Fix it now
Error 1722 is published as “The RPC server is unavailable”: a lower layer reported a connectivity failure and the call never reached the directory. Microsoft’s first documented check is not the firewall, it is the RPC ClientProtocols registry key on both machines.
reg query HKLM\Software\Microsoft\Rpc\ClientProtocols
netsh int ipv4 show dynamicport tcp
repadmin /bind <partner>
nltest /dsgetdc:<domain>
- Check ClientProtocols first. Microsoft documents four values that must be present, each REG_SZ with data
rpcrt4.dll:ncacn_http,ncacn_ip_tcp,ncacn_npandncacn_ip_udp. A missing entry produces 1722 with no network fault at all. - Test the endpoint mapper:
Test-NetConnection <partner> -Port 135. If that fails, nothing downstream will work. - Read the dynamic port range on both machines with
netsh int ipv4 show dynamicport tcp, and give the network team the range those machines actually report rather than a figure from an article. - Test the bind that replication really performs:
repadmin /bind <partner>, in both directions. - If the range cannot be opened, pin the ports instead:
TCP/IP PortunderHKLM\SYSTEM\CurrentControlSet\Services\NTDS\ParametersandDCTcpipPortunderHKLM\SYSTEM\CurrentControlSet\Services\Netlogon\Parameters, both REG_DWORD, then restart the controller.
Set the same pinned ports on every controller that talks across the restricted boundary. A half-configured pair fails in a way that looks exactly like a firewall problem all over again.
If repadmin /bind succeeds both ways and replication runs, you are done. The next section explains how an RPC call finds its port and what the neighbouring codes are telling you.
Why it happens
RPC does not use a fixed port for directory replication. The client connects to the endpoint mapper on TCP 135 and asks which port the directory replication interface is listening on right now. The mapper answers with a port from the dynamic range, the client opens a second connection to that port, and the real work happens there. Two connections, two ports, and a firewall that permits only the first produces exactly this error.
Microsoft frames 1722 broadly: it occurs when a lower-layer protocol reports a connectivity failure, and the named causes include link-local failure, DHCP, DNS, WINS, routing and blocked firewall ports, IPsec or network authentication failures, resource limitations, and a higher-layer protocol simply not running. That breadth is why the first documented resolution step is a registry check rather than a firewall rule: the ClientProtocols key under HKLM\Software\Microsoft\Rpc must list ncacn_http, ncacn_ip_tcp, ncacn_np and ncacn_ip_udp, each pointing at rpcrt4.dll. If one of those is missing the machine cannot make the call at all, and no amount of network work will help.
Because the ports move, this class of failure is often intermittent in a way that misleads people. Replication succeeds when the chosen port falls inside a permitted range and fails when it does not, so the same pair of controllers can look healthy for hours and then fail. Any inconsistency of that kind should push you towards the port range rather than towards the directory.
The RPC client protocols are not registered
You have this one if 1722 with no network fault, sometimes on a machine that has had a cleanup tool or a failed update run against it.
- Read the key on both machines:
reg query HKLM\Software\Microsoft\Rpc\ClientProtocols - Confirm all four values are present as REG_SZ with data
rpcrt4.dll:ncacn_http,ncacn_ip_tcp,ncacn_np,ncacn_ip_udp. - Restore any that are missing, then restart the machine.
- Re-test with
repadmin /bind <partner>before looking at anything else.
This is Microsoft’s first documented resolution step for 1722 and it costs thirty seconds. Do it before you raise a firewall change.
The dynamic RPC port range is not permitted end to end
You have this one if Port 135 tests fine, replication still fails, and the failure moves around rather than being constant.
- Read the range in use on both machines:
netsh int ipv4 show dynamicport tcp, and the udp equivalent. - Give the network team the exact range those machines report, from each controller to each partner, in both directions.
- Include the other ports directory replication needs on the same request: TCP 135, TCP 389 and 636 for LDAP, TCP 3268 and 3269 for the global catalog, TCP and UDP 88 for Kerberos, TCP and UDP 53 for DNS, and TCP and UDP 445 for SMB.
- Retest with
repadmin /bind <partner>before declaring it fixed.
Do not quote a default range from memory. Microsoft’s own 1722 page states 1024-5000 while current builds report a high range starting far above that, which is exactly why the netsh command exists.
The endpoint is not registered on the partner
You have this one if Error 1753, and the partner answers on port 135 but offers nothing for the replication interface.
- On the partner, confirm the directory service is running and the machine is advertising:
dcdiag /test:advertising. - Query the mapper for what is registered:
portqry -n <partner> -e 135, and look for the directory replication interface in the output. - You can also scan the classic range with
portqry -n <partner> -r 1024-5000if you are chasing a specific endpoint. - If the service is running and nothing is registered, restart the partner outside production hours; a controller in that state is usually unhealthy in other ways too.
The next fix writes to the NTDS and Netlogon parameters in the registry and needs a restart of the domain controller to take effect. Export the keys first, apply to one controller at a time, and confirm replication before moving to the next. A wrong value here stops replication rather than fixing it.
The network genuinely cannot allow a dynamic range
You have this one if The firewall team will not open the high range and replication has to cross that boundary.
- Choose an unused high port and set it as
TCP/IP Port, a REG_DWORD, underHKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters. - Set
DCTcpipPort, also REG_DWORD, underHKLM\SYSTEM\CurrentControlSet\Services\Netlogon\Parameters, so both halves of the conversation are predictable. - Restart each controller after the change, one at a time.
- Ask for TCP 135 and your chosen ports to be permitted, then confirm with
repadmin /bindfrom both directions.
Every controller that talks across the restricted boundary needs the same configuration. Microsoft’s examples use the same port number for both values, which is legitimate and easier to reason about.
Full reference
The six codes, as published
| Code | Symbolic name | Published text |
|---|---|---|
1722 |
RPC_S_SERVER_UNAVAILABLE | The RPC server is unavailable |
1753 |
EPT_S_NOT_REGISTERED | There are no more endpoints available from the endpoint mapper |
1727 |
RPC_S_CALL_FAILED_DNE | The remote procedure call failed and did not execute |
1726 |
RPC_S_CALL_FAILED | The remote procedure call failed |
1725 |
RPC_S_NO_CALL_ACTIVE | There are no remote procedure calls active on this thread |
1717 |
RPC_S_UNKNOWN_IF | The interface is unknown |
The pair worth internalising is 1726 and 1727. Both mean the call reached the server; only 1727 tells you it did not execute, which narrows the question from “did anything happen” to “why did it refuse”.
The registry values, written out
| Path | Value | Type | Effect |
|---|---|---|---|
HKLM\Software\Microsoft\Rpc\ClientProtocols |
ncacn_http, ncacn_ip_tcp, ncacn_np, ncacn_ip_udp |
REG_SZ, data rpcrt4.dll |
All four must be present or the machine cannot make RPC calls |
HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters |
TCP/IP Port |
REG_DWORD | Pins the AD replication RPC port. Restart required |
HKLM\SYSTEM\CurrentControlSet\Services\Netlogon\Parameters |
DCTcpipPort |
REG_DWORD | Pins the Netlogon RPC port. Restart required |
HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Diagnostics |
1 Replication |
REG_DWORD | 1 for basic, 2 or 3 for verbose, 5 for extended NTDS replication logging |
The dynamic port range, and why not to quote a number
Microsoft’s 1722 troubleshooting page states the default dynamic port range as 1024-5000, which was the default on older releases. Current Windows Server builds report a high range instead. Rather than adjudicate that in an article, read it on the machines in front of you:
| Command | What it does |
|---|---|
netsh int ipv4 show dynamicport tcp |
Reports the current TCP dynamic port range |
netsh int ipv4 show dynamicport udp |
The same for UDP |
netsh int ipv4 set dynamicport tcp start=<port> num=<count> |
Sets the range, if you need both ends to match |
Whatever the two machines report is what has to be permitted between them. A firewall rule written from an article that quoted the other range is precisely how you get a pair of controllers that replicate three times out of four.
Ports directory replication needs, beyond RPC
- TCP 135 for the RPC endpoint mapper.
- TCP 389 for LDAP and TCP 636 for LDAP over SSL.
- TCP 3268 and 3269 for the global catalog.
- TCP and UDP 88 for Kerberos.
- TCP and UDP 53 for DNS.
- TCP and UDP 445 for SMB.
- The dynamic RPC range, or the ports you pinned in its place.
One thing this article deliberately leaves out
Microsoft’s 1722 page carries a section recommending that SMB signing be disabled to resolve signing mismatches. That guidance is old and this article does not reproduce it. Disabling SMB signing on a domain controller weakens a protection that exists for good reason, and it is not needed to fix a port or endpoint problem. If you genuinely have a signing mismatch, resolve it by making both ends agree on signing rather than by turning it off.
Diagnostic order that saves a change window
reg query HKLM\Software\Microsoft\Rpc\ClientProtocolson both machines. Free, instant, and Microsoft’s first step.Test-NetConnection <partner> -Port 135. If this fails, nothing else matters yet.repadmin /bind <partner>in both directions. Success here means RPC is fine and the fault is elsewhere in the directory.netsh int ipv4 show dynamicport tcpon both, then a single firewall request with the real numbers in it.portqry -n <partner> -e 135if you need to prove which interfaces are actually registered.dcdiag /test:dns /v /e /f:<file>, because DNS is on Microsoft’s cause list for 1722 and looks nothing like a name resolution problem when it presents this way.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
1722 |
RPC_S_SERVER_UNAVAILABLE: the RPC server is unavailable | Microsoft Learn |
1753 |
EPT_S_NOT_REGISTERED: there are no more endpoints available from the endpoint mapper | Microsoft Learn |
1727 |
RPC_S_CALL_FAILED_DNE: the remote procedure call failed and did not execute | Microsoft Learn |
1726 |
RPC_S_CALL_FAILED: the remote procedure call failed | Microsoft Learn |
1725 |
RPC_S_NO_CALL_ACTIVE: there are no remote procedure calls active on this thread | Microsoft Learn |
1717 |
RPC_S_UNKNOWN_IF: the interface is unknown | Microsoft Learn |
Confirm the fix worked
repadmin /bind <partner>succeeds in both directions.reg query HKLM\Software\Microsoft\Rpc\ClientProtocolsreturns all four protocol values on both machines.repadmin /replsummaryshows every partner with a recent success.- Force a full sync and confirm it completes without errors.
- Restart one controller and re-test, particularly if you pinned a static port, so you know the fix survives a reboot.
Questions people ask about this
Do I need to buy anything to fix this?
No. This is registry values on your own domain controllers plus firewall configuration. Nothing about it is a licensing question.
What is the default dynamic port range?
It depends on the build, and that is why this article will not quote you one. Microsoft’s own 1722 page states 1024-5000, which is the older default, while current builds report a high range. Read it with netsh int ipv4 show dynamicport tcp on the two machines involved and use what they say.
Is pinning replication to a single port a good idea?
It is a reasonable trade when a firewall stands between controllers, and it makes the rules readable. It is unnecessary complexity on a flat internal network where the dynamic range can simply be permitted. If you do it, set both TCP/IP Port and DCTcpipPort, on every controller that crosses the boundary.
Why does replication work sometimes and fail others?
Because the port is chosen per call. If only part of the dynamic range is permitted, roughly that proportion of attempts succeeds, which is the signature of a partially opened range.
Should I disable SMB signing, as some guidance suggests?
No. That advice appears in older troubleshooting material and this article does not repeat it. It weakens a protection on a domain controller and it does not fix a port or endpoint problem.
