fix(prom): update prom-agent config

willianpaixao · Nov 19, 2024 · 9ed54bb · 9ed54bb
1 parent 7abc587
commit 9ed54bb
Show file tree

Hide file tree

Showing 2 changed files with 40 additions and 5 deletions.
diff --git a/kubernetes/raspberry/observability/kube-prometheus-stack/app/helmrelease.yaml b/kubernetes/raspberry/observability/kube-prometheus-stack/app/helmrelease.yaml
@@ -85,8 +85,6 @@ spec:
           - new-service-discovery-manager
           - remote-write-receiver
         enableRemoteWriteReceiver: true
-        externalLabels:
-          cluster: "raspberry"
         retention: 30d
         retentionSize: 64GiB
         nodeSelector:
@@ -98,6 +96,26 @@ spec:
               resources:
                 requests:
                   storage: 64Gi
+    kubeControllerManager:
+      enabled: false
+    kubeScheduler:
+      enabled: false
+    kubeEtcd:
+      enabled: false
+    kubeProxy:
+      enabled: false
+    kubeApiServer:
+      serviceMonitor:
+        relabelings:
+          - sourceLabels: [__address__]
+            targetLabel: cluster
+            replacement: raspberry
+    kubelet:
+      serviceMonitor:
+        relabelings:
+          - sourceLabels: [__address__]
+            targetLabel: cluster
+            replacement: raspberry
     nodeExporter:
       enabled: true
     prometheus-node-exporter:
@@ -112,6 +130,12 @@ spec:
       prometheus:
         monitor:
           enabled: true
+          relabelings:
+            - sourceLabels: [__address__]
+              targetLabel: cluster
+              replacement: raspberry
+            - source_labels: ["__metrics_path__"]
+              target_label: metrics_path
     grafana:
       enabled: true
       defaultDashboardsTimezone: ${TIMEZONE}
@@ -127,6 +151,13 @@ spec:
         tls:
           - hosts:
               - *host
+      sidecar:
+        dashboards:
+          multicluster:
+            global:
+              enabled: true
+            etcd:
+              enabled: true
       persistence:
         enabled: true
         storageClassName: longhorn

diff --git a/kubernetes/turing/observability/kube-prometheus-stack/app/helmrelease.yaml b/kubernetes/turing/observability/kube-prometheus-stack/app/helmrelease.yaml
@@ -6,7 +6,7 @@ metadata:
   name: &app prometheus-agent
   namespace: &namespace observability
 spec:
-  interval: 30m
+  interval: 1h
   timeout: 15m
   chart:
     spec:
@@ -17,9 +17,11 @@ spec:
         name: prometheus-community-charts
         namespace: flux-system
   install:
+    crds: Create
     remediation:
       retries: 3
   upgrade:
+    crds: CreateReplace
     cleanupOnFail: true
     remediation:
       strategy: rollback
@@ -30,8 +32,6 @@ spec:
       enabled: false
     grafana:
       enabled: false
-    kubeApiServer:
-      enabled: false
     kubeEtcd:
       enabled: false
     kubeScheduler:
@@ -41,8 +41,12 @@ spec:
     prometheus:
       agentMode: true
       prometheusSpec:
+        externalLabels:
+          cluster: "turing"
         scrapeInterval: 30s
+        evaluationInterval: 30s
         podMonitorSelectorNilUsesHelmValues: false
+        scrapeConfigSelectorNilUsesHelmValues: false
         serviceMonitorSelectorNilUsesHelmValues: false
         remoteWrite:
           - url: 'http://192.168.0.35:9090/api/v1/write'