🧪 Why a nested lab?
You want to learn Azure Local. You don't have validated servers sitting in a rack. So what do you do?
You build it inside a Hyper-V VM. One host, one domain controller VM, one node VM. It's slow, it's not supported for production, but it's the cheapest way to walk through the real deployment: Arc registration, the portal wizard, validation, Network ATC, the lot.
I've written guides like this before. The first version described the documented path. Then I ran it end to end, and reality had other ideas. This article is what actually happened: every error I hit, why it happened, and the fix that worked. The fixes are built into the steps, so if you follow them in order you shouldn't hit them yourself.
Lab and training use only. Microsoft doesn't support nested Azure Local for production, and it runs much slower than physical hardware. Use this to learn, to demo and to train, not to host workloads.
🗺️ What we're building
Here's the whole lab on one page. Keep it open in another tab; every IP address in the scripts below comes from it.
The deployment runs in phases, and each phase happens in a specific place. That's worth knowing up front, because a lot of the errors I hit came from running a command in the wrong place.
| Phase | What you do | Where |
|---|---|---|
| 0. Tools | Az modules and AD tools | Hyper-V host |
| 1. Azure | Resource providers, resource group, service principal + roles | Hyper-V host |
| 2. AD + time | OU, deployment account, NTP, DNS sanity check | DC, via PowerShell Direct |
| 3. VM | Create the node VM | Hyper-V host |
| 4. OS | Install from the ISO, set the local admin password | VM console |
| 5. Network | NIC names, static IP, time zone, NTP, rename, RDP | Hyper-V host (PowerShell Direct) |
| 6. Arc | Register with Arc; the node updates its OS | Interactive session on the node |
| 7. Portal | Extensions, validation, deployment | Azure portal |
| 8. Fix-ups | DNS fix if "Create the cluster" fails | Hyper-V host |
| 9. After | Update, clean up secrets | Portal + host |
What you need
| Component | Requirement | In my lab |
|---|---|---|
| Host OS | Windows Server 2022/2025 or Windows 11 with Hyper-V; CPU with nested virtualization | Windows Server 2025 |
| RAM | 48 GB free (the VM uses 32 GB static) | 32 GB to the node |
| CPU | 8 free logical processors (4 minimum) | 8 vCPUs |
| Disk | ~130 GB for the OS disk + two 1 TB dynamic data disks | OS on C:, data on E:\DataDisk |
| Network | External vSwitch on a 1 Gbps or faster link | Broadcom NetXtreme Gigabit |
| Azure | A subscription where you're Owner (a Visual Studio subscription works) and a region that supports Azure Local | westeurope |
| AD | A DC VM on the same host | MCT-DC, MCT.LAB, 192.168.100.80 |
| ISO | Azure Local 24H2 | AzureLocal24H2.26100.32230.LCM.12.2608 |
One more thing before we start. Every script in this article uses the same variables, so paste this block at the top of every new PowerShell window:
$Tenant = '<TENANT-ID>'
$Sub = '<SUBSCRIPTION-ID>'
$RG = 'AZL-MCT-RG'
$Region = 'westeurope'
$SPName = 'AzL-Arc-MCT-2610' # unique per deployment
$DomainDN = 'DC=MCT,DC=LAB'
$OU = "OU=AzureLocal,$DomainDN"
$DCVM = 'MCT-DC'
$DCIP = '192.168.100.80'
$NodeVM = 'AZLNode1'
$NodeIP = '192.168.100.50'
$Gateway = '192.168.100.1'
🔍 Three things to check before you touch PowerShell
Most guides start with "install the modules". I'd rather you spend five minutes here, because these three things caused the most pain later on.
1. Does your router hand out IPv6 DNS?
Why would a router matter? Because many home and office routers advertise themselves as an IPv6 DNS server. Windows then asks the router and your DC. The router answers "name does not exist" for your lab domain, Windows believes it, and the node loses the domain halfway through the deployment.
Check it from any Windows machine on the LAN:
Get-DnsClientServerAddress -AddressFamily IPv6 |
Where-Object ServerAddresses | Select-Object InterfaceAlias, ServerAddresses
# Anything like fe80::1 here = your router hands out IPv6 DNS
fe80::1, that's the router offering itself as a DNS server. The fec0:0:0:ffff::1–3 entries on the other adapters are Windows' built-in defaults when nothing is configured; you can ignore them.If you see something like fe80::1, turn off IPv6 DNS advertisement (RDNSS/DHCPv6) on the router now. Can't change the router? No problem, I'll show you the tested fix later.
2. What's your host's link speed?
Mine had negotiated 100 Mbps instead of 1 Gbps. Everything worked, just very, very slowly. Check the cable and switch port before you start downloading several gigabytes of updates.
3. Will your passwords pass the Azure Local rule?
Here's the trap. Active Directory accepted R00t123456789: 13 characters, no special character. Azure Local would have rejected it later in the deployment. The default domain policy is much weaker than the Azure Local rule.
| Account | Rule |
|---|---|
| Deployment account (AzLDeployUser) | 14+ characters with a lowercase letter, an uppercase letter, a number and a special character. Username: 1–20 characters (letters, numbers, - and _), can't start with a number or -, can't be admin, must differ from the local admin name. |
| Node local Administrator | The same rule |
| Practical advice | Avoid " ' $ & < > ; and the backtick, which break scripts and JSON. Turn on PasswordNeverExpires in a lab. |
So I wrote a small checker. Paste it into your host session; the AD step uses it.
function Test-AzLPassword ([securestring]$Secure) {
$p = [System.Net.NetworkCredential]::new('', $Secure).Password
$issues = @()
if ($p.Length -lt 14) { $issues += '14+ characters' }
if ($p -cnotmatch '[a-z]') { $issues += 'a lowercase letter' }
if ($p -cnotmatch '[A-Z]') { $issues += 'an uppercase letter' }
if ($p -notmatch '\d') { $issues += 'a number' }
if ($p -notmatch '[^a-zA-Z0-9]') { $issues += 'a special character' }
if ($p -match '["''$&<>;]' -or $p.Contains([string][char]96)) { $issues += 'none of " '' $ & < > ; or backtick' }
$issues
}
PHASE 0–1Azure preparation
First, the tools. Run this once in an elevated PowerShell on the host:
[Net.ServicePointManager]::SecurityProtocol = [Net.SecurityProtocolType]::Tls12
Install-Module Az -Scope AllUsers -Force -AllowClobber # includes Az.Accounts + Az.Resources
Import-Module Az.Accounts, Az.Resources
Get-Command Register-AzResourceProvider, New-AzADServicePrincipal | Select-Object Name, Source
What went wrong: I first had only Az.Accounts installed. Connect-AzAccount worked fine, but every *-AzResourceProvider command failed with "not recognized". If you're short on space, Install-Module Az.Resources is enough. PowerShell 7 and Windows PowerShell 5.1 both work, but each needs its own modules.
Register the 13 resource providers
This loop registers them all and waits until every one says Registered:
Connect-AzAccount -TenantId $Tenant -SubscriptionId $Sub
$providers = @(
'Microsoft.HybridCompute','Microsoft.GuestConfiguration','Microsoft.HybridConnectivity',
'Microsoft.AzureStackHCI','Microsoft.Kubernetes','Microsoft.KubernetesConfiguration',
'Microsoft.ExtendedLocation','Microsoft.ResourceConnector','Microsoft.HybridContainerService',
'Microsoft.Attestation','Microsoft.Storage','Microsoft.Insights','Microsoft.KeyVault'
)
foreach ($p in $providers) { Register-AzResourceProvider -ProviderNamespace $p | Out-Null }
# Wait until all 13 say Registered
do {
$pending = foreach ($p in $providers) {
$s = (Get-AzResourceProvider -ProviderNamespace $p | Select-Object -First 1).RegistrationState
if ($s -ne 'Registered') { "$p : $s" }
}
if ($pending) { Write-Host "Waiting on:"; $pending; Start-Sleep 30 }
} while ($pending)
Write-Host 'All 13 providers Registered' -ForegroundColor Green
A fresh resource group, every time
Use a new resource group for each attempt. I left my earlier group, with all its leftovers, and started clean in AZL-MCT-RG. Much easier than hunting for orphaned resources.
New-AzResourceGroup -Name $RG -Location $Region -Force | Out-Null
$Scope = (Get-AzResourceGroup -Name $RG).ResourceId
A uniquely named service principal
Why a service principal at all? Because a personal Microsoft account can't reliably sign in with device code on the node. The SP gets an access token on the host, and we pass that token to the node.
if (Get-AzADServicePrincipal -DisplayName $SPName) { throw "SP $SPName already exists - choose a new name" }
$sp = New-AzADServicePrincipal -DisplayName $SPName
$spAppId = $sp.AppId
$spSecret = $sp.PasswordCredentials.SecretText
Start-Sleep -Seconds 45 # Entra ID replication
$roles = @(
'Azure Connected Machine Onboarding','Azure Connected Machine Resource Administrator',
'Azure Stack HCI Administrator','Reader',
'Key Vault Data Access Administrator','Key Vault Secrets Officer',
'Key Vault Contributor','Storage Account Contributor'
)
foreach ($r in $roles) {
try {
New-AzRoleAssignment -ApplicationId $spAppId -RoleDefinitionName $r -Scope $Scope -ErrorAction Stop | Out-Null
Write-Host "Assigned: $r"
} catch {
if ($_.Exception.Message -match 'RoleAssignmentExists') { Write-Host "Already there: $r" }
else { Write-Host "FAILED: $r - $($_.Exception.Message)" -ForegroundColor Red }
}
}
Get-AzRoleAssignment -ServicePrincipalName $spAppId -Scope $Scope | Select-Object RoleDefinitionName
Write-Host "AppId : $spAppId"
Write-Host "Secret: $spSecret" # copy into a password manager - shown only once
Keep the secret in a password manager only. I pasted one into a chat and then had to plan to delete the SP straight after deployment. If you lose it, create a new one with New-AzADAppCredential. A "RoleAssignmentExists" conflict is harmless: the role was already there.
And why "unique"? Earlier attempts had left five apps all called AzureLocalArcSP. So $app.AppId returned five IDs, and the next command failed with Cannot convert System.Object[] to System.Guid. If that's you, clean up first:
Get-AzADApplication -DisplayName 'AzureLocalArcSP' | ForEach-Object {
Get-AzRoleAssignment -ServicePrincipalName $_.AppId -ErrorAction SilentlyContinue |
Remove-AzRoleAssignment -ErrorAction SilentlyContinue | Out-Null
Remove-AzADApplication -ObjectId $_.Id
Write-Host "Removed $($_.AppId)"
}
PHASE 2Active Directory and time, on the DC
Azure Local needs an OU and a deployment account, created by Microsoft's AsHciADArtifactsPreCreationTool module. I ran it on the host first. It failed: "The specified module 'ActiveDirectory' was not loaded".
Of course. The tool needs the ActiveDirectory and GroupPolicy modules, and those live on the DC. So we run it there, without RDP, using PowerShell Direct from the host. Notice how the loop refuses to continue until the password passes the checker:
$daCred = Get-Credential -UserName 'MCT\Administrator' -Message 'Domain admin'
do {
$lcmCred = Get-Credential -UserName 'AzLDeployUser' -Message 'Password for the deployment account'
$problems = Test-AzLPassword $lcmCred.Password
if ($problems) { Write-Host "Password needs: $($problems -join ', ')" -ForegroundColor Yellow }
} while ($problems)
Invoke-Command -VMName $DCVM -Credential $daCred -ArgumentList $lcmCred, $OU -ScriptBlock {
param($lcmCred, $OU)
[Net.ServicePointManager]::SecurityProtocol = [Net.SecurityProtocolType]::Tls12
Install-Module AsHciADArtifactsPreCreationTool -Repository PSGallery -Force
New-HciAdObjectsPreCreation -AzureStackLCMUserCredential $lcmCred -AsHciOUName $OU
Set-ADUser -Identity 'AzLDeployUser' -PasswordNeverExpires $true
Get-ADUser 'AzLDeployUser' -Properties PasswordNeverExpires | Select-Object SamAccountName, Enabled, PasswordNeverExpires, DistinguishedName
}
You should see it verify DC=MCT,DC=LAB, create the AzureLocal OU and AzLDeployUser, grant permissions and block GPO inheritance. Note that the tool creates the account; to change its password later, use Set-ADAccountPassword on the DC.
Time and DNS on the DC
Time matters more than you'd think. My DC was still on Pacific time, which made every log confusing. Point it at a real NTP source and turn off Hyper-V time sync:
Invoke-Command -VMName $DCVM -Credential $daCred -ScriptBlock {
Set-TimeZone -Id 'Morocco Standard Time'
w32tm /config /manualpeerlist:"pool.ntp.org" /syncfromflags:manual /reliable:YES /update
Restart-Service w32time
w32tm /resync /force
w32tm /query /status | Select-String 'Source|Stratum|Last Successful'
}
Disable-VMIntegrationService -VMName $DCVM -Name 'Time Synchronization'
Expect Source: pool.ntp.org and stratum 2 or 3. Then two quick DNS checks on the DC:
- Forwarders: there must be at least one, or the node can't resolve Azure names. Add one with
Add-DnsServerForwarder -IPAddress 8.8.8.8. - DNS client: the DC should list only itself. If you see
fe80::1, the router is handing out IPv6 DNS to the DC too.
PHASE 3Create the node VM
Now the node itself. A few settings here aren't optional for nested Azure Local, so let's look at them before the script:
- Nested virtualization (
-ExposeVirtualizationExtensions): without it, nothing inside the node can run VMs. - Static memory, 32 GB. Dynamic memory and Azure Local don't mix.
- vTPM: the Recommended security settings need it.
- Two NICs with MAC spoofing, teaming and VLAN trunking allowed: Network ATC needs them.
- Time sync off and checkpoints off.
$VMPath = "C:\HYPERVMs\$NodeVM" # VM config + OS disk
$DataPath = 'E:\DataDisk' # data disks for Storage Spaces Direct
$SwitchName = 'Broadcom NetXtreme Gigabit Ethernet #3 - Virtual Switch'
$vCPU = 8
$DataDiskSize = 1TB
$ISOPath = 'C:\Users\Administrator\Downloads\ISO\AzureLocal24H2.26100.32230.LCM.12.2608.0.3020.x64.en-us.iso'
if (-not (Test-Path $ISOPath)) { throw "ISO not found: $ISOPath" }
if (-not (Get-VMSwitch -Name $SwitchName -ErrorAction SilentlyContinue)) { throw "vSwitch not found" }
if (Get-VM -Name $NodeVM -ErrorAction SilentlyContinue) { throw "$NodeVM already exists" }
New-Item -Path $VMPath, $DataPath -ItemType Directory -Force | Out-Null
New-VM -Name $NodeVM -Generation 2 -MemoryStartupBytes 32GB -Path $VMPath -NoVHD -SwitchName $SwitchName
New-VHD -Path "$VMPath\$NodeVM-OS.vhdx" -SizeBytes 127GB -Dynamic | Out-Null
Add-VMHardDiskDrive -VMName $NodeVM -Path "$VMPath\$NodeVM-OS.vhdx"
foreach ($i in 1..2) {
$d = Join-Path $DataPath "$NodeVM-S2D$i.vhdx"
New-VHD -Path $d -SizeBytes $DataDiskSize -Dynamic | Out-Null
Add-VMHardDiskDrive -VMName $NodeVM -Path $d
}
Add-VMDvdDrive -VMName $NodeVM -Path $ISOPath
Set-VMFirmware -VMName $NodeVM -FirstBootDevice (Get-VMDvdDrive -VMName $NodeVM)
Set-VMMemory -VMName $NodeVM -DynamicMemoryEnabled $false
Set-VM -VMName $NodeVM -CheckpointType Disabled -AutomaticCheckpointsEnabled $false
Set-VMProcessor -VMName $NodeVM -Count $vCPU -ExposeVirtualizationExtensions $true
Set-VMKeyProtector -VMName $NodeVM -NewLocalKeyProtector
Enable-VMTPM -VMName $NodeVM
Get-VMNetworkAdapter -VMName $NodeVM | Remove-VMNetworkAdapter
Add-VMNetworkAdapter -VMName $NodeVM -Name 'NIC1' -SwitchName $SwitchName
Add-VMNetworkAdapter -VMName $NodeVM -Name 'NIC2' -SwitchName $SwitchName
Get-VMNetworkAdapter -VMName $NodeVM | Set-VMNetworkAdapter -MacAddressSpoofing On -AllowTeaming On
Get-VMNetworkAdapter -VMName $NodeVM | Set-VMNetworkAdapterVlan -Trunk -NativeVlanId 0 -AllowedVlanIdList '0-1000'
Disable-VMIntegrationService -VMName $NodeVM -Name 'Time Synchronization'
Get-VMHardDiskDrive -VMName $NodeVM | Select-Object ControllerLocation, Path
Before the first boot, verify it:
[pscustomobject]@{
vCPU = (Get-VMProcessor $NodeVM).Count
Nested = (Get-VMProcessor $NodeVM).ExposeVirtualizationExtensions
MemoryGB = (Get-VMMemory $NodeVM).Startup/1GB
DynamicMem = (Get-VMMemory $NodeVM).DynamicMemoryEnabled
vTPM = (Get-VMSecurity $NodeVM).TpmEnabled
SecureBoot = (Get-VMFirmware $NodeVM).SecureBoot
TimeSync = (Get-VMIntegrationService $NodeVM -Name 'Time Synchronization').Enabled
Checkpoint = (Get-VM $NodeVM).CheckpointType
}
Expected: 8, True, 32, False, True, On, False, Disabled.
Prefer to check it visually? Here's the same VM in Hyper-V Manager:
PHASE 4–5Install the OS, then network, time and name
Boot from the ISO (press a key as soon as the prompt appears), choose Custom, and install on the 127 GB disk. Leave the two 1 TB disks alone. Set the local Administrator password using the rule above, and type it carefully: a different keyboard layout in the VM console can quietly swap characters like @, ", y/z or #.
Don't join the domain and don't install updates. The Arc bootstrap takes care of the OS.
Use .\Administrator, with the dot
Everything from here runs from the host through PowerShell Direct. Use .\Administrator. A bare Administrator plus a mistyped password gave me "The credential is invalid" more times than I'd like to admit. When it happens, sign in on the VM console to check the password before trying again.
$nodeCred = Get-Credential -UserName '.\Administrator' -Message "$NodeVM local admin"
Invoke-Command -VMName $NodeVM -Credential $nodeCred -ScriptBlock { whoami } # expect <name>\administrator
$mac1 = (Get-VMNetworkAdapter -VMName $NodeVM -Name 'NIC1').MacAddress -replace '(..)(?!$)', '$1-'
$mac2 = (Get-VMNetworkAdapter -VMName $NodeVM -Name 'NIC2').MacAddress -replace '(..)(?!$)', '$1-'
Invoke-Command -VMName $NodeVM -Credential $nodeCred -ArgumentList $mac1,$mac2,$NodeIP,$Gateway,$DCIP -ScriptBlock {
param($mac1, $mac2, $ip, $gw, $dns)
Get-NetAdapter | Where-Object MacAddress -eq $mac1 | Rename-NetAdapter -NewName 'NIC1'
Get-NetAdapter | Where-Object MacAddress -eq $mac2 | Rename-NetAdapter -NewName 'NIC2'
Start-Sleep 2
Set-NetIPInterface -InterfaceAlias 'NIC1' -Dhcp Disabled
Get-NetIPAddress -InterfaceAlias 'NIC1' -AddressFamily IPv4 -ErrorAction SilentlyContinue | Remove-NetIPAddress -Confirm:$false
Get-NetRoute -InterfaceAlias 'NIC1' -AddressFamily IPv4 -ErrorAction SilentlyContinue | Remove-NetRoute -Confirm:$false
New-NetIPAddress -InterfaceAlias 'NIC1' -IPAddress $ip -PrefixLength 24 -DefaultGateway $gw | Out-Null
Set-DnsClientServerAddress -InterfaceAlias 'NIC1' -ServerAddresses $dns
Set-NetIPInterface -InterfaceAlias 'NIC2' -Dhcp Disabled # storage: no IP, no gateway
Set-TimeZone -Id 'Morocco Standard Time'
w32tm /config /manualpeerlist:"$dns" /syncfromflags:manual /update | Out-Null
Restart-Service w32time; Start-Sleep 5; w32tm /resync /force | Out-Null
Get-NetIPConfiguration -InterfaceAlias 'NIC1' | Select-Object InterfaceAlias, IPv4Address, IPv4DefaultGateway, DNSServer
w32tm /query /status | Select-String 'Source'
Get-TimeZone | Select-Object Id
}
Expect Source: 192.168.100.80 and Morocco Standard Time (or your own zone). The image defaults to Pacific, so check.
Connectivity, free IPs and link speed
Invoke-Command -VMName $NodeVM -Credential $nodeCred -ScriptBlock {
[pscustomobject]@{
DC_LDAP = Test-NetConnection 192.168.100.80 -Port 389 -InformationLevel Quiet
DNS = [bool](Resolve-DnsName management.azure.com -ErrorAction SilentlyContinue)
ARM_443 = Test-NetConnection management.azure.com -Port 443 -InformationLevel Quiet
EntraID_443 = Test-NetConnection login.microsoftonline.com -Port 443 -InformationLevel Quiet
}
51..59 | ForEach-Object { "192.168.100.$_ in use: " + (Test-Connection "192.168.100.$_" -Count 1 -Quiet) }
Get-NetAdapter | Select-Object Name, LinkSpeed
}
All four connectivity tests must be True, and every address from .51 to .59 must be False (free): the deployment needs that range for infrastructure.
Rename before Arc. Not after.
Invoke-Command -VMName $NodeVM -Credential $nodeCred -ScriptBlock {
Rename-Computer -NewName 'AZLNode1' -Force
Set-ItemProperty 'HKLM:\System\CurrentControlSet\Control\Terminal Server' -Name fDenyTSConnections -Value 0
Enable-NetFirewallRule -DisplayGroup 'Remote Desktop'
}
Restart-VM -Name $NodeVM -Force
Rename before Arc registration. The Arc resource takes the computer name, and the only clean way to change it afterwards is to reinstall.
PHASE 6Register with Azure Arc
This is where I made my most "reasonable" mistake. I wanted the latest release, so I forced it:
| Attempt | Result |
|---|---|
TargetSolutionVersion = 12.2610.1004.50 (the latest) | Rejected: "Given TargetSolutionVersion … is not valid". The zero-day update only offers certain versions to this ISO. |
| No TargetSolutionVersion, interactive session | Succeeded. The bootstrap updated the node to the ISO's release: 2608, OS 26100.33296. |
Leave TargetSolutionVersion out. Deploy the version the ISO gives you, then update from the portal.
And why interactive? So you can answer the version prompt if it appears. Here's the flow: get a token on the host with the SP, push it into a session on the node, then enter that session.
$spCred = Get-Credential -UserName $spAppId -Message 'Paste the SP secret as the password'
Connect-AzAccount -ServicePrincipal -Credential $spCred -TenantId $Tenant -SubscriptionId $Sub | Out-Null
$tok = Get-AzAccessToken -ResourceUrl 'https://management.azure.com/'
$ArmAccessToken = [System.Net.NetworkCredential]::new('', $tok.Token).Password # SecureString -> text
$s = New-PSSession -VMName $NodeVM -Credential $nodeCred
Invoke-Command -Session $s -ArgumentList $ArmAccessToken,$Tenant,$Sub,$RG,$Region -ScriptBlock {
param($t, $ten, $sub, $rg, $reg)
$global:params = @{ TenantId=$ten; SubscriptionID=$sub; ResourceGroup=$rg; Region=$reg; Cloud='AzureCloud'; ArmAccessToken=$t }
}
Enter-PSSession $s
The prompt changes to [AZLNode1]: PS>. Now run:
Invoke-AzStackHciArcInitialization @params
If the OS needs updating, this takes 40–45 minutes and the node reboots. After the reboot, get a new token and run the session step again. Success looks like ArcRegistration : Succeeded and Bootstrap succeeded:
(Get-ArcBootstrapStatus).Response.Status
(Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion').UBR # 33296 = 2608
& "$env:ProgramFiles\AzureConnectedMachineAgent\azcmagent.exe" show | Select-String 'Agent Status|Resource Group|Location'
Exit-PSSession
azcmagent show on my node: Agent Status Connected, westeurope, cloud provider AzSHCI, and every dependent service running. I've redacted the IDs.Two traps: (1) Get-ArcZeroDaySolutionUpdateConfig asks for a ConfigFilePath and waits forever. Press Ctrl+C, you don't need it. (2) Inside [AZLNode1], running Invoke-Command -VMName AZLNode1 fails with "does not resolve to a single virtual machine". Of course it does: you're already on the node. Run commands directly, or Exit-PSSession first.
PHASE 7Deploy from the portal
Go to Azure Local → All systems → + Create. Pick your subscription, AZL-MCT-RG, an instance name (for example AZLCluster1) and West Europe.
Machines and extensions
Click + Add machines and pick AZLNode1. It might say Not validated or Missing Arc extensions. Don't panic, that's normal. The portal installs the Arc extensions itself; in a nested VM it takes 10–20 minutes. Refresh now and then until the machine shows Ready, then Validate selected machines and create a new key vault.
Networking: turn RDMA off
| Setting | Value |
|---|---|
| Pattern | Group management and compute traffic |
| Management + compute intent | NIC1 |
| Storage intent | NIC2, VLAN 711 |
| Customize network settings (every intent) | RDMA protocol: Disabled. Jumbo packet: 1514 |
| IP allocation | Manual |
| Infra range | 192.168.100.51 – 192.168.100.59 |
| Mask / gateway / DNS | 255.255.255.0 / 192.168.100.1 / 192.168.100.80 |
What happens if you leave RDMA on? Validation fails twice. Once on RDMA operational, because a nested VM simply can't do RDMA. And once on symmetry and bandwidth, because with RDMA on, the storage NIC needs 10 Gbps. Disabling RDMA on each intent fixes both.
Management, security, advanced
| Field | Value |
|---|---|
| Custom location | e.g. AZLCluster1-cl |
| Witness | None (single node) |
| Domain / OU | MCT.LAB / OU=AzureLocal,DC=MCT,DC=LAB |
| Deployment account | AzLDeployUser (no MCT\ prefix) and its password |
| Local administrator | Administrator and the node password |
| Security | Recommended (needs the vTPM) |
| Advanced | Defaults |
Validation: when in doubt, run it again
Validation takes about 15–30 minutes. Mine failed with Azure Local Network Infra Connection: "The parameter is incorrect".
So I checked the node carefully: no leftover switches, no IP conflicts, no IPv6 problems. The same command succeeded in two isolated repeats. It was a one-off timing failure in the nested VM. I ran validation again, without changing anything, and it passed.
Rule of thumb: on a nested lab, re-run a failed validation once before you start "fixing" things. You can easily break a working configuration chasing a timing glitch.
Then Review + create → Create. A single physical node takes 1.5–2 hours; nested takes longer. Follow it under Deployments.
🛡️ The router that broke "Create the cluster"
This is the story of the lab. Hours into the deployment, the Create the cluster step failed with this:
Type 'TestClusterPrereq' of Role 'Cluster' raised an exception:
Unexpected error when attempting to resolve DNS for 'AZLNode1.mct.lab': DNS name does not exist
That made no sense. The DC was up. DNS queries sent straight to the DC (-Server 192.168.100.80) worked perfectly. But on the node, nltest /dsgetdc:MCT.LAB returned ERROR_NO_SUCH_DOMAIN, and Test-ComputerSecureChannel returned False. The node had lost its domain.
So what happened?
During deployment, Network ATC creates a new management virtual NIC: vManagement(compute_management). That brand-new adapter learned fe80::1 (my router, 192.168.100.1) as an IPv6 DNS server through stateless DHCPv6.
Think of it like asking two people for directions. One is the DC, who knows the lab domain. The other is the router, who has never heard of it and confidently says "doesn't exist". Windows took the router's answer, and Netlogon cached the whole domain as not found. Gone: the DC, the secure channel, even the node's own A record.
The fix (tested)
Run this from the host. One important detail: before deployment use .\Administrator, but once deployment has started, use a domain account (MCT\AzLDeployUser or a domain admin). In my lab the saved local credential stopped working once deployment had run.
$cred = Get-Credential -Message 'Node admin (.\Administrator before deployment, MCT\... after)'
Invoke-Command -VMName $NodeVM -Credential $cred -ScriptBlock {
# 1. Find interfaces that learned an IPv6 DNS server
$bad = Get-DnsClientServerAddress -AddressFamily IPv6 | Where-Object ServerAddresses
if (-not $bad) { 'No IPv6 DNS servers - nothing to fix'; return }
$bad | Select-Object InterfaceAlias, ServerAddresses
# 2. Stop learning DNS/routes from the router over IPv6 on those interfaces
foreach ($i in $bad) {
Set-NetIPInterface -InterfaceIndex $i.InterfaceIndex -AddressFamily IPv6 -RouterDiscovery Disabled -Dhcp Disabled -OtherStatefulConfiguration Disabled
}
# 3. Flush the poisoned caches and re-register
Clear-DnsClientCache
Restart-Service Netlogon
Start-Sleep 5
nltest /dsgetdc:MCT.LAB /force
nltest /sc_verify:MCT.LAB
ipconfig /registerdns | Out-Null
# 4. Verify
Resolve-DnsName _ldap._tcp.dc._msdcs.MCT.LAB -Type SRV | Select-Object -First 1 NameTarget
Resolve-DnsName AZLNode1.mct.lab | Select-Object Name, IPAddress
'SecureChannel: ' + (Test-ComputerSecureChannel)
Get-DnsClientServerAddress -AddressFamily IPv6 | Where-Object ServerAddresses
}
You want to see: the SRV record pointing to mct-dc.mct.lab, AZLNode1.mct.lab returning only 192.168.100.50, NERR_Success from nltest /sc_verify, SecureChannel: True, and no IPv6 DNS servers left. Then click Resume on the deployment in the portal. In my lab, the cluster was created right after.
How to prevent it next time
| Option | Effect | Status |
|---|---|---|
| Turn off IPv6 DNS advertisement (RDNSS/DHCPv6) on the router | Permanent; covers the DC, every node and every future lab | Recommended |
| Run the fix on each node as soon as vManagement(…) exists, and again if Network ATC recreates it | Fixes one node; can come back | Tested |
NRPT rule: Add-DnsClientNrptRule -Namespace ".mct.lab","mct.lab" -NameServers 192.168.100.80 | Sends every mct.lab query to the DC, whatever the adapter | Not tested in this lab |
| Put the lab on an isolated or NAT network | The router can't reach the lab | Bigger change |
✅ After deployment
Check the portal: Azure Local → your instance should show Connected with one machine. From release 2610 onward you can also run Get-SolutionHealthSummary on the node.
Update, in order
You deployed 2608. Go to Updates on the instance and install them in the order the portal offers them (cumulative before feature) until you reach the latest release. Never use Windows Update directly. A release is supported for six months.
Clean up secrets
The service principal isn't needed after deployment (12.2512+ clusters use a managed identity). Remove it, along with any saved credential files:
# Service principal - not needed after deployment (12.2512+ clusters use a managed identity)
$app = Get-AzADApplication -DisplayName $SPName
Get-AzRoleAssignment -ServicePrincipalName $app.AppId | Remove-AzRoleAssignment | Out-Null
Remove-AzADApplication -ObjectId $app.Id
# Saved credential files on the host
Remove-Item C:\AZL-Lab\*.cred.xml -ErrorAction SilentlyContinue
Then change every password that appeared in a chat or a script, and use a domain account for PowerShell Direct to the node from now on.
🧰 Every error I hit, in one table
Bookmark this one. If you hit an error, search this page for it first.
| Error | Cause | Fix |
|---|---|---|
| Get-/Register-AzResourceProvider not recognized | Az.Resources not installed | Install Az |
| Import-Module: … ActiveDirectory … not loaded | AD tool run on a machine without RSAT | Run it on the DC via PowerShell Direct |
| RoleAssignmentExists (Conflict) | Role already assigned | Harmless |
| Cannot convert System.Object[] … to System.Guid | Several SPs with the same name | Unique SP name; remove old ones |
| The credential is invalid (PowerShell Direct) | Wrong password, or bare username | .\Administrator; check on the console; domain account after deployment |
| … does not resolve to a single virtual machine | Running -VMName from inside the node session | Exit-PSSession first |
| Given TargetSolutionVersion : … is not valid | The ISO's zero-day update doesn't offer that version | Leave the target out |
| Get-ArcZeroDaySolutionUpdateConfig asks for ConfigFilePath | It needs a parameter | Ctrl+C; you don't need it |
| No extensions on the Arc machine | Expected before the portal step | The portal installs them |
| RDMA operational / symmetry and bandwidth | RDMA left on in the intents | Disable RDMA per intent |
| Network Infra Connection: "The parameter is incorrect" | Timing in the nested VM | Run validation again |
| Create the cluster: DNS name does not exist | Router's IPv6 DNS (fe80::1) | The DNS fix |
| Very slow downloads / LinkSpeed 100 Mbps | Host NIC negotiated 100 Mbps | Fix cable or switch port |
| Password accepted by AD but fails deployment | AD policy weaker than Azure Local's | Test-AzLPassword |
Where the logs are (on the node)
| Log | What it shows |
|---|---|
C:\CloudContent\MASLogs\AzStackHciEnvironmentChecker.log | Detailed trace of every validation test |
C:\MASLogs\ and C:\CloudDeployment\Logs\ | Deployment launcher and ECE steps |
C:\Deployment\Unattended.json | What the portal sent to the node (IPs, intents, RDMA override) |
C:\Windows\debug\netlogon.log | DC discovery failures (enable with nltest /dbflag:0x2080ffff) |
Collect-ArcBootstrapSupportLogs | Arc registration and bootstrap |
🎯 The ten lessons that matter most
Let's wrap up. If you keep only one list from this article, keep this one. Each lesson cost me somewhere between ten minutes and half a day.
- Your router can break the deployment. A home or office router that advertises itself as an IPv6 DNS server makes the node lose the domain at "Create the cluster". Details
- The Azure Local password rule (14+ characters, all four character types) is stricter than Active Directory's. AD happily accepts passwords that fail deployment later. Details
- Don't force TargetSolutionVersion. The ISO only accepts certain targets. Leave it out and register in an interactive session. Details
- You need Az.Resources on the host, and the AD tool needs the ActiveDirectory and GroupPolicy modules. Run the AD step on the DC. Details
- Give each deployment a uniquely named service principal. Duplicate names break every script that looks one up. Details
- Turn off RDMA on every network intent. A nested VM can't do RDMA, and leaving it on also triggers a 10 Gbps bandwidth check. Details
- "Network Infra Connection — The parameter is incorrect" was a one-off timing failure. Run validation again before you change anything. Details
- The portal installs the Arc extensions when you add the machine. You don't install them yourself. Details
- Watch the host's link speed. Mine negotiated 100 Mbps, which made everything painfully slow. Details
- Never paste secrets into a chat, a ticket or a script. Use
Get-CredentialorRead-Host -AsSecureString.
📦 Get the full guide
This article covers the story and the scripts that matter most. The complete guide, with the full checklist, the lab values table and the original Word version, is free on GitHub:
If you only remember one thing, make it this: check your router before you start. It took the most time to find, and it's the one problem no official doc warned me about.
Have you deployed Azure Local nested and hit something that's not in the table? Tell me on LinkedIn. I'll test it and add it to the guide.
🔗 Continue the conversation on LinkedIn
Follow Mohamed Dyabi for insights on Azure hybrid infrastructure, adaptive cloud, and Microsoft certifications across EMEA.
Follow on LinkedIn