Merge pull request #14 from GoogleCloudPlatform/master

bombastictranz/monitoring-dashboard-samples
bombastictranz · Mar 14, 2024 · 62b039a · 62b039a
2 parents 96c2a04 + 2d901a1
commit 62b039a
Show file tree

Hide file tree

Showing 45 changed files with 3,152 additions and 668 deletions.
diff --git a/alerts/google-gce/metadata.yaml b/alerts/google-gce/metadata.yaml
@@ -55,3 +55,17 @@ alert_policy_templates:
   related_integrations:
     - id: gce
       platform: GCP
+-
+  id: reservation-utilization-too-high
+  description: "Monitors reservation utilization across all GCE Reservations in the current project and will notify you if the utilization rises above 90%. Reservation utilization is (in use count / reserved count)."
+  version: 1
+  related_integrations:
+    - id: gce
+      platform: GCP
+-
+  id: reservation-utilization-too-low
+  description: "Monitors reservation utilization across all GCE Reservations in the current project and will notify you if the utilization falls below 10% for 20 of the past 23 hours. Reservation utilization is (in use count / reserved count)."
+  version: 1
+  related_integrations:
+    - id: gce
+      platform: GCP
diff --git a/alerts/google-gce/reservation-utilization-too-high.v1.json b/alerts/google-gce/reservation-utilization-too-high.v1.json
@@ -0,0 +1,21 @@
+{
+    "displayName": "Reservation - High Utilization",
+    "userLabels": {},
+    "conditions": [
+      {
+        "displayName": "High Reservation Utilization",
+        "conditionMonitoringQueryLanguage": {
+          "duration": "0s",
+          "query": "fetch compute.googleapis.com/Reservation\n|\n{ metric 'compute.googleapis.com/reservation/used'\n| align next_older(5m) | every 5m ;\nmetric 'compute.googleapis.com/reservation/reserved'\n| align next_older(5m) | every 5m\n}\n| ratio\n| condition val() >= 0.9",
+          "trigger": {
+            "count": 1
+          }
+        }
+      }
+    ],
+    "alertStrategy": {
+      "autoClose": "604800s"
+    },
+    "combiner": "OR",
+    "enabled": true
+  }
diff --git a/alerts/google-gce/reservation-utilization-too-low.v1.json b/alerts/google-gce/reservation-utilization-too-low.v1.json
@@ -0,0 +1,21 @@
+{
+    "displayName": "Reservation - Low Utilization",
+    "userLabels": {},
+    "conditions": [
+      {
+        "displayName": "Low Usage for 20 hours out of 23 hours",
+        "conditionMonitoringQueryLanguage": {
+          "duration": "0s",
+          "query": "fetch compute.googleapis.com/Reservation\n|\n{ metric 'compute.googleapis.com/reservation/used'\n| align next_older(5m) | every 5m ;\nmetric 'compute.googleapis.com/reservation/reserved'\n| align next_older(5m) | every 5m\n}\n| ratio\n| value val() <= 0.1\n| count_true_aligner(23h)\n| condition val() > 20 * 12 # 20 hours * (12 5 min intervals in hour)",
+          "trigger": {
+            "count": 1
+          }
+        }
+      }
+    ],
+    "alertStrategy": {
+      "autoClose": "604800s"
+    },
+    "combiner": "OR",
+    "enabled": true
+  }
diff --git a/dashboards/argo-workflows/README.md b/dashboards/argo-workflows/README.md
@@ -7,4 +7,4 @@
 |Argo Workflows Prometheus|
 |:------------------|
 |Filename: [argo-workflows-prometheus.json](argo-workflows-prometheus.json)|
-|This dashboard has charts displaying: `Running Workflows`, `Pending Workflows`, `Skipped Workflows`, `Succeeded Workflows`, `Failed Workflows`, `Errors`, `Operation Duration (seconds)`, `Queue Adds`, `Queue Depth`, and `Queue Latency`|
+|This dashboard has charts displaying: `Running Workflows`, `Pending Workflows`, `Skipped Workflows`, `Succeeded Workflows`, `Workflows With Pods Not Running`, `Failed Workflows`, `Errors`, `Operation Duration Seconds`, `Kubernetes Request Rates`, `Queue Adds`, `Queue Depth`, and `Queue Latency`|