Merge pull request #130 from appuio/dependencies/update-upstream-alerts

Update upstream alerts
appuio · May 2, 2024 · db8eb8f · db8eb8f
2 parents e8ff4d7 + 3d66f75
commit db8eb8f
Show file tree

Hide file tree

Showing 2 changed files with 26 additions and 0 deletions.
diff --git a/component/extracted_alerts/master/collector_prometheus_alerts.yaml b/component/extracted_alerts/master/collector_prometheus_alerts.yaml
@@ -85,6 +85,17 @@ spec:
       labels:
         service: visualization
         severity: Warning
+    - alert: DiskBufferUsage
+      annotations:
+        message: "Collectors potentially consuming too much node disk, {{ $value }}% "
+        summary: "Detected consuming too much node disk on $labels.hostname host"
+      expr: |
+        (label_replace(sum by(hostname) (vector_buffer_byte_size{component_kind='sink', buffer_type='disk'}), 'instance', '$1', 'hostname', '(.*)') 
+        / on(instance) group_left() sum by(instance) (node_filesystem_size_bytes{mountpoint='/var'})) * 100  > 0.15
+      for: 5m
+      labels:
+        service: collector
+        severity: Warning
   - name: logging_clusterlogging_telemetry.rules
     rules:
     - expr: |

diff --git a/tests/golden/master/openshift4-logging/openshift4-logging/60_collector_alerts.yaml b/tests/golden/master/openshift4-logging/openshift4-logging/60_collector_alerts.yaml
@@ -107,3 +107,18 @@ spec:
             severity: Warning
             syn: 'true'
             syn_component: openshift4-logging
+        - alert: SYN_DiskBufferUsage
+          annotations:
+            message: 'Collectors potentially consuming too much node disk, {{ $value
+              }}% '
+            summary: Detected consuming too much node disk on $labels.hostname host
+          expr: "(label_replace(sum by(hostname) (vector_buffer_byte_size{component_kind='sink',\
+            \ buffer_type='disk'}), 'instance', '$1', 'hostname', '(.*)') \n/ on(instance)\
+            \ group_left() sum by(instance) (node_filesystem_size_bytes{mountpoint='/var'}))\
+            \ * 100  > 0.15\n"
+          for: 5m
+          labels:
+            service: collector
+            severity: Warning
+            syn: 'true'
+            syn_component: openshift4-logging