Fix it now
Error 8614 means the time since the last replication with this partner has exceeded the tombstone lifetime, so the directory has stopped replicating with it. Microsoft documents two action plans and recommends the graceful one. Do not open by force-demoting: check for lingering objects first, then lift the quarantine deliberately.
repadmin /showattr <dc> "CN=Directory Service,CN=Windows NT,CN=Services,CN=Configuration,DC=example,DC=com" /atts:tombstonelifetime
repadmin /showrepl * /csv > showrepl.csv
repadmin /replsummary
- Read the forest’s tombstone lifetime from that first command. If the attribute is absent, an internal default is in use; Microsoft states the default is 60 or 180 days depending on the Windows version deployed in the forest.
- Open showrepl.csv in a spreadsheet, sort by last replication success, and find every DC that has been failing, not just the one that reported 8614.
- Check for a time jump before assuming an absence. A clock that moved a tombstone lifetime or more, in either direction, produces this error on a DC that was never offline.
- Find and remove lingering objects before you lift anything:
repadmin /removelingeringobjects <DC holding them> <GUID of a DC with a correct writable copy> <naming context> /advisory_mode, then again without advisory mode. - Only then lift the quarantine:
repadmin /regkey <dc> +allowDivergent, let replication run, fix whatever was breaking it, and remove the value again with-allowDivergent.
Microsoft states plainly that you should not assume force demotion is the only workable solution, especially when the underlying replication failure is easy to resolve or is outside Active Directory.
If replication resumes and the DC converges, remove the divergent-partner setting and stop here. The next section explains what the quarantine is protecting you from and when demotion really is the answer.
Why it happens
Deletions in Active Directory replicate as markers rather than as absences. When an object is deleted, a stub of it remains so every controller learns about the deletion, and that stub survives for the tombstone lifetime before garbage collection removes it. A controller that has not replicated for longer than that lifetime has missed the entire window in which it could have been told, so it may still hold objects everybody else deleted.
Error 8614 is that risk being refused. Event ID 2042 says it in words: it has been too long since this machine last replicated with the named source machine, the time between replications has exceeded the tombstone lifetime, and replication has been stopped with this source, because the two machines’ views of deleted objects may now differ and the source might return objects that have already been deleted.
The important thing about that refusal is what it is for. It is a checkpoint, not a verdict. Microsoft’s guidance is explicit: the point of the quarantine is to check for lingering objects and remove them if present before allowing replication to resume, and you should not assume a force demotion is the only workable solution. There are two documented action plans – force demotion, which is faster and easier and risks losing originating updates on the demoted DC, and a graceful recovery, which is the one Microsoft walks through in detail.
There is also a second root cause that has nothing to do with a machine being switched off. A system time jump of a tombstone lifetime or more, forwards or backwards, produces the appearance of failed replication. A DC that adopted a bad clock, replicated successfully, and then tried to replicate from a partner it last spoke to a long apparent time ago will report 8614 while having been online the whole time. Force-demoting that machine solves nothing and loses whatever it originated.
The DC really was out of contact for longer than the tombstone lifetime
You have this one if Event ID 2042 names the interval, and the machine’s power or uptime history matches a long absence.
- Work the graceful plan in order. Read the forest tombstone lifetime, confirm it has not been shortened from the default.
- Run
repadmin /showrepl * /csvacross the forest and find every DC in the same state, not just this one. - Remove lingering objects with
repadmin /removelingeringobjects, in advisory mode first, for each affected partition. - Set
+allowDivergenton the destination, let replication run, then remove it again once the queue has drained. - If the DC has originated nothing you would miss and the estate is large, force demotion remains a legitimate second option.
Whichever route you take, the underlying replication failure that let this happen still has to be fixed, or it will claim the next controller.
The system clock moved a tombstone lifetime or more
You have this one if The DC has been running continuously and its time is wrong by a very large margin, often after a hardware clock or virtualisation problem.
- Look for the signature in the logs: date stamps predating the operating system release or the installation, future-dated stamps, or gaps where events should be.
- Correct the time hierarchy so the forest root PDC emulator is the authority and everything else follows it, then resync.
- Check the virtual host’s time synchronisation settings, which are a frequent source of large jumps.
- Then work the graceful recovery. The DC’s data is probably fine; it is the arithmetic that broke.
Demoting a healthy DC because its clock was wrong is a real and avoidable loss. Establish which of the two causes you have before you touch it.
Replication has been failing for months and nobody was watching
You have this one if repadmin /showrepl * /csv shows several partners stale, and 8614 is simply the first error anyone looked at.
- Deal with the quarantined DC on its own merits, but do not stop there.
- Fix the underlying replication failure on the remaining controllers before it claims another one.
- Use Microsoft’s own thresholds for triage: a DC that has not replicated within 50 percent of the tombstone lifetime needs a strong push; one past 90 percent should be considered for force demotion.
- Put
repadmin /showrepl * /csvon a schedule so a stale partner raises an alert rather than an incident.
Full reference
Microsoft’s two action plans, side by side
| Force demotion | Graceful recovery | |
|---|---|---|
| Speed | Faster and easier | Slower, several steps |
| Risk | Can lose originating updates on the demoted DC | Preserves what the DC originated |
| Tooling | dcpromo /forceremoval, or Uninstall-ADDSDomainController -ForceRemoval |
repadmin, plus two registry settings |
| What Microsoft says | One of two options | Documented in detail, with the instruction not to assume demotion is the only workable solution |
| When it fits | The DC originated nothing you would miss, or is old and due for replacement anyway | The underlying replication failure is easy to fix, or is external to Active Directory |
The graceful recovery, in the documented order
- Check for a non-default tombstone lifetime with
repadmin /showattragainst the Directory Service object in the configuration partition. If the attribute is absent, an internal default applies. - Find every DC that has failed inbound replication for that long:
repadmin /showrepl * /csv, opened in a spreadsheet and sorted by last success. - Check for time jumps: date stamps predating the OS release or its installation, future-dated stamps, missing event logs across a range.
- Find and remove lingering objects with
repadmin /removelingeringobjects, advisory mode first. - Evaluate strict replication consistency on the destinations:
repadmin /regkey <DSA_LIST> +strict. - Disable the time-based quarantine:
repadmin /regkey <dc> +allowDivergent, orrepadmin /regkey * +allowDivergentacross the forest. - Resolve the replication failures that were being masked. The 8614 may have been hiding a tombstone lifetime’s worth of other errors.
- Remove the divergent-partner setting again, or set it to 0.
- Monitor replication daily from then on.
The two registry values these steps set
| Value | Path | Effect | repadmin equivalent |
|---|---|---|---|
Allow replication with divergent and corrupt partner |
HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters |
0 disallows, 1 allows. Lifts the time-based quarantine | +allowDivergent / -allowDivergent |
Strict Replication Consistency |
HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters |
REG_DWORD, 1 enabled, 0 disabled | +strict / -strict |
Lift the quarantine only after you have checked for lingering objects, and remove the setting again afterwards. Left in place, it disables the protection permanently on that controller, and the next time a partner falls behind you will get the reanimation the quarantine exists to prevent, silently.
Tombstone lifetime, read rather than assumed
Microsoft states that the default is either 60 or 180 days, depending on the Windows version deployed in your forest, so quoting one number is wrong roughly half the time. Read it:
| Command | Note |
|---|---|
repadmin /showattr <dc> "CN=Directory Service,CN=Windows NT,CN=Services,CN=Configuration,DC=<root>,DC=<tld>" |
Shows the whole object, including tombstoneLifetime if it is set |
... /atts:tombstonelifetime |
Narrows it to the one attribute |
If the attribute is not present in the output, the forest is using an internal default rather than an explicit value. That is normal, and it is also why you cannot tell what your threshold is by looking at the forest’s age.
Watch thresholds worth putting in monitoring
- A DC that has not replicated within 50 percent of the tombstone lifetime deserves a priority push to resolve whatever is wrong.
- A DC that has not replicated within 90 percent of it should be considered for force demotion before it crosses the line.
- Monitor replication rather than servers. A controller that pings happily can be failing replication for months.
repadmin /showrepl * /csvis the documented way to collect that, and it goes into a spreadsheet cleanly.
Two events people expect to find here
Event ID 1864 and Event ID 2089 are widely quoted in this context, as a periodic summary of partner staleness and a warning about partitions not backed up recently. Microsoft does not publish either of them in a place this article could point you at, so they are recorded below as unpublished. The documented way to get the first is repadmin /showrepl * /csv, and it is better than an event because it covers the whole forest at once.
When a licence is the actual fix
The recovery itself is free, and if the graceful route works you will not have replaced anything. What often comes with an 8614 is a decision about the machine, because a controller that sat unreachable for months is usually an old one in a forgotten location, and rebuilding it on the same unsupported build resets the clock on the same problem. If you do replace it, Windows Server 2025 Standard is the straightforward choice for a domain controller. Arco can supply the licence and, before you buy, confirm whether the machine is already covered by an agreement you hold. Two things worth settling in the same conversation: Standard limits how many Windows virtual instances you may run on the licensed hardware while Datacenter does not, so which is cheaper depends on your density rather than on a rule of thumb; and Client Access Licences are counted separately from the server licence itself.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
8614 |
ERROR_DS_REPL_LIFETIME_EXCEEDED: the directory service cannot replicate with this server because the time since the last replication has exceeded the tombstone lifetime | Microsoft Learn |
Event ID 2042 |
It has been too long since this machine last replicated with the named source machine; replication has been stopped with this source because the two views of deleted objects may now differ | Microsoft Learn |
Event ID 1864 |
Quoted as a periodic summary of partners that have not replicated within given intervals. No published meaning found; use repadmin /showrepl * /csv, which Microsoft does document for this |
not published by the vendor |
Event ID 2089 |
Quoted as a warning that a directory partition has not been backed up recently. No published meaning found; check your backup schedule against the tombstone lifetime directly | not published by the vendor |
Confirm the fix worked
repadmin /replsummaryshows the previously quarantined partner replicating in both directions.- Re-run
repadmin /removelingeringobjectsin advisory mode for each partition and confirm it reports nothing left to remove. - Confirm the
Allow replication with divergent and corrupt partnervalue has been removed or set to 0 on every DC you set it on. - Confirm
Strict Replication Consistencyis enabled across the forest. - Re-check
repadmin /showrepl * /csva week later and confirm no partner is drifting back towards the threshold.
Questions people ask about this
Do I have to demote the DC?
No, and this is where most advice on 8614 goes wrong. Microsoft documents two action plans and says explicitly not to assume force demotion is the only workable solution, particularly when the underlying replication failure is easy to resolve or is outside Active Directory. The graceful route is the one it walks through in detail.
Is it safe to set the divergent-partner registry value?
It is a documented step of Microsoft’s own recovery, taken after lingering objects have been found and removed, and removed again once replication has resumed. What is not safe is setting it first, or leaving it set, because then you have disabled the protection rather than passed through it.
What is my tombstone lifetime?
Read it rather than assume it. Microsoft states the default is 60 or 180 days depending on the Windows version deployed in the forest, and it can be changed. repadmin /showattr <dc> "CN=Directory Service,CN=Windows NT,CN=Services,CN=Configuration,DC=..." /atts:tombstonelifetime gives you the answer; an absent attribute means an internal default is in use.
The DC was never offline. How can it have exceeded the lifetime?
A system time jump. Microsoft names it as one of the two root causes: a clock that moved a tombstone lifetime or more, in either direction, creates the appearance of failed replication on a machine that was up the whole time. Look for date stamps that predate the OS installation, or lie in the future.
Can I not just extend the tombstone lifetime and reconnect it?
Raising the value now does nothing about deletion markers that have already been garbage collected. It changes how long future ones are kept. Work the documented recovery instead.
