chore(opentelemetry): revise prometheus monitoring (#460)

* chore(opentelemetry): revise prometheus monitoring - grouped the prometheus-related values into one directive - PodMonitor and ServiceMonitor are enabled by default - Added indication whether `monitoring.coreos.com/v1` is available on the cluster - plugindefinition updated * docs(opentelemetry): reword option value description * fix(opentelemetry): invoke built-in object * chore(opentelemetry) update README * chore(opentelemetry) adjust threshold --------- Co-authored-by: I504010 <[email protected]>
cloudoperators · Oct 24, 2024 · 73a4553 · 73a4553
1 parent 07c8f52
commit 73a4553
Show file tree

Hide file tree

Showing 9 changed files with 37 additions and 32 deletions.
diff --git a/opentelemetry/README.md b/opentelemetry/README.md
@@ -35,15 +35,15 @@ Contributors are welcome to provide additional collector configurations.
 `openTelemetry.openSearchLogs.endpoint` | Endpoint URL for OpenSearch      | secret | `false` | 
 `openTelemetry.region`                   | Region label for logging         | string | `false` |
 `openTelemetry.cluster`                  | Cluster label for logging        | string | `false` |
-`openTelemetry.prometheus.additionalLabels`               | Label for Prometheus Service Monitoring | map | `false` | 
-`openTelemetry.podMonitor.enabled`       | Activates the pod-monitoring for the LogsCollector | bool | `false` | 
-`opentelemetry-operator.admissionWbhooks.certManager.enabled` | Activate to use the CertManager for generating self-signed certificates | bool | `false` | 
+`openTelemetry.prometheus.additionalLabels`               | Label selector for Prometheus resources to be picked-up by the operator | map | `false` | 
+`openTelemetry.prometheus.serviceMonitor.enabled` | Activates the service-monitoring for the Logs Collector  | bool | `false` | 
+`openTelemetry.prometheus.podMonitor.enabled`       | Activates the pod-monitoring for the Logs Collector | bool | `false` | 
+`openTelemetry-operator.admissionWebhooks.certManager.enabled` | Activate to use the CertManager for generating self-signed certificates | bool | `false` | 
 `opentelemetry-operator.admissionWebhooks.autoGenerateCert.enabled` | Activate to use Helm to create self-signed certificates | bool | `false` | 
 `opentelemetry-operator.admissionWebhooks.autoGenerateCert.recreate` | Activate to recreate the cert after a defined period (certPeriodDays default is 365) | bool | `false` | 
 `opentelemetry-operator.kubeRBACProxy.enabled` | Activate to enable Kube-RBAC-Proxy for OpenTelemetry | bool | `false` | 
 `opentelemetry-operator.manager.prometheusRule.defaultRules.enabled` | Activate to enable default rules for monitoring the OpenTelemetry Manager | bool | `false` | 
 `opentelemetry-operator.manager.prometheusRule.enabled` | Activate to enable rules for monitoring the OpenTelemetry Manager | bool | `false` | 
-`opentelemetry-operator.manager.serviceMonitor.enabled` | Activate to enable the Service Monitor to ingest the metrics of the OpenTelemetry Collector in Prometheus  | bool | `false` | 
 
 ### Examples
 

diff --git a/opentelemetry/chart/Chart.yaml b/opentelemetry/chart/Chart.yaml
@@ -4,7 +4,7 @@
 apiVersion: v2
 appVersion: v0.110.0
 name: opentelemetry-operator
-version: 0.4.1
+version: 0.4.2
 description: OpenTelemetry Operator Helm chart for Kubernetes
 icon: https://raw.githubusercontent.com/cncf/artwork/a718fa97fffec1b9fd14147682e9e3ac0c8817cb/projects/opentelemetry/icon/color/opentelemetry-icon-color.png
 type: application

diff --git a/opentelemetry/chart/ci/test-values.yaml b/opentelemetry/chart/ci/test-values.yaml
@@ -36,16 +36,10 @@ openTelemetry:
     password: test
   cluster: test
   region: test
-  logsCollector:
-    enabled: true
-  metricsCollector:
-    enabled: false
-  podMonitor:
-    enabled: false
   prometheus:
     additionalLabels:
-      testkey1: test1
-      testkey2: test2
+      key1: value1
+      key2: value2
 testFramework:
   enabled: false
   image:

diff --git a/opentelemetry/chart/dashboards/otel-logs.json b/opentelemetry/chart/dashboards/otel-logs.json
@@ -16,7 +16,7 @@
   "gnetId": null,
   "graphTooltip": 0,
   "id": 291,
-  "iteration": 1729071061818,
+  "iteration": 1729766555603,
   "links": [],
   "panels": [
     {
@@ -62,7 +62,7 @@
               },
               {
                 "color": "green",
-                "value": 11
+                "value": 1
               }
             ]
           }
@@ -2760,7 +2760,7 @@
       {
         "allValue": null,
         "current": {
-          "selected": false,
+          "selected": true,
           "text": "opensearch/logs",
           "value": "opensearch/logs"
         },
@@ -2827,7 +2827,7 @@
       {
         "allValue": null,
         "current": {
-          "selected": false,
+          "selected": true,
           "text": "_total",
           "value": "_total"
         },

diff --git a/opentelemetry/chart/templates/logs-collector.yaml b/opentelemetry/chart/templates/logs-collector.yaml
@@ -38,7 +38,7 @@ spec:
   envFrom:
     - secretRef:
          name: otel-basic-auth
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if .Values.openTelemetry.prometheus.podMonitor.enabled }}
   ports:
     - name: prometheus
       port: 9999
@@ -71,7 +71,7 @@ spec:
             field: resource["container.runtime"]
             value: "containerd"
 
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if .Values.openTelemetry.prometheus.podMonitor.enabled }}
       prometheus/internal:
         config:
           scrape_configs:
@@ -87,7 +87,7 @@ spec:
         timeout: 5s
         send_batch_size : 10
 
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if .Values.openTelemetry.prometheus.podMonitor.enabled }}
       attributes/prometheus:
         actions:
           - action: insert
@@ -353,7 +353,7 @@ spec:
     service:
       extensions:
         - basicauth
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if .Values.openTelemetry.prometheus.podMonitor.enabled }}
       telemetry:
         metrics:
           address: 127.0.0.1:8888
@@ -372,7 +372,7 @@ spec:
           receivers: [journald]
           processors: [attributes/cluster,transform/journal,batch]
           exporters: [opensearch/logs]
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if .Values.openTelemetry.prometheus.podMonitor.enabled }}
         metrics/prometheus:
           receivers: [prometheus/internal]
           processors: [attributes/prometheus]

diff --git a/opentelemetry/chart/templates/pmon-filelog.yaml b/opentelemetry/chart/templates/pmon-filelog.yaml
@@ -2,7 +2,7 @@
 SPDX-FileCopyrightText: 2024 SAP SE or an SAP affiliate company and Greenhouse contributors
 SPDX-License-Identifier: Apache-2.0
 */}}
-{{- if .Values.openTelemetry.podMonitor.enabled }}
+{{- if and (.Capabilities.APIVersions.Has "monitoring.coreos.com/v1") .Values.openTelemetry.prometheus.podMonitor.enabled }}
 apiVersion: monitoring.coreos.com/v1
 kind: PodMonitor
 metadata:

diff --git a/opentelemetry/chart/templates/smon.yaml b/opentelemetry/chart/templates/smon.yaml
@@ -2,7 +2,7 @@
 SPDX-FileCopyrightText: 2024 SAP SE or an SAP affiliate company and Greenhouse contributors
 SPDX-License-Identifier: Apache-2.0
 */}}
-{{- if index .Values "opentelemetry-operator" "manager" "serviceMonitor" "enabled" }}
+{{- if and (.Capabilities.APIVersions.Has "monitoring.coreos.com/v1") .Values.openTelemetry.prometheus.serviceMonitor.enabled }}
 apiVersion: monitoring.coreos.com/v1
 kind: ServiceMonitor
 metadata:

diff --git a/opentelemetry/chart/values.yaml b/opentelemetry/chart/values.yaml
@@ -54,9 +54,15 @@ openTelemetry:
   region:
   logsCollector:
     enabled: true
+  metricsCollector:
+    enabled: false
   prometheus:
+    serviceMonitor:
+      enabled: true
+    podMonitor:
+      enabled: true
     ## Label selectors for the Prometheus resources to be picked up by prometheus-operator.
-    additionalLabels:
+    additionalLabels: {}
     # plugin: kube-monitoring
     # prometheus: infra
 testFramework:

diff --git a/opentelemetry/plugindefinition.yaml b/opentelemetry/plugindefinition.yaml
@@ -6,14 +6,14 @@ kind: PluginDefinition
 metadata:
     name: opentelemetry
 spec:
-    version: 0.4.1
+    version: 0.4.2
     displayName: OpenTelemetry
     description: Observability framework for instrumenting, generating, collecting, and exporting telemetry data such as traces, metrics, and logs.
     icon: https://raw.githubusercontent.com/cloudoperators/greenhouse-extensions/main/opentelemetry/logo.png
     helmChart:
         name: opentelemetry-operator
         repository: oci://ghcr.io/cloudoperators/greenhouse-extensions/charts
-        version: 0.4.1
+        version: 0.4.2
     options:
         - default: true
           description: Activates the standard configuration for logs
@@ -45,15 +45,20 @@ spec:
           name: openTelemetry.cluster
           required: false
           type: string
-        - description: Label for Prometheus Service Monitoring
-          name: openTelemetry.prometheus.additionalLabels
-          required: false
-          type: map
         - default: false
           description: Activates the pod-monitoring for the Logs Collector
-          name: openTelemetry.podMonitor.enabled
+          name: openTelemetry.prometheus.podMonitor.enabled
           required: false
           type: bool
+        - default: false
+          description: Activates the service-monitoring for the Logs Collector
+          name: openTelemetry.prometheus.serviceMonitor.enabled
+          required: false
+          type: bool
+        - description: Label selector for Prometheus resources to be picked-up by the operator
+          name: openTelemetry.prometheus.additionalLabels
+          required: false
+          type: map
         - default: false
           description: Activate to use the CertManager for generating self-signed certificates
           name: opentelemetry-operator.admissionWebhooks.certManager.enabled