DAG problems are usually cluster problems first
A Database Availability Group is an Exchange abstraction layered on a Windows Server Failover Cluster. When a database fails to activate or a copy stays in a failed state, the root cause is frequently the underlying cluster — lost quorum, a down witness, or a node that has been quarantined — rather than Exchange itself. The discipline that resolves DAG incidents fastest is to verify the cluster foundation before chasing mailbox database copy status.
This article covers the DAG problems I encounter most: failed or suspended database copies, copy and replay queue growth, witness server failures, and split-brain after a data-centre partition. For the DAG architecture and design choices, see the companion guide on Exchange failover clustering and DAGs.
Problem 1: Database copy stuck in Failed or Suspended state
A database copy showing Failed or Suspended means replication has stopped and the copy is not a viable failover target. The first diagnostic is `Get-MailboxDatabaseCopyStatus` — the Status and ErrorMessage fields tell you whether the failure is storage, network, or the log stream itself.
A common cause is a broken replication network or a full log drive on the passive node. If the log drive fills, the passive copy cannot replay logs and suspends. Clear space, resume the copy, and let it catch up before considering a re-seed.
Get-MailboxDatabaseCopyStatus * |
Select-Object Name, Status, CopyQueueLength, ReplayQueueLength, ErrorMessage |
Format-Table -AutoSize
# Resume a suspended copy after resolving the underlying cause
Resume-MailboxDatabaseCopy -Identity "DB01\EXCH-MB02"
# If the copy is corrupt or too far behind, re-seed:
Update-MailboxDatabaseCopy -Identity "DB01\EXCH-MB02" -DeleteExistingFilesProblem 2: Copy queue and replay queue growth
A growing copy queue (log files waiting to be shipped) or replay queue (log files received but not replayed) indicates the passive node cannot keep pace. Left unchecked, a failover to that copy will lose data or take an unacceptable time to mount.
Copy queue growth points to network throughput or the replication service; replay queue growth points to disk I/O on the passive node. Diagnose with the performance counters under MSExchange Database ==> Instances (log replay bytes/sec) and the replication network throughput. A replay queue consistently above a few hundred MB on a synchronous copy warrants investigation before the next failover.
| Symptom | Likely bottleneck | Action |
|---|---|---|
| Copy queue grows, replay queue stable | Replication network or service | Check network, restart replication service |
| Replay queue grows, copy queue stable | Passive disk I/O | Investigate disk latency, add spindles |
| Both grow after a restart | Normal catch-up | Monitor; should drain within SLA window |
| Queue grows on one copy only | Node-specific storage | Check that node’s disk health |
Problem 3: Witness server and quorum loss
A two-node DAG relies on a witness server to maintain quorum. If the witness is down and one DAG member fails, the remaining node loses quorum and every database dismounts — even though a healthy copy exists. This is the most common cause of a total mail outage on small DAGs.
The witness must be a server that is not a DAG member (typically a standalone Exchange server or a third node), reachable from all DAG members, and granted the necessary permissions. Monitor witness reachability as part of DAG health, and never let a two-node DAG run without a functional witness.
- Use a witness server that is not a DAG member and is always reachable
- Monitor quorum state with Get-DatabaseAvailabilityGroup -Status
- For multi-site DAGs, place the witness in the primary data centre
- Alert immediately if the cluster loses quorum or a node is quarantined
Problem 4: Split-brain after a data-centre partition
Without Datacenter Activation Coordination (DAC) mode, a network partition between two data centres can let both sites mount the databases independently — split-brain — producing divergent mail data that cannot be reconciled. DAC mode prevents this by requiring a quorum of DAG members plus the witness before databases can mount.
If split-brain has already occurred, recovery means choosing the authoritative site, dismounting and discarding the divergent copies, and re-seeding the discarded site from the authoritative copy. Enable DAC mode on every multi-site DAG before a partition happens; recovering after the fact is costly and risky.
Get-DatabaseAvailabilityGroup -Identity DAG01 -Status |
Select-Object Name, DatacenterActivationMode, WitnessServer, WitnessDirectory,
Servers, OperationalServers
# Cluster quorum and node state
Get-ClusterNode | Select-Object Name, State, FaultDomain
Get-ClusterQuorumTrade-offs and next steps
DAG design balances resilience against complexity. Three or more members with a witness give the most stable quorum; two members are acceptable only with a reliable witness and DAC mode. Blocking automatic activation on the DR copy makes a site failover deliberate but requires a runbook and a tested procedure — an untested manual failover is a liability during an incident.
When problems recur, the next step is usually to harden the foundation: add a third DAG member or a cloud witness for quorum, dedicate a replication network, and ensure the passive storage matches the primary’s performance. Pair the DAG with Exchange-aware backups and a documented DR runbook — the DAG provides availability, but only backups provide point-in-time recovery.
- Prefer three or more DAG members for stable quorum
- Block automatic activation on the DR copy and test the manual failover runbook
- Dedicate a replication network and monitor queue depth continuously
- Pair the DAG with Exchange-aware backups for point-in-time recovery
Lessons learned
- Verify cluster quorum before chasing database copy status
- A full log drive suspends the passive copy — clear space before re-seeding
- Never run a two-node DAG without a functional witness
- Enable DAC mode on every multi-site DAG before a partition happens
Written by
Jamshed Alam
Global MSP & IT Infrastructure Consultant with 20 years across Exchange, Microsoft 365, Azure, security and automation. Available for project and contract work worldwide.