feat(monitoring): add mpabi node-exporter scrape and agave dashboard

bootstrap/argocd/application-monitoring-stack.yaml

@@ -26,6 +26,12 @@ spec:
                 static_configs:
                   - targets:
                       - 10.66.66.1:8999
+              - job_name: mpabi-node-exporter
+                metrics_path: /metrics
+                scrape_interval: 15s
+                static_configs:
+                  - targets:
+                      - 10.66.66.1:9100
         prometheusOperator:
           admissionWebhooks:
             enabled: false

kustomize/infra/monitoring-extras/dashboard-agave-status.yaml

@@ -0,0 +1,122 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: grafana-dashboard-agave-status
+  namespace: monitoring
+  labels:
+    grafana_dashboard: "1"
+data:
+  agave-status.json: |-
+    {
+      "uid": "agave-status-mpabi",
+      "title": "Agave @ mpabi",
+      "timezone": "browser",
+      "schemaVersion": 39,
+      "version": 1,
+      "refresh": "10s",
+      "tags": ["agave", "solana", "mpabi"],
+      "templating": {
+        "list": [
+          {
+            "name": "instance",
+            "type": "query",
+            "label": "Node Exporter Instance",
+            "datasource": { "type": "prometheus", "uid": "prometheus" },
+            "query": {
+              "query": "label_values(up{job=\"mpabi-node-exporter\"}, instance)",
+              "refId": "PromVarInstance"
+            },
+            "current": { "selected": false, "text": "10.66.66.1:9100", "value": "10.66.66.1:9100" }
+          }
+        ]
+      },
+      "panels": [
+        {
+          "id": 1,
+          "type": "stat",
+          "title": "Geyser Metrics Target (Prometheus up)",
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "targets": [
+            {
+              "refId": "A",
+              "expr": "up{job=\"mpabi-yellowstone-geyser\"}"
+            }
+          ],
+          "options": {
+            "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false },
+            "orientation": "horizontal",
+            "textMode": "value"
+          },
+          "fieldConfig": {
+            "defaults": {
+              "thresholds": {
+                "mode": "absolute",
+                "steps": [
+                  { "color": "red", "value": null },
+                  { "color": "green", "value": 1 }
+                ]
+              }
+            },
+            "overrides": []
+          },
+          "gridPos": { "h": 6, "w": 12, "x": 0, "y": 0 }
+        },
+        {
+          "id": 2,
+          "type": "stat",
+          "title": "agave-validator.service (systemd active)",
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "targets": [
+            {
+              "refId": "B",
+              "expr": "node_systemd_unit_state{job=\"mpabi-node-exporter\",instance=\"$instance\",name=\"agave-validator.service\",state=\"active\"}"
+            }
+          ],
+          "options": {
+            "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false },
+            "orientation": "horizontal",
+            "textMode": "value"
+          },
+          "fieldConfig": {
+            "defaults": {
+              "thresholds": {
+                "mode": "absolute",
+                "steps": [
+                  { "color": "red", "value": null },
+                  { "color": "green", "value": 1 }
+                ]
+              }
+            },
+            "overrides": []
+          },
+          "gridPos": { "h": 6, "w": 12, "x": 12, "y": 0 }
+        },
+        {
+          "id": 3,
+          "type": "timeseries",
+          "title": "Load (1m)",
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "targets": [
+            {
+              "refId": "C",
+              "expr": "node_load1{job=\"mpabi-node-exporter\",instance=\"$instance\"}"
+            }
+          ],
+          "gridPos": { "h": 8, "w": 12, "x": 0, "y": 6 }
+        },
+        {
+          "id": 4,
+          "type": "timeseries",
+          "title": "Memory Used (%)",
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "targets": [
+            {
+              "refId": "D",
+              "expr": "100 - (node_memory_MemAvailable_bytes{job=\"mpabi-node-exporter\",instance=\"$instance\"} / node_memory_MemTotal_bytes{job=\"mpabi-node-exporter\",instance=\"$instance\"} * 100)"
+            }
+          ],
+          "gridPos": { "h": 8, "w": 12, "x": 12, "y": 6 }
+        }
+      ]
+    }
+

kustomize/infra/monitoring-extras/kustomization.yaml

@@ -7,3 +7,4 @@ resources:
   - ingressroute-grafana-http.yaml
   - ingressroute-prometheus.yaml
   - ingressroute-prometheus-http.yaml
+  - dashboard-agave-status.yaml