “The remote server has been paused or is in the process of being started”, give the cluster service a minute after creation and check the account and password on that node before anything else.

The Witness Needs Its Own Credential

Why bother with a witness on two nodes? Without one, dynamic quorum lets the cluster survive only some single-node failures, depending on which node holds the vote. New-Cluster already made one of the shared disks a disk witness (Cluster Disk 1 above). Microsoft’s workgroup cluster page lists three witness types:

  • disk witness on shared storage
  • cloud witness in an Azure storage account
  • USB witness on a router or NAS share

A disk witness is the simplest of the three when you already have shared storage, because there is no credential involved at all. You only need something else when the shared storage is the very thing you are trying to survive losing, or when the cluster has no shared disks.

A file share witness is where workgroup clusters trip people up. On a domain, the cluster grants its own computer account access to the share. A workgroup cluster has no computer account, so the plain command fails:

Set-ClusterQuorum -FileShareWitness \\stor01.lab.local\witness

The error names the share and says nothing about why:

Could not grant the cluster access to the file share '\\stor01.lab.local\witness'.

The fix is to hand the cluster a credential that’s valid on the server hosting the share. Microsoft’s quorum witness guide supports a file share witness on a non-domain device as long as the cluster runs Windows Server 2019 or later. Here the credential is stor01’s own local kenneth account, which is a different account from the kenneth on the nodes even though the names match:

$pw = ConvertTo-SecureString 'Str0ng!Passw0rd' -AsPlainText -Force
$cred = New-Object System.Management.Automation.PSCredential('stor01\kenneth', $pw)
Set-ClusterQuorum -FileShareWitness \\stor01.lab.local\witness -Credential $cred

This time the witness came online, owned by node02. Its parameters show where the credential went: the cluster stores the user name in the resource and connects as that account instead of impersonating a cluster name object, which is what ImpersonateCNO 0 means:

Cluster        : wgclus
QuorumResource : File Share Witness
QuorumType     : Majority

Name                State OwnerNode
----                ----- ---------
File Share Witness Online node02

Name             Value
----             -----
SharePath        \\stor01.lab.local\witness
ArbitrationDelay 6
UserName         stor01\kenneth
Password
ImpersonateCNO   0

On stor01 the cluster created a folder named after its ID with a 512-byte Witness.log inside, and that is all a file share witness ever stores. Treat stor01\kenneth as a service account from now on. If its password changes on stor01 and nobody re-runs Set-ClusterQuorum with the new one, the witness goes offline the next time it needs to reconnect, and you won’t notice until a node fails and the cluster has one vote too few.

Hyper-V Live Migration Skips Kerberos

Live migration is the part that is new in Windows Server 2025. Between domain-joined Hyper-V hosts, live migration authenticates with Kerberos or CredSSP, and a workgroup has neither, which is why Hyper-V wasn’t a supported workgroup cluster workload before 2025. Microsoft’s workgroup live migration article says the 2025 cluster uses self-signed PKU2U certificates between nodes instead, with the matching local accounts as the identity. You configure nothing for it beyond the usual Hyper-V steps.

Put the disk left in Available Storage (Cluster Disk 2 here) to work as a Cluster Shared Volume, so both nodes can open the VM’s files at the same time. In our lab this step ran while Cluster Disk 1 was still the disk witness, so Cluster Disk 2 was the only candidate. If you already switched to the file share witness, both disks sit in Available Storage; pick the larger one by its Capacity column in Failover Cluster Manager under Storage, Disks:

Add-ClusterSharedVolume -Name 'Cluster Disk 2'
Get-ClusterSharedVolume | Format-Table Name, State, OwnerNode

The CSV mounts under C:\ClusterStorage\Volume1 on both nodes:

Name            State OwnerNode
----            ----- ---------
Cluster Disk 2 Online node01

Next, create a VM on the CSV, start it and make it a clustered role. We kept the test VM deliberately small, a Generation 2 VM with 512 MB of RAM and an empty 2 GB disk, because the point is how the cluster moves it, not what runs inside it:

$csv = 'C:\ClusterStorage\Volume1'
New-Item -ItemType Directory -Force "$csv\vm01" | Out-Null
New-VHD -Path "$csv\vm01\vm01.vhdx" -SizeBytes 2GB -Dynamic | Out-Null
New-VM -Name vm01 -Generation 2 -MemoryStartupBytes 512MB -VHDPath "$csv\vm01\vm01.vhdx" -Path "$csv\vm01" | Out-Null
Start-VM vm01
Add-ClusterVirtualMachineRole -VirtualMachine vm01

Then turn on live migration on both hosts from node01. Invoke-Command reaching node02 by its short name is itself a check that the TrustedHosts list and the matching account work:

foreach ($n in 'node01', 'node02') {
  Invoke-Command -ComputerName $n -ScriptBlock { Enable-VMMigration; Set-VMHost -MaximumVirtualMachineMigrations 2 -MaximumStorageMigrations 2 }
}

Now move the running VM to node02 and time the whole operation:

Measure-Command { Move-ClusterVirtualMachineRole -Name vm01 -Node node02 -MigrationType Live }
Get-ClusterGroup vm01 | Format-Table Name, OwnerNode, State

The move took 23 seconds from the command to the prompt, and the role ended up online on node02:

Name OwnerNode  State
---- ---------  -----
vm01 node02    Online

Is 23 seconds the downtime? No. The number that matters is in the Hyper-V VMMS admin log on the source node, which records how long the VM was actually paused:

Get-WinEvent -LogName Microsoft-Windows-Hyper-V-VMMS-Admin -MaxEvents 15 | Where-Object Message -match 'migrat' | Format-List TimeCreated, Id, Message

Event 20413 marks the start and 20415 the finish, 15 seconds apart, and 20415 carries the blackout time: 4.3 seconds for a 512 MB VM over a single 10.0.1.0/24 network that also carried cluster heartbeats and iSCSI. The pause depends on how much memory the VM changes during the copy and on the network carrying it, so measure your own VMs rather than borrowing this number:

TimeCreated : 10/30/2026 6:44:31 PM
Id          : 20415
Message     : The Virtual Machine Management service successfully completed the live migration of
              virtual machine  'vm01' with a blackout time of 4.3 seconds (VMID
              F0581432-264E-4C88-A77E-0BB2C5AC655D).

TimeCreated : 10/30/2026 6:44:16 PM
Id          : 20413
Message     : The Virtual Machine Management service initiated the live migration of virtual
              machine  'vm01' to destination host 'node02' (VMID
              F0581432-264E-4C88-A77E-0BB2C5AC655D).

Nothing in that exchange touched a domain controller, because there is not one. If live migration fails on your cluster with an authentication error, check the boring things first: the same password on both nodes, LocalAccountTokenFilterPolicy on both, and whether you ran the move as the shared account rather than as a different local admin.

What Breaks Without a Domain

The domain-joined cluster in our step-by-step failover cluster guide gets several things for free that this one doesn’t. Some of these we hit in the lab; the rest come from Microsoft’s documentation and are marked that way, so you know which is which:

Domain featureOn a workgroup clusterSource
Cluster name object in ADNone. ObjectGUID and CreatingDC are empty, StatusKerberos is 0Lab
Secure DNS registrationNeeds a zone that accepts nonsecure updates, or static A recordsLab
File share witness via the cluster accountFails with “Could not grant the cluster access”; works with -Credential for an account on the share hostLab
One admin identity everywhereA local account with the same password on every node; a password change is a change on every nodeLab
Group Policy keeping nodes alikeNothing does it; validation warns about “higher risk of configuration drift”Lab
Clustered file serverNot supported, citing authentication issuesMicrosoft docs
SQL Server failover cluster instanceNot supported; use a domain-independent availability groupMicrosoft docs
Cluster-Aware UpdatingIts requirements say all nodes must be in the same AD domain, so plan patching per nodeMicrosoft docs

Read down the Source column and a pattern shows. Everything the lab hit is a place where a domain would have supplied an identity, and you supply it by hand instead: an account, a DNS record, a credential for the witness share. The items from the documentation are the ones that need clients or tools to authenticate to the cluster itself, and a workgroup can’t offer that. The Cluster-Aware Updating row matters most day to day, since its requirements state that “All cluster nodes must be in the same Active Directory domain.” On a workgroup cluster you patch one node at a time yourself: drain it, update it, restart it, resume it, then the next one.

That makes the decision short. If the site already has a domain controller you trust, join the nodes and use a domain cluster, because you get Kerberos, CAU and clustered file shares. If the whole point is two Hyper-V hosts somewhere with no domain and no plan to add one, the Windows Server 2025 workgroup cluster now covers that case properly, live migration included, as long as you treat the shared account, the DNS zone and the witness credential as the three things that keep it together. We drove everything here from PowerShell on node01; if you’d rather manage the pair from a browser, Windows Admin Center is the usual tool, though we didn’t connect it to this cluster.

Keep reading

Create a Bootable Windows 11 USB and Install Windows 11 Windows Create a Bootable Windows 11 USB and Install Windows 11 Set Up a Samba Share on Windows 11 and Access It from Linux Networking Set Up a Samba Share on Windows 11 and Access It from Linux Create Disk Partitions in Windows Using diskpart Windows Create Disk Partitions in Windows Using diskpart Windows LAPS Setup with Group Policy on Windows Server 2025 Security Windows LAPS Setup with Group Policy on Windows Server 2025 Windows Run Commands Cheat Sheet: .msc, .cpl, ms-settings (PDF) Cheat Sheets Windows Run Commands Cheat Sheet: .msc, .cpl, ms-settings (PDF) Install WSUS on Windows Server 2025 Windows Install WSUS on Windows Server 2025

Leave a Comment

Press ESC to close