# Infrastructure templates for an external Prometheus. App alerts: ../../observe/
# Opt-in Prometheus alert rules; not a Luma deployment manifest.
groups:
  - name: luma-infrastructure
    rules:
      - alert: LumaControlScrapeUnavailable
        expr: up{job="luma-control"} == 0
        for: 2m
        labels:
          severity: critical
        annotations:
          summary: Luma Control metrics are unavailable
          description: Check Control connectivity, TLS and metrics credentials; this does not prove application downtime.
      - alert: LumaNodeHeartbeatMissing
        expr: luma_node_agent_up == 0
        for: 2m
        labels:
          severity: warning
        annotations:
          summary: 'Node {{ $labels.node }} stopped reporting'
          description: Check the agent and node connectivity before judging application availability.
      - alert: LumaFilesystemAlmostFull
        expr: luma_node_filesystem_used_ratio >= 0.90
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: 'Node {{ $labels.node }} filesystem is over 90% full'
          description: 'Inspect the filesystem containing {{ $labels.path }}; remote mounts and Docker VM disks are separate.'
      - alert: LumaFilesystemInodesAlmostFull
        expr: luma_node_filesystem_inodes_used_ratio >= 0.90
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: 'Node {{ $labels.node }} inode capacity is over 90% used'
      - alert: LumaTaskQueueDelayed
        expr: luma_task_queue_oldest_age_seconds >= 600
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: 'Luma {{ $labels.kind }} queue has a task waiting over ten minutes'
          description: Compare queue demand with builder/agent availability; task kinds can refer to the same workflow.