Add brusnika-stage YC Kafka exporter monitoring

2026-06-16 11:33:38 +03:00 · 2026-06-16 11:33:38 +03:00 · 69d07ba476
commit 69d07ba476
parent a342d22cd7
5 changed files with 230 additions and 0 deletions
--- a/clusters/brusnika-stage/infrastructure/kafka-exporter-yc-dashboard.yaml
+++ b/clusters/brusnika-stage/infrastructure/kafka-exporter-yc-dashboard.yaml
--- a/clusters/brusnika-stage/infrastructure/kafka-exporter-yc-rules.yaml
+++ b/clusters/brusnika-stage/infrastructure/kafka-exporter-yc-rules.yaml
@ -0,0 +1,47 @@
+apiVersion: operator.victoriametrics.com/v1beta1
+kind: VMRule
+metadata:
+  name: kafka-exporter-yc-rules
+  namespace: vmstack
+spec:
+  groups:
+    - name: kafka-exporter-yc.rules
+      interval: 5m
+      rules:
+        - alert: KafkaExporterYcDown
+          expr: absent(up{kafka_instance="yc-kafka", cluster="brusnika-stage"} == 1)
+          for: 10m
+          labels:
+            severity: critical
+            team: infra
+            cluster: brusnika-stage
+            kafka_instance: yc-kafka
+            source_cluster: yc-kafka
+          annotations:
+            summary: YC Kafka exporter is down in brusnika-stage
+            description: No healthy kafka-exporter-yc target is scraped for 10 minutes.
+        - alert: KafkaTopicNoMessagesFor12h
+          expr: |
+            (
+              sum by (topic, kafka_instance, source_cluster, cluster) (
+                max_over_time(kafka_topic_partition_current_offset{kafka_instance="yc-kafka", cluster="brusnika-stage", topic!~"^__.*"}[12h])
+                -
+                min_over_time(kafka_topic_partition_current_offset{kafka_instance="yc-kafka", cluster="brusnika-stage", topic!~"^__.*"}[12h])
+              ) == 0
+            )
+            and on (topic, kafka_instance, source_cluster, cluster)
+            (
+              sum by (topic, kafka_instance, source_cluster, cluster) (
+                kafka_topic_partitions{kafka_instance="yc-kafka", cluster="brusnika-stage", topic!~"^__.*"}
+              ) > 0
+            )
+          for: 5m
+          labels:
+            severity: warning
+            team: infra
+            cluster: brusnika-stage
+            kafka_instance: yc-kafka
+            source_cluster: yc-kafka
+          annotations:
+            summary: No new messages in Kafka topic for 12h
+            description: Topic {{ $labels.topic }} in {{ $labels.kafka_instance }} has no offset growth for 12 hours.
--- a/clusters/brusnika-stage/infrastructure/kafka-exporter-yc.yaml
+++ b/clusters/brusnika-stage/infrastructure/kafka-exporter-yc.yaml
@ -0,0 +1,159 @@
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: kafka-exporter-yc
+  namespace: kafka-exporter
+---
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: kafka-exporter-yc
+  namespace: kafka-exporter
+spec:
+  dependsOn:
+    - name: prometheus-stack
+      namespace: prometheus-stack
+    - name: vault
+      namespace: vault
+  interval: 5m
+  timeout: 10m
+  chart:
+    spec:
+      chart: kafka-exporter-prod
+      version: "0.27.0"
+      sourceRef:
+        kind: HelmRepository
+        name: yc-oci-charts
+        namespace: flux-system
+      interval: 10m
+  install:
+    remediation:
+      retries: 3
+  upgrade:
+    remediation:
+      retries: 3
+  postRenderers:
+    - kustomize:
+        patches:
+          - target:
+              group: apps
+              version: v1
+              kind: Deployment
+              name: kafka-exporter-yc
+            patch: |-
+              - op: add
+                path: /spec/template/metadata/annotations
+                value:
+                  vault.hashicorp.com/agent-init-first: "true"
+                  vault.hashicorp.com/agent-inject: "true"
+                  vault.hashicorp.com/agent-pre-populate-only: "true"
+                  vault.hashicorp.com/auth-path: auth/kubernetes
+                  vault.hashicorp.com/role: kafka-exporter-yc
+                  vault.hashicorp.com/agent-inject-secret-kafka-bootstrap: secrets/data/vault/apps/kafka-exporter-yc
+                  vault.hashicorp.com/agent-inject-template-kafka-bootstrap: |-
+                    {{- with secret "secrets/data/vault/apps/kafka-exporter-yc" -}}
+                    {{ index .Data.data "KAFKA_BOOTSTRAP" }}
+                    {{- end -}}
+                  vault.hashicorp.com/agent-inject-secret-kafka-user: secrets/data/vault/apps/kafka-exporter-yc
+                  vault.hashicorp.com/agent-inject-template-kafka-user: |-
+                    {{- with secret "secrets/data/vault/apps/kafka-exporter-yc" -}}
+                    {{ index .Data.data "KAFKA_USER" }}
+                    {{- end -}}
+                  vault.hashicorp.com/agent-inject-secret-kafka-password: secrets/data/vault/apps/kafka-exporter-yc
+                  vault.hashicorp.com/agent-inject-template-kafka-password: |-
+                    {{- with secret "secrets/data/vault/apps/kafka-exporter-yc" -}}
+                    {{ index .Data.data "KAFKA_PASSWORD" }}
+                    {{- end -}}
+                  vault.hashicorp.com/agent-inject-secret-kafka-ca.pem: secrets/data/vault/apps/kafka-exporter-yc
+                  vault.hashicorp.com/agent-inject-template-kafka-ca.pem: |-
+                    {{- with secret "secrets/data/vault/apps/kafka-exporter-yc" -}}
+                    {{ index .Data.data "KAFKA_CA_PEM" }}
+                    {{- end -}}
+              - op: add
+                path: /spec/template/spec/serviceAccountName
+                value: kafka-exporter-yc
+              - op: add
+                path: /spec/template/spec/imagePullSecrets
+                value:
+                  - name: regcred
+              - op: add
+                path: /spec/template/spec/containers/0/command
+                value:
+                  - /bin/sh
+                  - -ec
+              - op: replace
+                path: /spec/template/spec/containers/0/args
+                value:
+                  - |-
+                    KAFKA_BOOTSTRAP="$(cat /vault/secrets/kafka-bootstrap)"
+                    KAFKA_USER="$(cat /vault/secrets/kafka-user)"
+                    KAFKA_PASSWORD="$(cat /vault/secrets/kafka-password)"
+                    if command -v kafka_exporter >/dev/null 2>&1; then
+                      KAFKA_EXPORTER_BIN="$(command -v kafka_exporter)"
+                    else
+                      KAFKA_EXPORTER_BIN=/bin/kafka_exporter
+                    fi
+                    OLD_IFS="${IFS}"
+                    IFS=,
+                    set --
+                    for broker in ${KAFKA_BOOTSTRAP}; do
+                      broker="$(printf '%s' "${broker}" | tr -d '[:space:]')"
+                      if [ -n "${broker}" ]; then
+                        set -- "$@" --kafka.server="${broker}"
+                      fi
+                    done
+                    IFS="${OLD_IFS}"
+                    exec "${KAFKA_EXPORTER_BIN}" \
+                      "$@" \
+                      --sasl.enabled \
+                      --sasl.username="${KAFKA_USER}" \
+                      --sasl.password="${KAFKA_PASSWORD}" \
+                      --sasl.mechanism=scram-sha512 \
+                      --tls.enabled \
+                      --tls.ca-file=/vault/secrets/kafka-ca.pem \
+                      --kafka.labels=yc-kafka \
+                      --topic.exclude='^__.*' \
+                      --verbosity=0
+          - target:
+              group: monitoring.coreos.com
+              version: v1
+              kind: ServiceMonitor
+              name: kafka-exporter-yc
+            patch: |-
+              - op: add
+                path: /spec/selector/matchLabels/app.kubernetes.io~1instance
+                value: kafka-exporter-yc
+  values:
+    fullnameOverride: kafka-exporter-yc
+    image:
+      repository: danielqsj/kafka-exporter
+      tag: latest
+      pullPolicy: IfNotPresent
+    kafkaExporter:
+      kafka:
+        servers:
+          - kafka-bootstrap.from-vault.invalid:9091
+      sasl:
+        enabled: false
+      tls:
+        enabled: false
+    prometheus:
+      serviceMonitor:
+        enabled: true
+        namespace: kafka-exporter
+        interval: 30s
+        additionalLabels:
+          app: kafka-exporter-yc
+        metricRelabelings:
+          - action: replace
+            targetLabel: kafka_instance
+            replacement: yc-kafka
+          - action: replace
+            targetLabel: source_cluster
+            replacement: yc-kafka
+          - action: replace
+            targetLabel: monitored_cluster
+            replacement: yc-kafka
+          - action: replace
+            targetLabel: cluster
+            replacement: brusnika-stage
--- a/clusters/brusnika-stage/infrastructure/kustomization.yaml
+++ b/clusters/brusnika-stage/infrastructure/kustomization.yaml
@ -21,6 +21,9 @@ resources:
  - ./node-exporter-vmnodescrape.yaml
  - ./istio-gateway-stats-scrape.yaml
  - ./istio-dashboard-compat-vmrule.yaml
+  - ./kafka-exporter-yc.yaml
+  - ./kafka-exporter-yc-rules.yaml
+  - ./kafka-exporter-yc-dashboard.yaml
  - ../../../infrastructure/failed-pod-cleanup
 patches:
  - path: ./patches/istio-gateway.yaml
--- a/clusters/brusnika-stage/infrastructure/patches/kafka-exporter.yaml
+++ b/clusters/brusnika-stage/infrastructure/patches/kafka-exporter.yaml
@ -22,6 +22,15 @@ spec:
                path: /spec/template/spec/imagePullSecrets
                value:
                  - name: regcred
+          - target:
+              group: monitoring.coreos.com
+              version: v1
+              kind: ServiceMonitor
+              name: kafka-exporter-kafka-exporter-prod
+            patch: |-
+              - op: add
+                path: /spec/selector/matchLabels/app.kubernetes.io~1instance
+                value: kafka-exporter
  values:
    image:
      repository: danielqsj/kafka-exporter