About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
A node shown as quarantined in a Windows Server failover cluster cannot host cluster roles, such as virtual machines (VMs), until it leaves quarantine. Quarantine is a protective feature: it isolates a node that keeps failing so that one unstable server stops disrupting the workloads on the others. Microsoft's page names repeated health check failures, persistent service failures, hardware issues, and storage or network communication problems as triggers.
Find out why the node was isolated before you clear it. Clearing quarantine with Start-ClusterNode -Name node -ClearQuarantine returns the node to service, but if the cause is still there the node can be isolated again. The checks below read information only. The fixes that change anything are marked.
What the error means
Microsoft describes cluster node quarantine as "a protective feature in Windows Server Failover Clustering that temporarily isolates unstable or problematic nodes to safeguard cluster health and application workloads". A quarantined node "can't host cluster roles, such as virtual machines (VMs), until it exits quarantine, either automatically or through administrator intervention."
Microsoft's page applies to "Supported versions of Windows Server". It does not give the number of failures that triggers quarantine, or how long the node stays out before it leaves on its own, so neither is stated here. The state shows in Failover Cluster Manager or with Get-ClusterNode, and Microsoft's checklist starts with exactly that step: identify the quarantined nodes. The events that point to the reason, per Microsoft's page, include 1641 ("Node quarantine activated"), 1647 and 1649, with 7031 (Service Control Manager: "unexpected service termination") for repeated service failures.
For quarantine caused by network communication problems Microsoft's page lists event 1135 ("Node removed from cluster") and event 1177.
Likely causes
| Cause | What you see | How to confirm |
|---|---|---|
| Repeated Resource Hosting Subsystem (RHS) or service failures | Node in Quarantine; roles keep failing over; events 1641 and repeated 7031 | System log events; check whether rhs.exe or another process keeps crashing |
| Network communication problems | Events 1135 and 1177; warnings about broken communication channels (status 10054); failed connections to port 3343 | Test-NetConnection to the key ports; MTU values across cluster adapters |
| Storage or CSV problems | Events 5120 and 5142; Cluster Shared Volume (CSV) paused or disconnected; storage path failures | Get-ClusterSharedVolumeState on the affected node; storage vendor logs |
| Security or antivirus software blocking cluster traffic or services | Silent packet drops; cluster service blocked from running | Security software logs; exclusions for cluster directories |
| Configuration gaps | Quarantine after a restart or update; DHCP used for cluster IP; missing cluster certificate; failing quorum witness | Cluster IP resource, node certificates, witness configuration |
Prove the cause before changing anything
Record the current state first. These commands only read.
Get-ClusterNode
Get-ClusterGroup
Get-ClusterSharedVolumeState
Get-ClusterNode shows the name, id and state of each node, and Get-ClusterGroup the state and owner node of each clustered role. Keep the output. It shows which node owns which VM before you touch anything, and that is what you restore if roles move.
Then read the System and FailoverClustering event logs on the quarantined node for the events above. Microsoft's page says to look at "Event IDs 1641, 1647, 1649" for the reason, and at 5120 and 5142 for storage.
For the network, Microsoft's page recommends Test-NetConnection for key ports and lists "UDP 3343, SMB, TCP 6600 for live migration" as ports to keep open.
Test-NetConnection -ComputerName hv02 -Port 3343
The Test-NetConnection page describes a ping test, a TCP test and route tracing, and the port is listed as "UDP 3343", so a TCP pass does not show that the UDP path is open. A TCP failure on 3343 does not by itself show a fault either; Microsoft's quarantine page says only "Run Test-NetConnection for key ports".
Collect cluster logs before any change. Create the destination folder first (this is the author's step; the Get-ClusterLog page does not mention it). The Get-ClusterLog page names -Destination, -TimeSpan (minutes) and -UseLocalTime:
Get-ClusterLog -Destination C:\ClusterLogs -UseLocalTime -TimeSpan 120
Example (illustrative): a three-node cluster clu01 at contoso.com.au has nodes hv01, hv02 and hv03. hv02 shows Quarantine after a Tuesday-night patch run. The System log on hv02 shows 7031 several times for the same security agent service, then event 1641. Get-ClusterNode shows the other two nodes Up. The cause is the security agent, not the cluster, and clearing quarantine at once would repeat the failures.
Fix
Fix the cause shown by the events. Do not clear quarantine first.
RHS or service failures. Microsoft's page says to read the failure details in Cluster.log and the System and Application event logs, and, if a specific process such as rhs.exe is implicated, to obtain memory dump files and analyse them. It lists: apply the latest updates for Windows, Hyper-V and third-party software (the marked update step below carries its safeguards); add antivirus exclusions for .avhdx, .vhdx and cluster-related files; and check known bugs, applying a hotfix if one is available.
Risk: an antivirus exclusion changes your security cover for those paths. Record the existing exclusion list first and add only the paths Microsoft's page names. Back-out: remove the entries you added.
Disruptive step: installing updates can require a restart of the node, and the VMs and roles it hosts stop or move while it restarts. Save first: keep the Get-ClusterGroup output above and take a current backup of the VMs on the node. Update one node at a time. Pause and drain the node first, so roles move off it in an orderly way:
Suspend-ClusterNode -Name hv02 -Drain
Install the updates, restart the node, then bring it back:
Resume-ClusterNode -Name hv02 -Failback NoFailback
The Suspend-ClusterNode page says that with -Drain "clustered roles currently running on the node will be drained before the node is paused". NoFailback keeps drained roles off the node until you decide, because the cause of the quarantine may still be present; -Failback Immediate, the form on the Resume-ClusterNode page, brings them back at once, so use it only after the cause is fixed. Resume-ClusterNode ends the pause. Microsoft's quarantine page names only Start-ClusterNode -ClearQuarantine for clearing quarantine. Back-out: the saved Get-ClusterGroup output shows where each role was. An installed update is removed through the update's own uninstall, which Microsoft's quarantine page does not cover.
Network problems.
Save first: record the current MTU on every cluster adapter.
Risk: changing MTU on a cluster adapter, one of the fixes below, can break traffic between nodes until all adapters match. Change one node at a time.
Make sure the ports above are open and not blocked by a firewall or security software, check for MTU mismatches, and update adapter drivers and firmware. Microsoft's page says to "Set consistent MTU values across all cluster adapter settings". To undo an MTU change, put the recorded value back if traffic fails.
Storage or CSV problems.
Save first: run Get-ClusterSharedVolumeState and Get-ClusterGroup and keep both outputs, note the current multipath settings with the storage vendor's tool, and take a current backup of the VMs on the affected volumes.
Disruptive step: the storage fixes below, reconnecting or restarting a storage path, can pause Cluster Shared Volume (CSV) input and output, and firmware or driver updates restart hardware or the node. Work on one path or one node at a time, with the node paused and drained as above.
Verify storage health with vendor tools and logs, reconnect or restart storage paths, check multipath I/O, and apply firmware or driver updates. Microsoft's page also says to exclude cluster storage directories from antivirus. To undo it, reconnect the path you changed and restore the recorded multipath settings; a firmware or driver update is reversed only by the vendor's own rollback, which Microsoft's page does not cover.
Security software.
Save first: record which product and policy you are changing. If you uninstall, keep the installer and licence to put it back.
Disruptive step: disabling security software on a cluster node, as Microsoft's resolution below does, removes its protection while it is off. Do it on the quarantined node only.
Microsoft's resolution is to "Disable/remove offending security software", retest cluster stability, and reinstall after cluster communication is shown to work. It also says to add the cluster service, process and directories to the exclusion list. To undo it, re-enable the software (or restore the policy) when the test is over.
Configuration gaps. Microsoft's page lists: set the cluster IP resource to static instead of DHCP (with Set-ClusterParameter), and the certificate and quorum witness steps further down.
Save first: record the current values of the cluster IP address resource. The Get-ClusterParameter page shows the pattern of piping a Get-ClusterResource result into it, which for an IP address resource returns its DHCP-related information:
Get-ClusterResource -Name "Cluster IP Address" | Get-ClusterParameter
Use the resource name that Get-ClusterResource lists for your cluster. Keep the output: its parameter names and values are what you set back. Note the witness type and its disk or file share in Failover Cluster Manager, and which certificates each node holds in its certificate console.
Disruptive step: changing the cluster IP resource from DHCP to a static address changes the address clients and nodes use to reach the cluster, and a wrong address can drop the cluster name from the network. Choose an unused address on the cluster subnet and do it in a maintenance window. The Set-ClusterParameter page shows no IP address example, so take the parameter names from the saved Get-ClusterParameter output. The form on that page is a name and a value, piped from the resource:
Get-ClusterResource -Name "Cluster IP Address" | Set-ClusterParameter -Name ParameterName -Value NewValue
Back-out: run the same command for each parameter you changed, with the value from the saved output.
Save first for the certificate and witness steps: export the certificate from the healthy node and keep the file, and keep the witness settings you noted above.
Risk: importing a certificate on a node adds one to that node's store, and repairing the quorum witness changes how the cluster counts votes, so a mistake can cost the cluster its quorum. Change the witness only while every node is Up and you hold the recorded settings.
Microsoft's page also lists: verify cluster node certificates, export and import from a healthy node if missing, and validate and repair the quorum witness configuration. To undo it, remove the certificate you imported, and set the witness back to the recorded type and location.
Clear quarantine.
Disruptive step: the next command lets the node host roles again, so clusters may move VMs to it, and you cannot undo that move except by moving them back.
When the cause is fixed, Microsoft's page gives one command:
Start-ClusterNode -Name hv02 -ClearQuarantine
Run it on a cluster node: the Start-ClusterNode page says it "cannot be run remotely without Credential Security Service Provider (CredSSP) authentication on the server computer".
To undo it: Microsoft's pages show no command that puts a node back into quarantine. To keep roles off the node, move them to another node, or stop the Cluster service on that node with Stop-ClusterNode -Name hv02. The Stop-ClusterNode page says it stops the Cluster service on a node, and refuses if that would bring the cluster below quorum. Stopping the service interrupts whatever the node hosts.
Microsoft's page also lists, for the general recovery steps: "Evict and re-add node; restart cluster service, if it's necessary" and "Restart node after clearing issues". It gives no commands for these. Risk: evicting a node removes it from the cluster, and the page gives no steps for adding it back. Treat this as a last resort, and only with the saved Get-ClusterNode and Get-ClusterGroup output and a current backup of the VMs at hand.
If none of the five causes fits, Microsoft's checklist adds hardware and resource checks: disk status, memory, CPU use and hardware alerts, and whether the node can reach the other cluster members without packet loss. Run those on the quarantined node before you decide the cause is unknown, and keep the cluster logs for a support case.
Confirm the fix
Get-ClusterNodeshows the node as Up, not Quarantine.- Run
Get-ClusterGroupand compare it with your saved copy. Roles are hosted where you expect. - The System log has no new 1641 or 7031 events for the node.
- Microsoft's page says: "Monitor logs for confirmation: node rejoins cluster normally."
What to watch afterwards
- A node that is isolated a second time may still have its cause in place. Collect
Get-ClusterLogagain and compare the time of the first failure. - Watch for new antivirus or endpoint agent versions; Microsoft's page names security software as a cause of dropped cluster traffic.
- After patching, check each node before the next one starts.
Read next
Related articles you may want next.
- TroubleshootingFailover cluster event 1135: cluster node was removed from the active failover cluster membershipEvent ID 1135: a cluster node lost heartbeats and was removed. Read the pattern, the cluster log and the network, before changing any threshold.Read article
- Backup & RecoveryWhen a Windows failover cluster loses quorum: why forcing it to start is the last resortUnderstand quorum loss, witness recovery and the isolation needed before a forced Windows cluster start. No executable recovery recipe.Read article
- TroubleshootingMigration fails: "the processor is not compatible with the host" (SCVMM 2008 error, Hyper-V fix)A Hyper-V or VMM migration fails with 'the processor is not compatible with the host'. Find out why, prove it, and use the documented workarounds.Read article
Risks and limits
These commands have not been run or lab-tested.
Microsoft's page does not state how many failures cause quarantine, how long it lasts, or what each of 1647 and 1649 means beyond listing them as quarantine reasons. Its table of reference event IDs includes 1069, 1558 and 121 for configuration and quorum issues without further explanation. If a cluster shows a state or event that Microsoft does not describe, keep the cluster logs from every node and open a support case.
Sources
- Guidance for troubleshooting cluster node quarantine issues — Microsoft Learn, page dated 2026-02-12, accessed 20 November 2026
- Start-ClusterNode — Microsoft Learn, date not shown, accessed 20 November 2026
- Stop-ClusterNode — Microsoft Learn, date not shown, accessed 20 November 2026
- Test-NetConnection — Microsoft Learn, date not shown, accessed 20 November 2026
- Set-ClusterParameter — Microsoft Learn, date not shown, accessed 20 November 2026
- Get-ClusterNode — Microsoft Learn, date not shown, accessed 20 November 2026
- Get-ClusterGroup — Microsoft Learn, date not shown, accessed 20 November 2026
- Get-ClusterLog — Microsoft Learn, date not shown, accessed 20 November 2026
- Suspend-ClusterNode — Microsoft Learn, date not shown, accessed 20 November 2026
- Resume-ClusterNode — Microsoft Learn, date not shown, accessed 20 November 2026
- Get-ClusterParameter — Microsoft Learn, date not shown, accessed 20 November 2026
- Get-ClusterResource — Microsoft Learn, date not shown, accessed 20 November 2026
This article was originally published on .