feat: mixin / add loki compaction not successfull alert

production/loki-mixin-compiled-ssd/alerts.yaml

@@ -43,3 +43,36 @@ groups:
           for: 5m
           labels:
             severity: warning
+        - alert: LokiCompactorHasNotSuccessfullyRunCompaction
+          annotations:
+            description: |
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3 hours since the last compaction. This may indicate a problem with the compactor.
+            summary: Loki compaction has not run in the last 3 hours since the last compaction.
+          expr: |
+            # The "last successful run" metric is updated even if the compactor owns no tenants,
+            # so this alert correctly doesn't fire if compactor has nothing to do.
+            min (
+              time() - (loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{} > 0)
+            )
+            by (cluster, namespace)
+            > 60 * 60 * 3
+          for: 1h
+          labels:
+            severity: critical
+        - alert: LokiCompactorHasNotSuccessfullyRunCompaction
+          annotations:
+            description: |
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3h since startup. This may indicate a problem with the compactor.
+            summary: Loki compaction has not run in the last 3h since startup.
+          expr: |
+            # The "last successful run" metric is updated even if the compactor owns no tenants,
+            # so this alert correctly doesn't fire if compactor has nothing to do.
+            max(
+              max_over_time(
+                loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{}[3h]
+              )
+            ) by (cluster, namespace)
+            == 0
+          for: 1h
+          labels:
+            severity: critical

production/loki-mixin-compiled/alerts.yaml

@@ -43,3 +43,36 @@ groups:
           for: 5m
           labels:
             severity: warning
+        - alert: LokiCompactorHasNotSuccessfullyRunCompaction
+          annotations:
+            description: |
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3 hours since the last compaction. This may indicate a problem with the compactor.
+            summary: Loki compaction has not run in the last 3 hours since the last compaction.
+          expr: |
+            # The "last successful run" metric is updated even if the compactor owns no tenants,
+            # so this alert correctly doesn't fire if compactor has nothing to do.
+            min (
+              time() - (loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{} > 0)
+            )
+            by (cluster, namespace)
+            > 60 * 60 * 3
+          for: 1h
+          labels:
+            severity: critical
+        - alert: LokiCompactorHasNotSuccessfullyRunCompaction
+          annotations:
+            description: |
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3h since startup. This may indicate a problem with the compactor.
+            summary: Loki compaction has not run in the last 3h since startup.
+          expr: |
+            # The "last successful run" metric is updated even if the compactor owns no tenants,
+            # so this alert correctly doesn't fire if compactor has nothing to do.
+            max(
+              max_over_time(
+                loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{}[3h]
+              )
+            ) by (cluster, namespace)
+            == 0
+          for: 1h
+          labels:
+            severity: critical

production/loki-mixin/alerts.libsonnet

@@ -70,6 +70,53 @@
               |||, 'cluster', $._config.per_cluster_label),
             },
           },
+          {
+            // Alert if the compactor has not successfully run compaction in the last 3h since the last compaction.
+            alert: 'LokiCompactorHasNotSuccessfullyRunCompaction',
+            expr: |||
+              # The "last successful run" metric is updated even if the compactor owns no tenants,
+              # so this alert correctly doesn't fire if compactor has nothing to do.
+              min (
+                time() - (loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{} > 0)
+              )
+              by (%s, namespace)
+              > 60 * 60 * 3
+            ||| % $._config.per_cluster_label,
+            'for': '1h',
+            labels: {
+              severity: 'critical',
+            },
+            annotations: {
+              summary: 'Loki compaction has not run in the last 3 hours since the last compaction.',
+              description: std.strReplace(|||
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3 hours since the last compaction. This may indicate a problem with the compactor.
+              |||, 'cluster', $._config.per_cluster_label),
+            },
+          },
+          {
+            // Alert if the compactor has not successfully run compaction in the last 3h since startup.
+            alert: 'LokiCompactorHasNotSuccessfullyRunCompaction',
+            expr: |||
+              # The "last successful run" metric is updated even if the compactor owns no tenants,
+              # so this alert correctly doesn't fire if compactor has nothing to do.
+              max(
+                max_over_time(
+                  loki_boltdb_shipper_compact_tables_operation_last_successful_run_timestamp_seconds{}[3h]
+                )
+              ) by (%s, namespace)
+              == 0
+            ||| % $._config.per_cluster_label,
+            'for': '1h',
+            labels: {
+              severity: 'critical',
+            },
+            annotations: {
+              summary: 'Loki compaction has not run in the last 3h since startup.',
+              description: std.strReplace(|||
+                {{ $labels.cluster }} {{ $labels.namespace }} has not run compaction in the last 3h since startup. This may indicate a problem with the compactor.
+              |||, 'cluster', $._config.per_cluster_label),
+            },
+          },
         ],
       },
     ],