Fix it now
Microsoft publishes 8456 as the source server currently rejecting replication requests and 8457 as the destination server currently rejecting them. Which number you have tells you which machine to look at, and checking the wrong one is the most common way to conclude, wrongly, that nothing is set.
repadmin /options <dc>
repadmin /replsummary
repadmin /queue
- Read the error first. An 8456 reported by a destination points at the source, because it is the source that is refusing to send. An 8457 points at the destination itself.
- Before changing anything, read the Directory Service log on the controller with the flag set. If replication was disabled because the directory detected a USN rollback, stop and read the section on that below.
- If a person or a tool set the flags, clear the one that is set:
repadmin /options <dc> -DISABLE_INBOUND_REPLorrepadmin /options <dc> -DISABLE_OUTBOUND_REPL. - Confirm with
repadmin /options <dc>and no arguments that the flag is gone. - Force a sync and watch the backlog drain with
repadmin /queue, then confirm withrepadmin /replsummary.
repadmin /options and the DISABLE_INBOUND_REPL and DISABLE_OUTBOUND_REPL flag names could not be confirmed against a Microsoft documentation page for this article. Confirm the exact syntax against repadmin /? on your own machine before running it in anger.
If the queue drains and both partners report a recent success, you are done. The next section covers where the switch lives and the one case where clearing it causes real damage.
Why it happens
Each controller carries a set of flags on its own settings object, and two of them control replication directly. One stops the controller accepting inbound changes; the other stops it sending outbound changes. They exist for good reasons. An authoritative restore sets them so restored data is not overwritten before the operator is ready. Migrations and bulk imports use them so a large change can be staged. The trouble is that they persist until somebody removes them, and nothing about day-to-day operation reminds you they are there.
Which error you get tells you which end to look at, and that saves a lot of time. Microsoft publishes 8456 as ERROR_DS_DRA_SOURCE_DISABLED, “The source server is currently rejecting replication requests”, and 8457 as ERROR_DS_DRA_SINK_DISABLED, “The destination server is currently rejecting replication requests”. A destination reporting 8456 is telling you the flag is on the other machine. Checking only the machine that reported the error is how people conclude nothing is set.
Maintenance flags were never cleared
You have this one if The flags are present, an authoritative restore or a bulk import happened recently, and the Directory Service log shows nothing alarming.
- Confirm the current options on both ends before changing anything.
- Clear the flag that is actually set, on the machine that actually has it, and only that one.
- Force replication and watch the queue drain rather than assuming it has.
- Write down that you cleared it, and when, so the next person is not repeating this.
Clear one controller at a time and let the queue drain before moving on. Releasing several at once on a slow link produces a burst you will have to wait out anyway.
If the Directory Service log shows that replication was disabled because a USN rollback was detected, do not clear the flags. The directory set them because the controller was restored by an unsupported method, such as copying or reverting a virtual machine, and its update sequence numbers now conflict with what its partners have already recorded. Re-enabling replication on a controller in that state spreads inconsistent data across the forest.
The directory disabled replication itself after a rollback
You have this one if Directory Service entries recording a USN rollback, usually after a virtual machine was reverted to a checkpoint or a system state was restored incorrectly.
- Leave the flags exactly as they are. They are the protection working.
- Confirm the quarantine directly:
reg query "HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters" /v "Dsa Not Writable". A value of 0x4 confirms a detected rollback. - Follow the documented recovery for that condition rather than this article’s: remove the DC from the domain, or restore a good system state backup taken before the rollback.
- Review how the rollback happened, because reverting checkpoints on domain controllers by hand will keep producing it.
Modern hypervisors that expose a VM generation identifier to a supported guest handle a revert safely. Copying a virtual disk back by hand bypasses that entirely.
You are checking the wrong controller
You have this one if An 8456 on the destination, and reading the options on that destination shows nothing set.
- Read the published meanings again: 8456 says the source is rejecting requests, 8457 says the destination is.
- Read the options on the source controller named in the failure.
- Clear the outbound flag there, then retry from the destination.
- If the source is unreachable rather than refusing, you have a different problem and the options are a red herring.
The error is 8438 and nobody set anything
You have this one if Replication fails intermittently under load and the code is 8438 rather than 8456 or 8457.
- Read it correctly: the directory service is too busy to complete the replication operation at this time.
- Look at what the controller is doing rather than at its flags. This is a load condition.
- Check whether a hub bridgehead is serving more branch sites than it can, or whether site link schedules are firing more often than the link can carry.
- Retry the operation once the load has passed, and treat a persistent 8438 as a capacity problem.
This one is regularly listed alongside 8456 and 8457 as though it were another disable flag. It is not, and treating it as one sends you looking for a switch that was never thrown.
Full reference
The four codes, as published
| Code | Symbolic name | Published text | What it makes you do |
|---|---|---|---|
8456 |
ERROR_DS_DRA_SOURCE_DISABLED | The source server is currently rejecting replication requests | Go and look at the source |
8457 |
ERROR_DS_DRA_SINK_DISABLED | The destination server is currently rejecting replication requests | Look at the machine that reported it |
8477 |
ERROR_DS_DRA_REPL_PENDING | The replication request has been posted; waiting for reply | Nothing. This is a normal in-progress state |
8438 |
ERROR_DS_DRA_BUSY | The directory service is too busy to complete the replication operation at this time | Look at load and topology, not at flags |
Reading those four side by side is most of the article. Two of them describe a switch someone threw, one describes work in progress, and one describes a busy server. They are routinely presented as four faces of the same problem and they are not.
A note on the commands in this article
repadmin /options, and the flag names DISABLE_INBOUND_REPL and DISABLE_OUTBOUND_REPL, are the standard way to read and change this state, and they are what every experienced administrator will tell you to run. This article could not find a Microsoft documentation page that publishes that subcommand or those flag names, so it is telling you that rather than presenting them as documented. Two practical consequences:
- Run
repadmin /?on the machine in front of you and read the syntax it prints. That is authoritative for your build, and it takes ten seconds. - Read the current state before you write anything.
repadmin /options <dc>with no further arguments reports rather than changes, so there is no risk in looking. - The same caution applies to the
/syncallflag letters and to/queue, which this article uses for confirmation rather than for the fix itself. repadmin /replsummaryis documented by Microsoft among the commands that report replication status, and it is the safest thing to trust in this list.
Which end is refusing
| What you see | Where the flag is |
|---|---|
| 8456 reported by the destination | Outbound replication is disabled on the source |
| 8457 reported by the destination | Inbound replication is disabled on the destination itself |
| Both, on different partners | More than one controller has been left with flags set |
| 8477 in repadmin output | A request is in progress. Not a fault |
| 8438 under load | The directory is too busy. Not a flag |
| USN rollback entries in the Directory Service log | The directory set the flags itself. Do not clear them |
Telling a deliberate flag from a protective one
- Read the Directory Service log on the controller that holds the flag, for the period around when replication stopped.
- Check the quarantine value directly:
reg query "HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Parameters" /v "Dsa Not Writable". 0x4 means a rollback was detected. - Check whether the Net Logon service is paused on that controller. That is the documented consequence of a detected rollback and it is visible without reading a single event.
- If either of those is true, the flags are protection, not an oversight. Follow the USN rollback recovery instead.
- If neither is true, and the timing lines up with a restore, a migration or a maintenance window somebody ran, clearing the flag is routine.
Clearing flags without causing a second incident
- Clear one controller at a time and let its queue drain before moving to the next. A burst of catch-up traffic on a slow site link is still a burst.
- Do it inside a window if replication has been off for a long time, because the volume is proportional to the outage.
- Confirm afterwards with
repadmin /replsummaryrather than assuming, and check both directions. - Create a test object on one controller and confirm it appears on the previously blocked partner. That is the only proof that means anything.
Using the flags deliberately, later
Disabling replication for maintenance is a legitimate technique when a procedure calls for it: an authoritative restore, a staged bulk import, a migration where you want changes held until you are ready. The discipline that matters is writing down that you set it, on which machine and in which direction, and confirming the options are clear before you close the change. Almost every 8456 in the wild is a change record that ended one step early.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
8456 |
ERROR_DS_DRA_SOURCE_DISABLED: the source server is currently rejecting replication requests | Microsoft Learn |
8457 |
ERROR_DS_DRA_SINK_DISABLED: the destination server is currently rejecting replication requests | Microsoft Learn |
8477 |
ERROR_DS_DRA_REPL_PENDING: the replication request has been posted and is waiting for a reply. An in-progress state, not a failure | Microsoft Learn |
8438 |
ERROR_DS_DRA_BUSY: the directory service is too busy to complete the replication operation at this time. A load condition, not a disable flag | Microsoft Learn |
Confirm the fix worked
- Read the options on both controllers and confirm neither disable flag is listed.
repadmin /replsummaryreports a recent success for every partner, in both directions.- The replication backlog has drained rather than merely started moving.
- Create a test object on one controller and confirm it appears on the previously blocked partner.
- Confirm
Dsa Not Writableis not set under the NTDS Parameters key on either machine, so you know you cleared a maintenance flag and not a quarantine.
Questions people ask about this
Does anything here need to be bought?
No. This is a flag on an object in your own directory and a single command to change it. Nothing about it is a licensing question.
How do I know whether it is safe to clear the flags?
Read the Directory Service log first, and check Dsa Not Writable under the NTDS Parameters key. If the value is 0x4, the directory quarantined the controller after detecting a USN rollback, and clearing the flags spreads inconsistent data across the forest. If it is absent and the timing matches a maintenance window, clearing is routine.
I have 8438, not 8456. Is it the same thing?
No. 8438 is published as the directory service being too busy to complete the replication operation at this time. Nobody set a flag; the server is loaded. Look at bridgehead capacity and site link schedules rather than at options.
Will clearing them flood the network?
There will be a catch-up burst proportional to how long replication was off. On a slow site link, clear one controller at a time and let the queue drain rather than releasing everything at once.
Why does this article hedge on the repadmin syntax?
Because this rewrite could not find a Microsoft page that publishes repadmin /options or the flag names, and it would rather tell you that than present them as documented. Run repadmin /? on your own server: it prints the syntax for your build, and it is authoritative in a way an article cannot be.
