Fix it now
Monitoring is telling you a database has fewer usable copies than it is supposed to. Nobody is complaining yet, which is exactly the problem: this is an outage that has not happened. Read the copy status, fix whichever copy is not healthy, and check the backup as well as the copies.
Get-MailboxDatabaseCopyStatus * | Format-Table Name,Status,CopyQueueLength,ReplayQueueLength,ContentIndexState
Test-ReplicationHealth
Get-MailboxDatabase -Status | Format-List Name,LastFullBackup
- Read the event on your own server. Microsoft publishes no description for these entries, so the database and copy they name are the diagnosis, not the number.
- Treat any copy that is not healthy or mounted as the thing to fix. Suspended means resume it; failed means find the error that failed it first.
- Resume a suspended copy with Resume-MailboxDatabaseCopy -Identity DB01\SERVER2, and watch the copy and replay queues drain.
- If a copy cannot catch up, suspend it, reseed with Update-MailboxDatabaseCopy -DeleteExistingFiles -SourceServer <passive source>, then resume it.
- Check the health set as well as the raw event: Get-ServerHealth -Identity SERVER1 and Get-HealthReport -Identity SERVER1.
Do not put -ConnectionStatus on Get-MailboxDatabaseCopyStatus. Microsoft documents that parameter as deprecated and no longer used, and older runbooks still carry it.
If every copy is healthy with queues near zero and the backup is current, you are done. If the alert persists, the next section covers what the check is actually counting.
Why it happens
A database availability group provides protection through copies, and the monitoring counts how many of them are actually usable rather than how many exist. A database with three copies of which two have failed is one bad disk away from an outage while everyone is working normally and nobody has complained. That gap between apparent health and real resilience is the whole reason the check exists.
The check looks at more than the status word. The copy queue counts logs shipped to a copy but not yet copied across; the replay queue counts logs copied but not yet replayed into the database; the content index state says whether search will work if that copy is activated. A copy with a growing replay queue is healthy in name only, because activating it means waiting for that queue to drain first.
Protection also includes backups. A database with every copy healthy but no successful backup for weeks is not protected against logical damage, an accidental deletion or a site loss, which is why the alert sometimes fires when every copy looks perfect. Read LastFullBackup on the database itself rather than trusting the backup product’s own console.
Microsoft publishes no event descriptions for the entries this article covers. What it does publish is the tooling: Test-ReplicationHealth is documented as proactive monitoring of continuous replication and the replication pipeline, the availability of Active Manager, and the health and status of the underlying cluster service, quorum and network components, and it can be run locally or remotely against any Mailbox server in the DAG. That is the check to trust.
A copy was suspended for maintenance and never resumed
You have this one if The copy status is suspended and the timing matches a patch window or a hardware job.
- Resume the copy and watch the queues drain:
Resume-MailboxDatabaseCopy -Identity DB01\SERVER2. - If it has been suspended long enough that the source has truncated logs it still needs, it will not catch up and must be reseeded.
- Use the maintenance scripts that ship with Exchange to take a member in and out of service, so copies are handled consistently instead of by hand.
A suspended copy is not a harmless state. It blocks log truncation on the active server while it is suspended, so it can turn into a full log volume as well as a redundancy gap.
A copy failed on an error and stopped replicating
You have this one if The status is failed or failed and suspended, and a failure item was logged against that copy.
- Find the error before you touch the copy. A copy that failed on a storage fault will fail again immediately after a reseed onto the same disk.
- Look in the MailboxDatabaseFailureItems channel, which Microsoft documents as logging events associated with any failures affecting a replicated mailbox database.
- Once the cause is dealt with, suspend, reseed with
Update-MailboxDatabaseCopy -DeleteExistingFiles, and resume. - Watch the copy and replay queues afterwards until they settle near zero.
The content index for a copy has failed
You have this one if The database copy is healthy and the content index state is not, and search results are missing or stale on that copy.
- Where the search catalogue is a separate structure,
Update-MailboxDatabaseCopy -CatalogOnlyseeds the content index catalogue alone, which is far cheaper than rebuilding the whole copy. - Note that catalogue seeding always occurs over the MAPI network, even if you name a DAG network with -Network.
- Confirm the search services are running on that member and that the volume holding the index has free space.
If that copy is activated with a failed index, users get a working mailbox with broken search, which generates more helpdesk traffic than most outages.
Copies are healthy but backups are not
You have this one if Every copy is healthy with empty queues and the alert persists.
- Check when Exchange last recorded a backup:
Get-MailboxDatabase -Status | Format-List Name,LastFullBackup. - Fix the backup rather than the monitoring. Copies protect against a server or disk failure; they do not protect against something bad that replicates to all of them.
- Alert on the age of the last successful backup, not on whether the backup job reported success.
Full reference
Reading a copy status
| What you see | What to do |
|---|---|
| Healthy with queues near zero | Nothing. This copy is doing its job |
| Suspended | Someone suspended it, usually for maintenance. Resume it |
| Failed | Replication hit an error. Read the failure item before resuming |
| FailedAndSuspended | Deal with the underlying error first, then reseed. It will not resume on its own |
| ServiceDown | The Microsoft Exchange Replication service on that member is not running |
| Healthy, content index not healthy | The database is protected, search is not. Seed the catalogue |
Microsoft documents the -ConnectionStatus parameter on Get-MailboxDatabaseCopyStatus as deprecated and no longer used. If a script in your runbook still passes it, that script has not been reviewed since it mattered.
Where the evidence lives
| Crimson channel | What Microsoft says it holds |
|---|---|
HighAvailability |
Startup and shutdown of the Microsoft Exchange Replication service and its components, Active Manager role monitoring, database action events such as mount operations and log truncation, and events about the DAG’s underlying cluster |
MailboxDatabaseFailureItems |
Events associated with any failures that affect a replicated mailbox database |
ActiveMonitoring |
Definition and result events for Managed Availability probes, monitors and responders |
ManagedAvailability |
Recovery action logs and results, and related events |
All four sit under Applications and Services Logs, then Microsoft, then Exchange, in Event Viewer. When an event number carries no published description, these channels are where the actual explanation is: the failure item that suspended a copy will say what the copy hit, and the HighAvailability channel will say what Active Manager did about it.
Two scripts worth knowing
- CollectOverMetrics.ps1 reads DAG member event logs and gathers database operations – mounts, moves and failovers – over a period, recording the database, the start and finish times, the servers involved, the reason, whether it succeeded and the error detail if not. It writes one CSV per DAG.
- CollectReplicationMetrics.ps1 collects performance counter data about replication in real time from multiple Mailbox servers, and reports statistics such as how long each copy was failed or suspended and the average copy or replay queue length.
- Both ship in the Exchange Scripts folder. The first is the fastest way to answer ‘how often has this been happening’, which is usually the question behind a redundancy alert.
When every copy is healthy and the alert stays
- Check the backup age on the database, not in the backup console.
- Run Test-ReplicationHealth on every member, not just the one that raised the alert. It checks the cluster, quorum and network as well as the copies.
- Confirm the content index state on every copy you would be willing to activate.
- Remember that a lagged copy is deliberately behind. It protects against logical damage rather than against a server failure, and it cannot be activated without first replaying its queue.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
Event ID 2153 |
No description is published by Microsoft. Read the database and copy your own entry names, then check that copy with Get-MailboxDatabaseCopyStatus | not published by the vendor |
Event ID 4113 |
No description is published by Microsoft. Correlate it with the copy status for the database it names rather than treating the number as a diagnosis | not published by the vendor |
Event ID 4074 |
No description is published by Microsoft. Match it by database and timestamp against the HighAvailability and MailboxDatabaseFailureItems channels | not published by the vendor |
Event ID 3154 |
No description is published by Microsoft. Read it alongside the Active Manager entries in the HighAvailability channel for the same database | not published by the vendor |
Confirm the fix worked
Get-MailboxDatabaseCopyStatus *shows every copy healthy or mounted with copy and replay queues near zero.Test-ReplicationHealthpasses on every member of the group.- The content index state is healthy for every copy you would be willing to activate.
LastFullBackupon each database is recent, and the health report for the server has returned to healthy.
Questions people ask about this
Where is the official description of these events?
There is not one. Microsoft publishes no event descriptions for them, which is why this article works from copy status, Test-ReplicationHealth and the crimson channels instead of from the numbers.
Can I just suppress the alert?
You can, and you will then find out about the missing copies during the next hardware failure. The loss of redundancy is invisible in every other way: mail keeps flowing right up to the moment it does not.
How many copies should a database have?
Enough that losing one still leaves you protected. Adding a copy means adding a server, and that server needs its own Windows Server and Exchange licensing, so the decision is a real one rather than a checkbox.
Does a lagged copy count towards redundancy?
Not in the way a normal copy does. It is deliberately behind, so it protects against logical damage rather than against a server failure, and it cannot be activated without first replaying its queue.
Is a failed content index worth an alert?
Yes. If that copy is activated, users get a working mailbox with broken search. It is also cheap to fix: -CatalogOnly seeds the catalogue without rebuilding the database copy.
