Fix it now
0x800706D9 is Windows error 1753: there are no more endpoints available from the endpoint mapper, so the cluster’s RPC call never reached a listener. Event ID 5398 is a different failure entirely – the cluster will not start because no available node holds the latest copy of the cluster configuration.
Test-NetConnection <other-node> -Port 135
Test-NetConnection <other-node> -Port 3343
Test-NetConnection <other-node> -Port 445
Get-Service ClusSvc -ComputerName <other-node>
- If 135 fails, nothing else on this page matters yet. Microsoft’s documented port list for cluster nodes is TCP and UDP 135, TCP and UDP 3343, TCP 445 and TCP 139, plus 5000-5099 where Event ID 1721 appears and 8011-8031 for internode RPC.
- Confirm the cluster service is actually running on the node you are joining to. An endpoint mapper with nothing registered behind it produces the same 1753.
- If the failure is Event ID 5398 instead, do not force anything yet. Start the cluster service on every node in the cluster, including the ones you think are irrelevant – the node holding the current copy may simply be switched off.
- Retry the join with
Start-ClusterNode -Name <node>, or Add Node in Failover Cluster Manager. - If the cluster service will not start at all on a recently hardened node, check whether Group Policy has disabled “Allow Custom SSPs and APs to be loaded into LSASS” under Computer Configuration, Administrative Templates, System, Local Security Authority.
Event ID 5377 is often in the same log and is not this problem. It records an internal cluster service operation exceeding its threshold, after which the service is terminated to recover, restarted by Service Control Manager, and the node rejoins on its own.
If every node reads Up and survives a restart, you can stop here. The next section separates the three failures that all look like “the node will not join”.
Why it happens
Forming a cluster and joining one are different operations that fail in similar ways. Forming means a node decides it has enough of the cluster to start: it reads its local copy of the cluster database, checks the quorum configuration, and brings the cluster up. Joining means contacting a node that is already up and asking for a current copy of that database and a place in the membership.
Joining depends on RPC. The joining node resolves the peer, connects to the endpoint mapper on TCP 135, is told which dynamic port the cluster service is listening on, and connects there. 0x800706D9 is Windows error 1753, EPT_S_NOT_REGISTERED, “There are no more endpoints available from the endpoint mapper” – the conversation ended at that first step. Either nothing answered on 135, or the endpoint mapper had no registration for the cluster service because the service is not running. Heartbeats then use 3343, and it is entirely possible to have 135 open and 3343 blocked, which produces a node that joins and is promptly evicted.
Event ID 5398 has nothing to do with the network. Clustering replicates its configuration database and tracks which copy is newest. If changes were made while some nodes were down, and only those out-of-date nodes are available now, the cluster refuses to start rather than discard the newer configuration. The event prints the arithmetic: votes required to start the cluster, votes available, and how many nodes have votes.
The RPC path is blocked between the nodes
You have this one if 0x800706D9 on a node that resolves names and answers ping. Port 135 fails a connection test.
- Test each port in turn from the joining node: 135, 445 and 3343.
- Have the network team permit Microsoft’s documented set between all nodes: TCP and UDP 135, TCP and UDP 3343, TCP 445, TCP 139, and the internode RPC range 8011-8031.
- If connecting to the cluster as an administrator produces Event ID 1721, open 5000-5099 as well.
- Retest every port before retrying the join; a third-party endpoint firewall drops packets before the Windows rules ever see them.
No available node holds the current cluster configuration
You have this one if Event ID 5398 rather than an RPC error. The cluster service starts and then stops, on every node you try.
- Start the cluster service on every node in the cluster before doing anything else.
- Read the votes required, votes available and nodes with votes printed in the event; they tell you how far short you are.
- If a node is genuinely unrecoverable and you accept losing configuration changes made after your surviving copy was current, start with
Start-ClusterNode -ForceQuorum. - Once up, review the quorum configuration and the resource list immediately, because anything changed after that copy was current is gone.
Forcing quorum tells the cluster to trust the copy you have. Do it only when you know why the other nodes are unavailable.
Security hardening stopped the cluster service loading its authentication provider
You have this one if The cluster service will not start at all on a node that was working before a baseline was applied.
- Check Computer Configuration, Administrative Templates, System, Local Security Authority for “Allow Custom SSPs and APs to be loaded into LSASS”.
- If it is Disabled, LSASS cannot load the Cluster Authentication Manager, and the cluster cannot authenticate or start.
- Set it to Enabled, run
gpupdate /force, and start the cluster withStart-ClusterorStart-ClusterNode. - Fix it in the domain policy rather than locally, or the next refresh undoes you.
The cluster service is timing out on itself
You have this one if Event ID 5377, and the node drops out and comes back without anyone doing anything.
- Read the threshold in the event body: an internal operation exceeded that many seconds and the service was terminated to recover.
- Treat it as a performance problem on the node, not a membership problem. Look at storage latency and CPU pressure in that window.
- Collect the cluster log for a narrow window:
Get-ClusterLog -Node '<node>' -Destination c:\temp -UseLocalTime
Full reference
Which failure you actually have
| What you see | What it means |
|---|---|
| 0x800706D9 | Windows 1753, EPT_S_NOT_REGISTERED: no endpoints available from the endpoint mapper |
| Event ID 5398 | The cluster failed to start because the latest copy of the configuration data was not available among the nodes trying |
| Event ID 5377 | An internal cluster service operation exceeded its threshold; the service was terminated to recover and will restart and rejoin |
| 0x8007138F | Windows 5007, ERROR_CLUSTER_RESOURCE_NOT_FOUND: the cluster resource could not be found |
| Event ID 1721 | Logged when connecting to the cluster as administrator; Microsoft’s remedy is to open TCP 5000-5099 for RPC |
The ports, as Microsoft publishes them
| Port | Used for |
|---|---|
| TCP 135 | RPC endpoint mapper and DCOM |
| UDP 135 | RPC endpoint mapper over UDP |
| TCP and UDP 3343 | Cluster network driver |
| TCP 445 | SMB |
| TCP 139 | NetBIOS session service |
| TCP 5000-5099 | RPC traffic, where Event ID 1721 appears |
| TCP 8011-8031 | Internode RPC traffic between cluster nodes |
Test them from the node that is failing, in the direction that is failing. A path that works from the server you are logged on to tells you nothing about the subnet the failing node lives in, and cluster traffic is bidirectional.
Rights the cluster service needs on the node
Hardening baselines break cluster startup in two distinct ways, and both are worth checking on a node that used to work. The first is the Local Security Authority policy that stops LSASS loading custom security support providers: with it disabled, the Cluster Authentication Manager never loads and the service cannot authenticate. The second is user rights assignment. Microsoft’s guidance for a cluster service that will not start is to confirm the accounts involved still hold “Access this computer from the network” and “Allow log on locally”, and that policies denying access to local accounts or to Everyone have not swept them up.
Collecting evidence before you change anything
| Command | What it gives you |
|---|---|
Get-ClusterLog -Node '<node>' -Destination c:\temp -UseLocalTime |
The cluster log for that node, in local time |
Test-NetConnection <node> -Port <n> |
Whether the port is reachable from where you are standing |
Get-Service ClusSvc |
Whether the service on the far end is even running |
NET HELPMSG <code> |
The text behind a numeric status carried in a cluster event |
Forcing quorum and clearing a node’s local cluster configuration are destructive in different ways. Forcing quorum can discard configuration changes made after the copy you force; clearing a node wipes its local cluster state and should only ever be done on a node that has already been evicted. Take a documented copy of the cluster configuration and confirm your backups before either.
When the ports are open and it still will not join
- Confirm both nodes agree on time. Kerberos will not authenticate across a large skew, and the failure surfaces as an RPC or access error rather than as a clock message.
- Confirm the cluster networks in Failover Cluster Manager are still enabled for cluster communication on the network you expect to carry it.
- Check for a NIC teaming or virtual switch change that quietly moved cluster traffic onto a different adapter and subnet.
- If one node fails while every other node forms the cluster happily, and the failure survives reboots and firewall changes, evict it and re-add it rather than continuing to debug its local state in place.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
0x800706D9 |
Windows error 1753, EPT_S_NOT_REGISTERED: there are no more endpoints available from the endpoint mapper, so the RPC call never reached a listener | Microsoft Learn |
Event ID 5377 |
An internal cluster service operation exceeded its defined threshold in seconds; the cluster service was terminated to recover, and Service Control Manager restarts it so the node rejoins | Microsoft Learn |
0x8007138F |
Windows error 5007, ERROR_CLUSTER_RESOURCE_NOT_FOUND: the cluster resource could not be found | Microsoft Learn |
Event ID 5398 |
The cluster failed to start because the latest copy of the cluster configuration data was not available within the set of nodes attempting to start it; the event prints votes required, votes available and nodes with votes | Microsoft Learn |
Confirm the fix worked
Get-ClusterNodereports every node as Up.Test-NetConnection <peer> -Port 3343succeeds in both directions between every pair of nodes.- Restarting the cluster service on one node lets it rejoin with no manual intervention.
- A full node reboot produces no new 5398 and no RPC failures in the FailoverClustering log.
- If 5377 was the original symptom, it does not recur through a full backup window.
Questions people ask about this
Is it safe to force quorum to get the cluster up?
It is safe in the narrow sense that the cluster will start. It is not safe in the sense that it may discard configuration changes made after the copy you are forcing. Use it when you understand why the other nodes are gone and have accepted that trade.
Why does the cluster need high ports as well as 135?
That is ordinary RPC behaviour: the service registers with the endpoint mapper on 135 and is given a dynamic port. A firewall that permits only 135 lets the first packet through and blocks everything after it, which is why the failure looks so much like a working connection.
Can I turn the firewall off instead of opening ports?
You can, and you should not. The port list is short and published, and turning a host firewall off to fix one port is a far larger change than opening it.
A node joined and was immediately evicted. Same problem?
Usually a different one. Joining uses 135 and a dynamic port; membership uses 3343. If 3343 is unreliable the node joins and then loses heartbeats, which is the pattern you are describing.
Do I need a licence or a bigger edition to fix this?
No. Every fix here is a firewall rule, a policy setting or a service state. Failover Clustering is included in both Standard and Datacenter.
