---
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: thanos.rules
  namespace: monitoring
spec:
  groups:
    - name: thanos.rules
      rules:
        - alert: ThanosCompactionHalted
          expr: |
            thanos_compactor_halted == 1
          for: 0m
          labels:
            severity: critical
          annotations:
            summary: "Thanos compaction halted on {{ $labels.instance }}"
            description: "Thanos compaction has failed to run and is now halted.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
        - alert: ThanosCompactBucketOperationFailure
          expr: |
            rate(thanos_objstore_bucket_operation_failures_total[1m])
              > 0
          for: 0m
          labels:
            severity: critical
          annotations:
            summary: "Thanos compact bucket operation failure on {{ $labels.instance }}"
            description: "Thanos compaction has failing storage operations\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
        - alert: ThanosCompactNotRun
          expr: |
            (time() - thanos_objstore_bucket_last_successful_upload_time)
              > 24*60*60
          for: 0m
          labels:
            severity: critical
          annotations:
            summary: "Thanos compact not run on {{ $labels.instance }}"
            description: "Thanos compaction has not run in 24 hours.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"