Fix it now
The machine at the top of the time hierarchy has been told to take its time from the domain hierarchy, and there is nothing above it to ask. Nothing has failed yet: the warning is that the clock the whole forest trusts is a hardware clock nobody is checking. Point it at real NTP peers.
w32tm /config /manualpeerlist:"ntp1.example.net ntp2.example.net" /syncfromflags:manual /update
net stop w32time
net start w32time
w32tm /resync /rediscover
w32tm /query /status
- Confirm you are on the right machine first:
Get-ADDomain | Select-Object PDCEmulator, run against the forest root domain. - Check the result in
w32tm /query /status: the source should be one of the peers you configured, with a recent successful sync time and a low stratum. - Check
w32tm /query /configurationshows Type as NTP rather than NT5DS on this one machine. Every other domain member stays on NT5DS. - If this domain controller is a virtual machine, disable the hypervisor’s time synchronisation integration for it, or the host clock will keep overriding what you just configured.
Configure this on the machine that holds the PDC emulator role, not on a name. If the role moves, the configuration has to move with it, which is why many sites script it as part of role transfer.
Once the status output names an external peer, the hierarchy has a root and you can stop. The next section explains how the rest of the forest inherits from here.
Why it happens
Windows time in a domain is a hierarchy rather than a set of clients. Members synchronise with a domain controller in their domain, domain controllers synchronise with the PDC emulator of their domain, and each domain’s PDC emulator synchronises with the forest root’s. The machine holding the PDC emulator role in the forest root has nothing above it, so unless you give it something external it is authoritative on the strength of its own hardware clock.
That is what the warning is about. The service is not failing; it is telling you the top of the hierarchy is unsourced. Everything below it will faithfully follow whatever that clock says, which is fine until the hardware drifts, the virtual machine is restored, or someone sets the clock by hand and the whole forest follows.
The configuration has two halves, and doing only one is the usual mistake. The PDC emulator must take time from real peers (Type NTP, a manual peer list, synchronising from manual flags), and it must announce itself as a reliable source so that the domain controllers below it accept it. Those are separate settings, and a machine that is correctly synchronised but not announcing will still leave the rest of the domain looking for a source.
The role holder has never been configured with peers
You have this one if w32tm /query /configuration shows Type as NT5DS on the forest root PDC emulator, and w32tm /query /source names a local clock.
- Set the peers and the sync mode:
w32tm /config /manualpeerlist:"<peers>" /syncfromflags:manual /update. - Restart the service with
net stop w32timethennet start w32time, which is the documented way to make the change take. - Force a synchronisation with
w32tm /resync /rediscoverand read the status.
It is synchronised but not advertising
You have this one if The status output looks correct, and domain controllers below it still report no usable source.
- Confirm the NTP server provider is enabled:
Enabledset to 1 underHKLM\SYSTEM\CurrentControlSet\Services\W32Time\TimeProviders\NtpServer. - Set
AnnounceFlagsto 5 under...\W32Time\Configon the PDC emulator. Microsoft’s guidance is 0xA instead where the server does not synchronise with an upstream source, or synchronises at a fixed interval through SpecialPollInterval. - Restart the service and check whether the advertisement events stop being written.
A hypervisor is setting the clock underneath the service
You have this one if The domain controller is virtual, and the time changes in ways the configuration does not explain.
- Disable the time synchronisation integration service for that virtual machine on the host.
- Restart the guest and confirm the time source is the peer you configured, not the host’s provider.
- Leave the guest on the normal Windows Time configuration afterwards; the point is one source, not none.
Microsoft’s guidance for a virtualised domain controller is to partially disable host-guest time synchronisation, so the guest follows the domain hierarchy and is protected from a jump if it is restored from a saved state.
The peers are unreachable rather than unconfigured
You have this one if The configuration is right, and the status shows no successful synchronisation.
- Confirm UDP 123 is open outbound to the peers you chose, in both directions for the reply.
- Test a specific peer with
w32tm /stripchart /computer:<peer> /dataonly /samples:5and read the offsets. - Where the peers resolve but never answer, choose sources your firewall team will actually allow, including an appliance on your own network.
Full reference
The registry values behind the commands
| Value | Key | What to set on the forest root PDC emulator |
|---|---|---|
Type |
...\W32Time\Parameters |
NTP, so it uses the manual peer list rather than the domain hierarchy |
NtpServer |
...\W32Time\Parameters |
The peer list, space-delimited, with a flag appended to each name – the authoritative time server guidance uses ,0x1 |
AnnounceFlags |
...\W32Time\Config |
5, or 0xA where the server does not synchronise upstream or uses SpecialPollInterval |
Enabled |
...\W32Time\TimeProviders\NtpServer |
1, so the machine answers time requests from the domain |
MaxPosPhaseCorrection / MaxNegPhaseCorrection |
...\W32Time\Config |
The largest correction the service will make. 48 hours on domain controllers by default |
All of these sit under HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\W32Time. Restart the service after changing them – net stop w32time && net start w32time – because several are only read at start-up.
The advertisement events, in order
| Event ID | Published text |
|---|---|
| 139 | The time service has started advertising as a time source |
| 140 | Stopped advertising because the local machine is not an Active Directory domain controller |
| 141 | Stopped advertising because there are no providers running |
| 142 | Stopped advertising because the local clock is not synchronised |
| 143 | Started advertising as a good time source |
| 144 | Stopped advertising as a good time source |
| 145 | Stopped advertising as a time source |
These are about whether this machine will serve time to others, not about whether it can get time for itself. A domain controller that logs 142 has lost its own source; a domain full of clients that cannot find a source is usually the consequence, one layer down.
What the rest of the forest should look like
- Domain members and non-PDC domain controllers stay on NT5DS and follow the hierarchy. Do not configure manual peers on them.
- A domain controller that is not the PDC emulator uses AnnounceFlags 10 (0xA) rather than 5.
- Every domain controller should have the NTP server provider enabled, so it can answer the machines below it.
- Check the whole domain at once with
w32tm /monitor, which reports offsets for the domain controllers in the domain you run it in. Run it in each domain if you have several. dcdiag /test:Advertisingconfirms a domain controller is advertising the services it should, which includes the time service on the role holder.
When the role moves
The time configuration follows the machine, not the role. Transfer the PDC emulator to another domain controller and the new holder is at the top of the hierarchy with whatever configuration it happened to have – usually NT5DS, pointing at nothing. Either script the reconfiguration into your role transfer procedure or, at minimum, add a check of w32tm /query /configuration on the new holder to the same runbook.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
Event ID 12 |
Logged by the Windows Time service on a machine following the domain hierarchy with nothing above it to follow. No Microsoft page publishing this ID’s text was found, so confirm the condition with w32tm /query /source and /configuration rather than on the number alone |
not published by the vendor |
Event ID 47 |
Written by the time service when a configured peer stops being usable as a source. No published text was found for this ID; check the peer with w32tm /stripchart instead |
not published by the vendor |
Event ID 29 |
No per-event text is published. Microsoft documents 24, 29 and 38 together as the set logged on a virtualised domain controller that is synchronising from both the hypervisor and the domain hierarchy | not published by the vendor |
Event ID 144 |
The time service has stopped advertising as a good time source | Microsoft Learn |
Confirm the fix worked
w32tm /query /statuson the forest root PDC emulator names an external peer and a recent successful sync.w32tm /query /configurationshows Type NTP on that machine, and NT5DS everywhere else.w32tm /monitorfrom a domain-joined machine shows small offsets across the domain controllers.- No new advertisement events in the 140 to 145 range on the role holder.
- On a virtual domain controller, the host’s time synchronisation integration is off and the source is still the configured peer after a reboot.
Questions people ask about this
Which machine needs external time sources?
The one holding the PDC emulator role in the forest root domain. Everything else follows the hierarchy, and configuring peers elsewhere creates competing sources.
Does this need a paid time source?
No. Any reachable NTP source will do, including an appliance on your own network. Nothing here is a licensing question.
Why is AnnounceFlags sometimes 5 and sometimes 0xA?
Microsoft’s guidance is 5 on the authoritative server, and 0xA where that server does not synchronise with an upstream source or synchronises at a fixed interval using SpecialPollInterval.
Our domain controllers are virtual. Is host time synchronisation acceptable?
Not alongside the domain hierarchy. Microsoft’s guidance is to partially disable host-guest time synchronisation for a virtualised domain controller so it follows the hierarchy and is protected from time jumps when it is restored from a saved state.
What happens if I ignore this?
Nothing, until the hardware clock drifts or the machine is restored. Then the whole forest follows an unchecked clock, and Kerberos failures appear across everything at once.
