M365Con 2027 – free online conference · 11–22 Jan 2027 · 120 sessions→ m365con.net

Blog · Mirko Peters

Azure Virtual Desktop Session Host Shows Unavailable: A Step-by-Step Checklist

By ·

A session host that is running but marked Unavailable in the host pool cannot take users. Read the failed health check first, then check domain join, services, URLs, autoscale and the agent registration.

You open the host pool in the Azure portal. The VMs are running. The session hosts show Unavailable. Users cannot connect, or they connect to the one host that is still healthy.

Key points

  • A VM in the Running state tells you nothing about the AVD agent. The host status comes from the agent's health checks.
  • A failed domain join or a broken side-by-side stack listener is a fatal check. A failed monitoring agent, metadata service or URL check is listed as non-fatal and leads to Needs Assistance.
  • After an autoscale plan starts and stops hosts, check roles at subscription level, drain mode and the agent services before you reinstall anything.
  • Reinstall the agent and boot loader with a new registration key only after the cheaper checks have failed.

Microsoft Q&A has several 2025 and 2026 threads on this. One asker with a new environment fixed it by joining the VM to the domain by hand. Another saw hosts go Unavailable one after another after turning on a scaling plan, and the thread did not end with a confirmed fix. This article follows the order that Microsoft Learn suggests and adds the points from the threads. It is not an official Microsoft procedure. Check each step against the current Learn pages, because the agent troubleshooting article and the session host configuration article are older than some of the other pages.

What you see

  • Host pool, then Session hosts: status Unavailable, sometimes Needs Assistance or Upgrading.
  • The VM itself is in the Running state.
  • After a restart, the VM runs again but the host stays Unavailable.
  • In the scaling plan case, hosts turn Unavailable gradually, not all at once.

Step 1: Read the failing health check

In the portal, open the host pool, select Session hosts, and look at the details for one affected host. Learn describes the statuses like this:

  • Available: all checks passed.
  • Needs Assistance: a non-fatal check failed. These are the monitoring agent, the Azure Instance Metadata Service and the URL check. Users can still connect.
  • Unavailable: the host is off, or it failed a fatal check.
  • Upgrading or Upgrade Failed: the agent update is in progress or has failed.

The fatal checks on the Learn page are the domain joined check, the domain trust check and the side-by-side stack listener check. The first one is where the 2025 thread ended.

Step 2: Check the domain join

The thread with a confirmed fix started with a user who could not connect to a brand-new environment. A volunteer moderator suggested the standard deployment guide and AVD Insights. An independent advisor listed generic checks (event logs, domain or Azure AD join, open ports, custom image agent, service health). The asker later wrote that the VM had not joined the domain on its own, and joining it by hand and restarting solved the problem.

Check this on the VM:

  1. Sign in to the VM with a local administrator account and check whether the machine is a member of your domain.
  2. From a command line on the VM, ping the domain name to check that it resolves. Learn gives this as its basic test.

Microsoft Learn lists typical reasons for a failed join in the session host configuration article: wrong credentials in the deployment template, MFA on the join account, missing permissions, and DNS problems. For DNS, the virtual network of the VMs must reach the domain's network (for example by peering), the network's DNS must point to the domain controllers, and the NIC must inherit or use the same DNS servers.

Also check the point about reusing a computer account name. Learn says the reuse can be blocked when Windows updates released on or after 11 October 2022 are installed and the joining account lacks the rights to reuse computer accounts. Learn names three ways out: use the account that created the computer account, use a member of Domain Administrators, or use an account covered by the Group Policy setting that allows computer account reuse during domain join.

Step 3: Check the agent services

Connect to the VM. If RDP does not work, the Azure portal's Run Command is an option, as one answer suggested.

  1. Open services.msc and check that Remote Desktop Agent Loader (the boot loader, service name RDAgentBootLoader) is running. Start it if it is stopped. One answer in the scaling plan thread also named an RDAgent service to check. The Learn agent troubleshooting page names only the Remote Desktop Agent Loader service in this step, so treat that as reported in a community thread, not confirmed in Microsoft Learn.
  2. Open Event Viewer, then Windows Logs, then Application. Learn names the sources WVD-Agent, WVD-Agent-Updater, RDAgentBootLoader and MsiInstaller. Most errors are event ID 3277 with a keyword in the message.
  3. If the loader stops again after you start it, suspect a registration problem (step 6).

If the VM starts slowly, the Learn page also describes a boot timeout case: set the loader's startup type to Automatic (Delayed Start), and if needed raise the ServicesPipeTimeout value, starting at 60000 milliseconds.

Step 4: Check connectivity to the service

Session hosts only need outbound access. Learn's required endpoint list includes .wvd.microsoft.com on TCP 443, login.microsoftonline.com, the monitoring endpoints gcs.prod.monitoring.core.windows.net and .prod.warm.ingest.monitor.core.windows.net, and the Azure fabric addresses 169.254.169.254 (metadata) and 168.63.129.16 (WireServer). Learn says traffic to those two addresses must not be intercepted, proxied or redirected. It recommends service tags and FQDN tags, because it does not publish a list of IP address ranges to allow instead of the FQDNs.

Test in this order:

  1. Run the Azure Virtual Desktop Agent URL Tool on the session host. It checks each required endpoint.
  2. For an INVALID_FORM event, Learn tells you to read the BrokerResourceIdURI and BrokerResourceIdURIGlobal values under HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\RDInfraAgent, add /api/health to each address and open it in a browser. A healthy result says the broker is healthy.
  3. Look for TLS inspection, a proxy or custom DNS in the path.

A scaling plan thread suggested testing a different health address on rdweb.wvd.microsoft.com. I could not find that address on the Learn pages, so I do not recommend it. Use the broker values from the registry.

That thread also claimed that missing monitoring endpoints cause heartbeat failures. Learn lists the monitoring agent check among the non-fatal checks that produce Needs Assistance, not Unavailable (its health check table calls the check semi-fatal). Open those endpoints anyway, but do not stop looking if the host stays Unavailable.

Step 5: If it started after you enabled a scaling plan

The thread on scaling plans was posted in June 2026. Hosts moved to Unavailable one by one after the plan was enabled. Opening the firewall names and restarting the VMs did not fix most of them. Two answers listed checks, and no answer was accepted.

Microsoft Learn explains how autoscale behaves, and it helps to compare:

  1. Roles at subscription level. The service principal or host pool managed identity needs Desktop Virtualization Power On Off Contributor, assigned at subscription level. Learn warns that a lower scope can keep autoscale from working properly.
  2. Drain mode. In ramp-down, autoscale puts the host with the fewest sessions into drain mode, signs users out after the wait time and deallocates the VM. An answer in the scaling plan thread suggested checking the Allow New Sessions column in the portal after a restart, because drain mode can persist. The Learn autoscale page does not describe this, so treat it as reported in a community thread.
  3. Exclusion tag. Hosts with the exclusion tag are left alone by autoscale. Use it on a test host to see whether the plan is the cause.
  4. Minimum percentage of hosts. Autoscale respects it, and it shuts a host down only if the remaining hosts can take the sessions. I found no Learn text about a host that is started by autoscale but then fails to report to the service, so after a power-on, repeat steps 2 to 4 on that host.
  5. No mixed scaling. Learn says you cannot use autoscale and the older Azure Automation or Logic Apps scaling on the same host pool.

For diagnosis, Learn has a page on setting up diagnostics for autoscale (Set up diagnostics for Autoscale), which links to an autoscale FAQ.

Step 6: Re-register or reinstall the agent

Do this when the checks above do not show a cause, or when the log shows INVALID_REGISTRATION_TOKEN or EXPIRED_MACHINE_TOKEN. For NAME_ALREADY_REGISTERED, Learn says the VM name is probably a duplicate: remove the session host from the pool, create a VM with a unique name and add it again. Learn's full reinstall procedure is:

  1. Uninstall these from Programs and Features, then restart: Remote Desktop Agent Boot Loader, Remote Desktop Services Infrastructure Agent, Remote Desktop Services Infrastructure Geneva Agent and Remote Desktop Services SxS Network Stack. If you connect by RDP, choose not to close applications when removing the SxS stack.
  2. Remove the session host from the host pool: Azure Virtual Desktop, Host pools, your pool, Session Hosts, select the host, Remove.
  3. Generate a new registration key on the host pool's Registration key page. It must be valid for at least one hour and at most 27 days.
  4. Install the current agent and boot loader. Paste the key into the agent installer, then run the boot loader installer and restart.
  5. Wait for the host to show Available.

A thread answer suggested a lighter variant. Learn shows registry values under HKLM:\SOFTWARE\Microsoft\RDInfraAgent: set IsRegistered to 0, put the new key in RegistrationToken, restart the boot loader, and then check that IsRegistered is 1 and the token field is empty. Use it only for the INVALID_REGISTRATION_TOKEN and EXPIRED_MACHINE_TOKEN errors.

Do not join a golden image VM to a host pool by deploying the agent on it. The Learn page on golden images says that session hosts created from such an image later fail to join the host pool because the registration token has expired, and that the host pool deployment process joins the session hosts during provisioning.

Verification

  • The host shows Available within a few minutes. After the full reinstall, Learn says to wait 5 to 10 minutes.
  • In the portal, the details show no failed check.
  • A test user can connect and the session lands on that host.
  • Event Viewer shows no new 3277 events.

Limits

  • Only one of the threads has a confirmed cause, a missing domain join.
  • The scaling plan thread ended without a fix, so steps in step 5 are a comparison with Learn, not a proven cure.
  • If the host is still Unavailable after a clean reinstall, collect the failed check name, event IDs with UTC times and the host pool name, and open a support case.

Related

Sources

Follow the showApple PodcastsSpotify

Entities in this article

Azure Virtual Desktop Host pool (Azure Virtual Desktop) Azure Autoscale Azure Monitor

Related articles

RSS