Requirements and boundaries
Service monitoring is opt-in. It requires Linux agent 0.9.0 or Windows agent 0.11.0 or later with this capability. It is not offered on macOS. No service is selected automatically. Allow two check-ins after selecting names or upgrading an older agent.
The collector queries state with its existing restricted account. It does not restart a service, execute an arbitrary script or run a remediation action. An unreadable service stays Unknown.
Find the exact service name
On Ubuntu, list loaded system services and inspect a candidate:
systemctl list-units --type=service --all
systemctl show ssh.service --property=LoadState,ActiveState,SubState
On Windows, use the Name field, not DisplayName:
Get-Service | Select-Object Name,DisplayName,Status
Get-Service -Name Spooler
ssh.service and Spooler are examples, not default recommendations. Only choose services expected to remain running on that device. A demand-start or intentionally idle service may stop normally. Linux unit names are case-sensitive and must end in .service; Windows names are case-insensitive. Linux user-session units, oneshot jobs, sockets and timers are excluded. Unusual escaped or Unicode names are not supported in this release.
Configure the selection
Open Policies → Services for organization or site rules, or Monitoring policy → Services for a machine override. Enter up to ten exact names per platform, one per line. Choose a stopped duration from 1 to 60 minutes; the default is one minute. Empty lists disable service monitoring.
The complete selection and duration inherit together. A machine override replaces the shared selection. Review the impact preview and current rule source before saving.
Interpret the timing
Stopped, Failed, Paused or Not installed observations start a timing window. An incident opens after the configured elapsed duration with fresh observations no more than 90 seconds apart. The default one-minute rule typically needs one or two check-in intervals, plus worker time.
Unknown, access denied, unsupported, incomplete, stale and transition states reset a pending window. They do not create a stopped-service incident or falsely recover an active one. A confirmed Running reading closes the incident immediately and cancels unsent mail.
Each service has its own incident. Brief failures between polling intervals can be missed. An offline device follows its offline policy; service delivery waits for fresh service readings.
Test safely
Review a healthy observation first. For an end-to-end stop/recovery test, use a disposable service on an authorized test device. Do not stop SSH, security software or a business-critical service just to exercise an alert. Verify the selected name, incident duration, one notification attempt and confirmed recovery.
References: systemctl queries, Microsoft Get-Service.