Add jmx-exporter monitoring functionality with servicemonitor

valeriano-manassero · Jan 30, 2023 · a3ae5db · a3ae5db
1 parent 549224c
commit a3ae5db
Show file tree

Hide file tree

Showing 8 changed files with 211 additions and 0 deletions.
diff --git a/valeriano-manassero/trino/templates/configmap-coordinator.yaml b/valeriano-manassero/trino/templates/configmap-coordinator.yaml
@@ -21,6 +21,9 @@ data:
     -XX:+HeapDumpOnOutOfMemoryError
     -XX:+ExitOnOutOfMemoryError
     -Djdk.attach.allowAttachSelf=true
+{{- if .Values.jmx-exporter.enabled }}
+    -javaagent:{{ .Values.jmx-exporter.path }}/{{ .Values.jmx-exporter.jarfile }}={{ .Values.jmx-exporter.port }}:{{ .Values.jmx-exporter.path }}/etc/trino.yaml
+{{- end }}
 {{ .Values.config.coordinator.jvmExtraConfig | indent 4 }}
 
   config.properties: |

diff --git a/valeriano-manassero/trino/templates/configmap-jmx-exporter.yaml b/valeriano-manassero/trino/templates/configmap-jmx-exporter.yaml
@@ -0,0 +1,63 @@
+{{- if .Values.jmx-exporter.enabled }}
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ template "trino.fullname" . }}-jmx-exporter
+  labels:
+    {{- include "trino.labels" . | nindent 4 }}
+data:
+  trino.yaml: |
+    startDelaySeconds: 5
+    rules:
+    - pattern : trino.execution<name=SqlTaskManager><>InputDataSize.OneMinute.Count
+      name: trino_coord_Input_bytes_sci
+
+    - pattern : trino.execution<name=SqlTaskManager><>InputPositions.OneMinute.Count
+      name: trino_coord_Input_rows
+      
+    - pattern : trino.execution<name=SqlTaskManager><>OutputDataSize.OneMinute.Count
+      name: trino_coord_Output_bytes_sci
+      
+    - pattern : trino.execution<name=SqlTaskManager><>OutputPositions.OneMinute.Count
+      name: trino_coord_Output_rows
+      
+    - pattern : trino.memory<type=ClusterMemoryPool, name=general><>TotalDistributedBytes
+      name: trino_TotalDistributedBytes
+      
+    - pattern : trino.memory<type=ClusterMemoryPool, name=general><>ReservedDistributedBytes
+      name: trino_ReservedDistributedBytes
+
+    - pattern : trino.execution<name=QueryManager><>FailedQueries.OneMinute.Count
+      name: trino_Failed_Queries
+      
+    - pattern : trino.execution<name=QueryManager><>RunningQueries
+      name: trino_running_queries
+
+    - pattern : trino.failuredetector<name=HeartbeatFailureDetector><>ActiveCount
+      name: trino_active_nodes
+
+    - pattern : trino.memory<type=ClusterMemoryPool, name=general><>FreeDistributedBytes
+      name: trino_cluster_free_memory
+      
+    - pattern : trino.execution<name=QueryExecution><>Executor.QueuedTaskCount
+      name: trino_queued_task_count
+
+    - pattern : trino.execution<name=QueryManager><>StartedQueries.FiveMinute.Count
+      name: trino_started_queries
+
+    - pattern : java.lang<type=Memory><HeapMemoryUsage>committed
+      name: trino_jvm_heap_memory_usage
+
+    - pattern : java.lang<type=Threading><>ThreadCount
+      name: trino_jvm_thread_count
+
+    - pattern : trino.execution<name=QueryManager><>InternalFailures.OneMinute.Count
+      name: trino_failed_queries_internal
+
+    - pattern : trino.execution<name=QueryManager><>ExternalFailures.OneMinute.Count
+      name: trino_failed_queries_external
+
+    - pattern : trino.execution<name=QueryManager><>UserErrorFailures.OneMinute.Count
+      name: trino_failed_queries_user
+
+{{- end }}
diff --git a/valeriano-manassero/trino/templates/deployment-coordinator.yaml b/valeriano-manassero/trino/templates/deployment-coordinator.yaml
@@ -87,6 +87,13 @@ spec:
           persistentVolumeClaim:
             claimName: {{ .Values.groupProvider.pvcName }}
         {{- end }}{{- end }}
+        {{- if .Values.jmx-exporter.enabled }}
+        - name: jmx-exporter-config
+          configMap:
+            name: {{ template "trino.fullname" . }}-jmx-exporter
+        {{- end }}
+        - name: jmx-exporter-lib
+          emptyDir: {}
         {{- range .Values.secretMounts }}
         - name: {{ .name }}
           secret:
@@ -99,6 +106,23 @@ spec:
         {{- end }}
       {{- if or .Values.config.coordinator.initContainers (include "trino.tlsEncryption" . ) }}
       initContainers:
+      {{- if .Values.jmx-exporter.enabled }}
+        - name: download-jmx-exporter
+          image: {{ .Values.jmx-exporter.image.repository }}:{{ .Values.jmx-exporter.image.tag }}
+          imagePullPolicy: {{ .Values.jmx-exporter.image.pullPolicy }}
+          {{- with .Values.image.securityContext }}
+          securityContext:
+            runAsUser: {{ .runAsUser }}
+            runAsGroup: {{ .runAsGroup }}
+          {{- end }}
+          args:
+            - "-o"
+            - "/jmx-exporter/lib/" 
+            - "https://repo1.maven.org/maven2/io/prometheus/jmx/jmx_prometheus_javaagent/0.17.2/jmx_prometheus_javaagent-0.17.2.jar" 
+          volumeMounts:
+          - name: jmx-exporter-lib
+            mountPath: /jmx-exporter/lib/
+      {{- end }}
       {{- if (include "trino.tlsEncryption" . ) }}
         - name: init-certs
           image: {{ .Values.initKeystore.image.repository }}:{{ .Values.initKeystore.image.tag }}
@@ -211,6 +235,12 @@ spec:
             - mountPath: {{ .Values.config.general.path }}/group-provider
               name: {{ .Values.groupProvider.pvcVolumeName | default "group-provider-pvc-volume" }}
             {{- end }}{{- end }}
+            {{- if .Values.jmx-exporter.enabled }}
+            - mountPath: {{ .Values.jmx-exporter.path }}/etc/
+              name: jmx-exporter-config
+            - name: jmx-exporter-lib
+              mountPath: /jmx-exporter/lib/
+            {{- end }}
             {{- range .Values.secretMounts }}
             - name: {{ .name }}
               mountPath: {{ .path }}
@@ -234,6 +264,11 @@ spec:
               containerPort: {{ .Values.config.general.httpsServer.port }}
               protocol: TCP
             {{- end }}
+            {{- if .Values.jmx-exporter.enabled }}
+            - name: jmx-exporter
+              containerPort: {{ .Values.jmx-exporter.port | default 9000 }}
+              protocol: TCP
+            {{- end }}
           livenessProbe:
             httpGet:
               path: /v1/status

diff --git a/valeriano-manassero/trino/templates/service-worker.yaml b/valeriano-manassero/trino/templates/service-worker.yaml
@@ -0,0 +1,18 @@
+{{- if .Values.jmx-exporter.enabled }}
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ template "trino.worker" . }}
+  labels:
+    {{- include "trino.labels" . | nindent 4 }}
+spec:
+  type: {{ .Values.service.type }}
+  ports:
+    - port: {{ .Values.jmx-exporter.port | default 9000 }}
+      targetPort: jmx-exporter
+      protocol: TCP
+      name: jmx-exporter
+  selector:
+    {{- include "trino.selectorLabels" . | nindent 4 }}
+    app.kubernetes.io/component: worker
+{{- end }}
diff --git a/valeriano-manassero/trino/templates/service.yaml b/valeriano-manassero/trino/templates/service.yaml
@@ -17,6 +17,12 @@ spec:
       protocol: TCP
       name: https-coord
     {{- end }}
+    {{- if .Values.jmx-exporter.enabled }}
+    - port: {{ .Values.jmx-exporter.port | default 9000 }}
+      targetPort: jmx-exporter
+      protocol: TCP
+      name: jmx-exporter
+    {{- end }}
   selector:
     {{- include "trino.selectorLabels" . | nindent 4 }}
     app.kubernetes.io/component: coordinator
diff --git a/valeriano-manassero/trino/templates/servicemonitor-coordinator.yaml b/valeriano-manassero/trino/templates/servicemonitor-coordinator.yaml
@@ -0,0 +1,32 @@
+{{- if.Values.jmx-exporter.serviceMonitor.enabled -}}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "trino.coordinator" . }}
+  labels:
+    {{- include "trino.labels" . | nindent 4 }}
+spec:
+  jobLabel: {{ .Release.Name }}-coordinator
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace }}
+  selector:
+    matchLabels:
+      {{- include "this.selectorLabels" . | nindent 6 }}
+      app.kubernetes.io/component: coordinator
+  endpoints:
+    - port: {{ .Values.jmx-exporter.serviceMonitor.port }}
+      {{- with .Values.jmx-exporter.serviceMonitor.path }}
+      path: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.interval }}
+      interval: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.scrapeTimeout  }}
+      scrapeTimeout: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.relabelings  }}
+      relabelings:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+{{- end }}
diff --git a/valeriano-manassero/trino/templates/servicemonitor-worker.yaml b/valeriano-manassero/trino/templates/servicemonitor-worker.yaml
@@ -0,0 +1,32 @@
+{{- if.Values.jmx-exporter.serviceMonitor.enabled -}}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "trino.worker" . }}
+  labels:
+    {{- include "trino.labels" . | nindent 4 }}
+spec:
+  jobLabel: {{ .Release.Name }}-worker
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace }}
+  selector:
+    matchLabels:
+      {{- include "this.selectorLabels" . | nindent 6 }}
+      app.kubernetes.io/component: worker
+  endpoints:
+    - port: {{ .Values.jmx-exporter.serviceMonitor.port }}
+      {{- with .Values.jmx-exporter.serviceMonitor.path }}
+      path: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.interval }}
+      interval: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.scrapeTimeout  }}
+      scrapeTimeout: {{ . }}
+      {{- end }}
+      {{- with .Values.jmx-exporter.serviceMonitor.relabelings  }}
+      relabelings:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+{{- end }}
diff --git a/valeriano-manassero/trino/values.yaml b/valeriano-manassero/trino/values.yaml
@@ -449,3 +449,25 @@ tls:
     # requests:
     #  cpu: 100m
     #  memory: 128Mi
+
+jmx-exporter:
+  enable: true
+  port: 9000
+  path: /prometheus
+  jarfile: jmx_prometheus_javaagent-0.17.2.jar
+  image:
+    repository: curlimages/curl
+    tag: 7.87.0
+    pullPolicy: IfNotPresent
+  serviceMonitor:
+    enabled: false
+    port: "jmx-exporter"
+    additionalLabels: {}
+    interval: 1m
+    scrapeTimeout: 10s
+    path: /metrics
+    # Additional relabeling configs to the ones that Prometheus Operator
+    # creates by default.
+    # relabelings:
+    #   - sourceLabels: [__meta_kubernetes_endpoints_label_app_kubernetes_io_version]
+    #     targetLabel: version