Availability group checks
The sqlserver_ag module runs these checks against each availability group, through its listener when one is known. The rule, interval and default severity below are generated from the product's check catalog at build time, so they always match the version you are reading. Thresholds are defaults; your entities and resources may override them.
| Check | Every | Default severity | Rule |
|---|---|---|---|
sqlserver_ag.replica_stateReplica role and synchronization state | 30s | critical | synchronization_health_desc = HEALTHY, connected_state_desc = CONNECTED |
sqlserver_ag.database_syncDatabase synchronization, queues and data loss | 30s | warning | warning > 51200 kilobytes · 5m / critical > 512000 kilobytes · 5m |
sqlserver_ag.secondary_lagSecondary lag (estimated data loss) | 30s | warning | warning > 60 seconds / critical > 300 seconds |
sqlserver_ag.role_changeRole change / failover detected | 30s | critical | any change |
sqlserver_ag.listener_dnsListener IP online | 60s | critical | state_desc = ONLINE |
sqlserver_ag.backup_preferenceBackup preference honoured | 3600s | warning | matches_preference = 1 |
Checks
sqlserver_ag.replica_state
Replica role and synchronization state
- Measures
replica_state· every30s·tsql- Rule
synchronization_health_desc = HEALTHY, connected_state_desc = CONNECTED· Default severity: critical- Cause
- A replica lost its connection to the primary, or its synchronization health is partially healthy or not healthy.
- Impact
- A replica that is not healthy cannot take over without data loss; a disconnected one cannot take over at all.
- Action
- Check the replica's SQL Server error log and the cluster's health; look for network, endpoint or service account problems between the replicas.
- Baseline
- PARTIALLY_HEALTHY → warning; NOT_HEALTHY or DISCONNECTED → critical. · source
sqlserver_ag.database_sync
Database synchronization, queues and data loss
- Measures
redo_queue_kb(kilobytes) · every30s·tsql- Rule
warning > 51200 kilobytes · 5m / critical > 512000 kilobytes · 5m· Default severity: warning- Cause
- The secondary receives or applies log more slowly than the primary produces it: a busy primary, a slow network or slow disks on the secondary.
- Impact
- The send queue is data a failover would lose; the redo queue is time a failover would take to recover.
- Action
- Compare log send rate and redo rate with the primary's log generation; check the secondary's disk latency and the network between replicas.
- Baseline
- Redo queue above 50 MB sustained indicates the secondary cannot keep up; 500 MB is critical. Send queue uses 10 MB / 100 MB.
sqlserver_ag.secondary_lag
Secondary lag (estimated data loss)
- Measures
secondary_lag_seconds(seconds) · every30s·tsql- Rule
warning > 60 seconds / critical > 300 seconds· Default severity: warning- Cause
- The secondary is behind the primary by more seconds than your RPO allows.
- Impact
- If the primary failed now, you would lose about this many seconds of committed transactions.
- Action
- Compare with the entity's RPO. Synchronous replicas should stay near zero; for asynchronous replicas, investigate log send and redo rates.
- Baseline
- Compare with the entity RPO; synchronous replicas should stay near 0.
sqlserver_ag.role_change
Role change / failover detected
- Measures
role· every30s·tsql- Rule
any change· Default severity: critical- Cause
- The primary moved to another replica: a planned failover, an automatic failover or a forced one.
- Impact
- Applications reconnect; connections that do not use the listener may break.
- Action
- Confirm whether the failover was planned; check cluster and SQL error logs on both nodes; verify listener DNS.
- Baseline
- Any role change raises an event that stays open until someone resolves it; critical when it happened outside a maintenance window.
sqlserver_ag.listener_dns
Listener IP online
- Measures
listener_state· every60s·tsql- Rule
state_desc = ONLINE· Default severity: critical- Cause
- No listener IP is online, or Excubius cannot reach the IP of the primary's subnet.
- Impact
- Applications that connect through the listener cannot reach the group.
- Action
- Check the listener resource in the cluster; in multi-subnet groups, allow the Excubius server to reach every listener IP.
- Baseline
- One listener IP must be ONLINE on the primary; synthetics.ag_listener_resolve also resolves the DNS name from the collector.
sqlserver_ag.backup_preference
Backup preference honoured
- Measures
backup_preference· every3600s·tsql- Rule
matches_preference = 1· Default severity: warning- Cause
- The primary took full backups of the group's databases in the last 26 hours although it is not their preferred backup replica.
- Impact
- Backup jobs that respect the preference may silently skip, or backups load the primary.
- Action
- Run full backups on the replica the group's automated backup preference names, or change the preference.
- Baseline
- Backups should run where the AG preference says, otherwise jobs silently skip.
Related content
- Prerequisites for availability groups
- Entities, tags and inheritance: how a default threshold is overridden