adding service monitors and fixing mistake on discoverable alertmanager

Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
cortexproject · Nov 9, 2020 · 36f0d5e · 36f0d5e
1 parent 304a311
commit 36f0d5e
Show file tree

Hide file tree

Showing 14 changed files with 490 additions and 1 deletion.
diff --git a/README.md b/README.md
@@ -109,6 +109,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | alertmanager.securityContext | object | `{}` |  |
 | alertmanager.service.annotations | object | `{}` |  |
 | alertmanager.service.labels | object | `{}` |  |
+| alertmanager.serviceMonitor.additionalLabels | object | `{}` |  |
+| alertmanager.serviceMonitor.annotations | object | `{}` |  |
+| alertmanager.serviceMonitor.enabled | bool | `false` |  |
+| alertmanager.serviceMonitor.interval | string | `""` |  |
 | alertmanager.statefulSet.enabled | bool | `false` |  |
 | alertmanager.statefulStrategy.type | string | `"RollingUpdate"` |  |
 | alertmanager.strategy.rollingUpdate.maxSurge | int | `0` |  |
@@ -160,6 +164,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | compactor.securityContext | object | `{}` |  |
 | compactor.service.annotations | object | `{}` |  |
 | compactor.service.labels | object | `{}` |  |
+| compactor.serviceMonitor.additionalLabels | object | `{}` |  |
+| compactor.serviceMonitor.annotations | object | `{}` |  |
+| compactor.serviceMonitor.enabled | bool | `false` |  |
+| compactor.serviceMonitor.interval | string | `""` |  |
 | compactor.strategy.type | string | `"RollingUpdate"` |  |
 | compactor.terminationGracePeriodSeconds | int | `240` |  |
 | compactor.tolerations | list | `[]` |  |
@@ -249,6 +257,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | configs.securityContext | object | `{}` |  |
 | configs.service.annotations | object | `{}` |  |
 | configs.service.labels | object | `{}` |  |
+| configs.serviceMonitor.additionalLabels | object | `{}` |  |
+| configs.serviceMonitor.annotations | object | `{}` |  |
+| configs.serviceMonitor.enabled | bool | `false` |  |
+| configs.serviceMonitor.interval | string | `""` |  |
 | configs.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | configs.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | configs.strategy.type | string | `"RollingUpdate"` |  |
@@ -292,6 +304,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | distributor.securityContext | object | `{}` |  |
 | distributor.service.annotations | object | `{}` |  |
 | distributor.service.labels | object | `{}` |  |
+| distributor.serviceMonitor.additionalLabels | object | `{}` |  |
+| distributor.serviceMonitor.annotations | object | `{}` |  |
+| distributor.serviceMonitor.enabled | bool | `false` |  |
+| distributor.serviceMonitor.interval | string | `""` |  |
 | distributor.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | distributor.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | distributor.strategy.type | string | `"RollingUpdate"` |  |
@@ -345,6 +361,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | ingester.securityContext | object | `{}` |  |
 | ingester.service.annotations | object | `{}` |  |
 | ingester.service.labels | object | `{}` |  |
+| ingester.serviceMonitor.additionalLabels | object | `{}` |  |
+| ingester.serviceMonitor.annotations | object | `{}` |  |
+| ingester.serviceMonitor.enabled | bool | `false` |  |
+| ingester.serviceMonitor.interval | string | `""` |  |
 | ingester.statefulSet.enabled | bool | `false` |  |
 | ingester.statefulStrategy.type | string | `"RollingUpdate"` |  |
 | ingester.strategy.rollingUpdate.maxSurge | int | `0` |  |
@@ -461,6 +481,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | querier.securityContext | object | `{}` |  |
 | querier.service.annotations | object | `{}` |  |
 | querier.service.labels | object | `{}` |  |
+| querier.serviceMonitor.additionalLabels | object | `{}` |  |
+| querier.serviceMonitor.annotations | object | `{}` |  |
+| querier.serviceMonitor.enabled | bool | `false` |  |
+| querier.serviceMonitor.interval | string | `""` |  |
 | querier.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | querier.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | querier.strategy.type | string | `"RollingUpdate"` |  |
@@ -499,6 +523,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | query_frontend.securityContext | object | `{}` |  |
 | query_frontend.service.annotations | object | `{}` |  |
 | query_frontend.service.labels | object | `{}` |  |
+| query_frontend.serviceMonitor.additionalLabels | object | `{}` |  |
+| query_frontend.serviceMonitor.annotations | object | `{}` |  |
+| query_frontend.serviceMonitor.enabled | bool | `false` |  |
+| query_frontend.serviceMonitor.interval | string | `""` |  |
 | query_frontend.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | query_frontend.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | query_frontend.strategy.type | string | `"RollingUpdate"` |  |
@@ -535,6 +563,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | ruler.securityContext | object | `{}` |  |
 | ruler.service.annotations | object | `{}` |  |
 | ruler.service.labels | object | `{}` |  |
+| ruler.serviceMonitor.additionalLabels | object | `{}` |  |
+| ruler.serviceMonitor.annotations | object | `{}` |  |
+| ruler.serviceMonitor.enabled | bool | `false` |  |
+| ruler.serviceMonitor.interval | string | `""` |  |
 | ruler.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | ruler.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | ruler.strategy.type | string | `"RollingUpdate"` |  |
@@ -543,6 +575,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | serviceAccount.annotations | object | `{}` |  |
 | serviceAccount.create | bool | `true` |  |
 | serviceAccount.name | string | `nil` |  |
+| serviceMonitor.additionalLabels | object | `{}` |  |
+| serviceMonitor.annotations | object | `{}` |  |
+| serviceMonitor.enabled | bool | `false` |  |
+| serviceMonitor.interval | string | `""` |  |
 | store_gateway.affinity.podAntiAffinity.preferredDuringSchedulingIgnoredDuringExecution[0].podAffinityTerm.labelSelector.matchExpressions[0].key | string | `"target"` |  |
 | store_gateway.affinity.podAntiAffinity.preferredDuringSchedulingIgnoredDuringExecution[0].podAffinityTerm.labelSelector.matchExpressions[0].operator | string | `"In"` |  |
 | store_gateway.affinity.podAntiAffinity.preferredDuringSchedulingIgnoredDuringExecution[0].podAffinityTerm.labelSelector.matchExpressions[0].values[0] | string | `"store-gateway"` |  |
@@ -585,6 +621,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | store_gateway.securityContext | object | `{}` |  |
 | store_gateway.service.annotations | object | `{}` |  |
 | store_gateway.service.labels | object | `{}` |  |
+| store_gateway.serviceMonitor.additionalLabels | object | `{}` |  |
+| store_gateway.serviceMonitor.annotations | object | `{}` |  |
+| store_gateway.serviceMonitor.enabled | bool | `false` |  |
+| store_gateway.serviceMonitor.interval | string | `""` |  |
 | store_gateway.strategy.type | string | `"RollingUpdate"` |  |
 | store_gateway.terminationGracePeriodSeconds | int | `240` |  |
 | store_gateway.tolerations | list | `[]` |  |
@@ -617,6 +657,10 @@ Source code can be found [here](https://cortexmetrics.io/)
 | table_manager.securityContext | object | `{}` |  |
 | table_manager.service.annotations | object | `{}` |  |
 | table_manager.service.labels | object | `{}` |  |
+| table_manager.serviceMonitor.additionalLabels | object | `{}` |  |
+| table_manager.serviceMonitor.annotations | object | `{}` |  |
+| table_manager.serviceMonitor.enabled | bool | `false` |  |
+| table_manager.serviceMonitor.interval | string | `""` |  |
 | table_manager.strategy.rollingUpdate.maxSurge | int | `0` |  |
 | table_manager.strategy.rollingUpdate.maxUnavailable | int | `1` |  |
 | table_manager.strategy.type | string | `"RollingUpdate"` |  |

diff --git a/templates/alertmanager-servicemonitor.yaml b/templates/alertmanager-servicemonitor.yaml
@@ -0,0 +1,34 @@
+{{- if .Values.alertmanager.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "cortex.fullname" . }}-alertmanager
+  labels:
+    app: {{ template "cortex.name" . }}
+    chart: {{ template "cortex.chart" . }}
+    heritage: {{ .Release.Service }}
+    release: {{ .Release.Name }}
+    {{- if .Values.alertmanager.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.alertmanager.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+  {{- if .Values.alertmanager.serviceMonitor.annotations }}
+  annotations:
+{{ toYaml .Values.alertmanager.serviceMonitor.annotations | indent 4 }}
+  {{- end }}
+spec:
+  selector:
+    matchLabels:
+      app: {{ template "cortex.name" . }}-alertmanager
+      release: {{ .Release.Name | quote }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace | quote }}
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.alertmanager.serviceMonitor.interval }}
+    interval: {{ .Values.alertmanager.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.alertmanager.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.alertmanager.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+{{- end }}
diff --git a/templates/alertmanager-statefulset.yaml b/templates/alertmanager-statefulset.yaml
@@ -19,7 +19,7 @@ spec:
       app: {{ template "cortex.name" . }}-alertmanager
       release: {{ .Release.Name }}
   updateStrategy:
-    {{- toYaml .Values.alertmanager.statefulStrategy | nindent 4 }}          
+    {{- toYaml .Values.alertmanager.statefulStrategy | nindent 4 }}
   serviceName: {{ template "cortex.fullname" . }}-alertmanager
   {{- if .Values.alertmanager.persistentVolume.enabled }}
   volumeClaimTemplates:
@@ -85,6 +85,8 @@ spec:
         {{- toYaml .Values.alertmanager.tolerations | nindent 8 }}
       terminationGracePeriodSeconds: {{ .Values.alertmanager.terminationGracePeriodSeconds }}
       volumes:
+        - emptyDir: {}
+          name: tmp
         - name: config
           secret:
             {{- if .Values.useExternalConfig }}
@@ -112,7 +114,11 @@ spec:
             - "-alertmanager.configs.url=http://{{ template "cortex.fullname" . }}-configs.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}:{{ .Values.config.server.http_listen_port }}"
             {{- if gt (.Values.alertmanager.replicas | int) 1 }}
             {{- range $n := until (.Values.alertmanager.replicas |int ) }}
+<<<<<<< HEAD
             - -cluster.peer={{ template "cortex.fullname" $ }}-alertmanager-{{ $n }}.{{ template "cortex.fullname" $ }}-alertmanager-headless.{{ $.Release.Namespace }}.svc.{{ .Values.clusterDomain }}:{{ $clusterPort }}
+=======
+            - -cluster.peer={{ template "cortex.fullname" $ }}-alertmanager-{{ $n }}.{{ template "cortex.fullname" $ }}-alertmanager.{{ $.Release.Namespace }}.svc.cluster.local:{{ $clusterPort }}
+>>>>>>> e3b2884 (I messed up)
             {{- end }}
             {{- end }}
             {{- range $key, $value := .Values.alertmanager.extraArgs }}
@@ -124,6 +130,8 @@ spec:
             {{- end }}
             - name: config
               mountPath: /etc/cortex
+            - mountPath: /tmp
+              name: tmp
             - name: storage
               mountPath: "/data"
               {{- if .Values.alertmanager.persistentVolume.subPath }}

diff --git a/templates/compactor-servicemonitor.yaml b/templates/compactor-servicemonitor.yaml
@@ -0,0 +1,34 @@
+{{- if .Values.compactor.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "cortex.fullname" . }}-compactor
+  labels:
+    app: {{ template "cortex.name" . }}
+    chart: {{ template "cortex.chart" . }}
+    heritage: {{ .Release.Service }}
+    release: {{ .Release.Name }}
+    {{- if .Values.compactor.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.compactor.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+  {{- if .Values.compactor.serviceMonitor.annotations }}
+  annotations:
+{{ toYaml .Values.compactor.serviceMonitor.annotations | indent 4 }}
+  {{- end }}
+spec:
+  selector:
+    matchLabels:
+      app: {{ template "cortex.name" . }}-compactor
+      release: {{ .Release.Name | quote }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace | quote }}
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.compactor.serviceMonitor.interval }}
+    interval: {{ .Values.compactor.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.compactor.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.compactor.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+{{- end }}
diff --git a/templates/compactor-svc.yaml b/templates/compactor-svc.yaml
@@ -0,0 +1,25 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ template "cortex.fullname" . }}-compactor
+  namespace: {{ .Release.Namespace }}
+  labels:
+    app: {{ template "cortex.name" . }}-compactor
+    chart: {{ template "cortex.chart" . }}
+    release: {{ .Release.Name }}
+    heritage: {{ .Release.Service }}
+    {{- with .Values.compactor.service.labels }}
+    {{- toYaml . | nindent 4 }}
+    {{- end }}
+  annotations:
+    {{- toYaml .Values.compactor.service.annotations | nindent 4 }}
+spec:
+  type: ClusterIP
+  ports:
+    - port: {{ .Values.config.server.http_listen_port }}
+      protocol: TCP
+      name: http-metrics
+      targetPort: http-metrics
+  selector:
+    app: {{ template "cortex.name" . }}-compactor
+    release: {{ .Release.Name }}
diff --git a/templates/configs-servicemonitor.yaml b/templates/configs-servicemonitor.yaml
@@ -0,0 +1,34 @@
+{{- if .Values.configs.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "cortex.fullname" . }}-configs
+  labels:
+    app: {{ template "cortex.name" . }}
+    chart: {{ template "cortex.chart" . }}
+    heritage: {{ .Release.Service }}
+    release: {{ .Release.Name }}
+    {{- if .Values.configs.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.configs.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+  {{- if .Values.configs.serviceMonitor.annotations }}
+  annotations:
+{{ toYaml .Values.configs.serviceMonitor.annotations | indent 4 }}
+  {{- end }}
+spec:
+  selector:
+    matchLabels:
+      app: {{ template "cortex.name" . }}-configs
+      release: {{ .Release.Name | quote }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace | quote }}
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.configs.serviceMonitor.interval }}
+    interval: {{ .Values.configs.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.configs.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.configs.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+{{- end }}
diff --git a/templates/distributor-servicemonitor.yaml b/templates/distributor-servicemonitor.yaml
@@ -0,0 +1,34 @@
+{{- if .Values.distributor.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "cortex.fullname" . }}-distributor
+  labels:
+    app: {{ template "cortex.name" . }}
+    chart: {{ template "cortex.chart" . }}
+    heritage: {{ .Release.Service }}
+    release: {{ .Release.Name }}
+    {{- if .Values.distributor.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.distributor.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+  {{- if .Values.distributor.serviceMonitor.annotations }}
+  annotations:
+{{ toYaml .Values.distributor.serviceMonitor.annotations | indent 4 }}
+  {{- end }}
+spec:
+  selector:
+    matchLabels:
+      app: {{ template "cortex.name" . }}-distributor
+      release: {{ .Release.Name | quote }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace | quote }}
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.distributor.serviceMonitor.interval }}
+    interval: {{ .Values.distributor.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.distributor.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.distributor.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+{{- end }}
diff --git a/templates/ingester-servicemonitor.yaml b/templates/ingester-servicemonitor.yaml
@@ -0,0 +1,34 @@
+{{- if .Values.ingester.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "cortex.fullname" . }}-ingester
+  labels:
+    app: {{ template "cortex.name" . }}
+    chart: {{ template "cortex.chart" . }}
+    heritage: {{ .Release.Service }}
+    release: {{ .Release.Name }}
+    {{- if .Values.ingester.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.ingester.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+  {{- if .Values.ingester.serviceMonitor.annotations }}
+  annotations:
+{{ toYaml .Values.ingester.serviceMonitor.annotations | indent 4 }}
+  {{- end }}
+spec:
+  selector:
+    matchLabels:
+      app: {{ template "cortex.name" . }}-ingester
+      release: {{ .Release.Name | quote }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace | quote }}
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.ingester.serviceMonitor.interval }}
+    interval: {{ .Values.ingester.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.ingester.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.ingester.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+{{- end }}