在 kubernetes 集群外访问 bitnami/kafka

Question

我目前正在使用 bitnami/kafka image(https://hub.docker.com/r/bitnami/kafka) 并将其部署在 kubernetes 上。

kubernetes 大师：1
kubernetes 工人：3

在集群中，其他应用程序能够找到 kafka。尝试从集群外部访问 kafka 容器时会出现问题。在阅读时，我读到我们需要为外部 kafka 客户端设置属性 "advertised.listener=PLAINTTEXT://hostname:port_number"。

我目前正在引用“https://github.com/bitnami/charts/tree/master/bitnami/kafka”。在我的 values.yaml 文件中添加了

values.yaml

advertisedListeners1: 10.21.0.191

和statefulset.yaml

    - name: KAFKA_CFG_ADVERTISED_LISTENERS
      value: 'PLAINTEXT://{{ .Values.advertisedListeners }}:9092'

对于单个 kafka 实例，它工作正常。

但对于 3 节点的 kafka 集群，我更改了一些配置，如下所示： values.yaml

advertisedListeners1: 10.21.0.191
advertisedListeners2: 10.21.0.192
advertisedListeners3: 10.21.0.193

和Statefulset.yaml

    - name: KAFKA_CFG_ADVERTISED_LISTENERS
      {{- if $MY_POD_NAME := "kafka-0" }}
      value: 'PLAINTEXT://{{ .Values.advertisedListeners1 }}:9092'
      {{- else if $MY_POD_NAME := "kafka-1" }}
      value: 'PLAINTEXT://{{ .Values.advertisedListeners2 }}:9092'
      {{- else if $MY_POD_NAME := "kafka-2" }}
      value: 'PLAINTEXT://{{ .Values.advertisedListeners3 }}:9092'
      {{- end }}

预期结果是所有 3 个 kafka 实例都应 advertised.listener 属性设置为工作节点 ip 地址。

示例：

kafka-0 --> "PLAINTEXT://10.21.0.191:9092"
kafka-1 --> "PLAINTEXT://10.21.0.192:9092"
kafka-3 --> "PLAINTEXT://10.21.0.193:9092"

目前只有一个kafka pod处于up状态，运行另外两个将进入crashloopbackoff状态。

和另外两个 pods 显示错误为：

[2019-10-20 13:09:37,753] 信息 [LogDirFailureHandler]：开始（kafka.server.ReplicaManager$LogDirFailureHandler） [2019-10-20 13:09:37,786] 错误 [KafkaServer id=1002] KafkaServer 启动期间出现致命错误。准备关机 (kafka.server.KafkaServer) java.lang.IllegalArgumentException：要求失败：已在广告侦听器中配置的端点 10.21.0.191:9092 已由代理 1001 注册在 scala.Predef$.require(Predef.scala:224) 在 kafka.server.KafkaServer$$anonfun$createBrokerInfo$2.apply(KafkaServer.scala:399) 在 kafka.server.KafkaServer$$anonfun$createBrokerInfo$2.apply(KafkaServer.scala:397) 在 scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59) 在 scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48) 在 kafka.server.KafkaServer.createBrokerInfo(KafkaServer.scala:397) 在 kafka.server.KafkaServer.startup(KafkaServer.scala:261) 在 kafka.server.KafkaServerStartable.startup(KafkaServerStartable.scala:38) 在 kafka.Kafka$.main(Kafka.scala:84) 在 kafka.Kafka.main(Kafka.scala)

这意味着 statefulset.yaml 中应用的逻辑不起作用。谁能帮我解决这个问题..？

任何帮助将不胜感激..

kubectl get statefulset kafka -o yaml

的输出

apiVersion: apps/v1
kind: StatefulSet
metadata:
  creationTimestamp: "2019-10-29T07:04:12Z"
  generation: 1
  labels:
    app.kubernetes.io/component: kafka
    app.kubernetes.io/instance: kafka
    app.kubernetes.io/managed-by: Tiller
    app.kubernetes.io/name: kafka
    helm.sh/chart: kafka-6.0.1
  name: kafka
  namespace: default
  resourceVersion: "12189730"
  selfLink: /apis/apps/v1/namespaces/default/statefulsets/kafka
  uid: d40cfd5f-46a6-49d0-a9d3-e3a851356063
spec:
  podManagementPolicy: Parallel
  replicas: 3
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app.kubernetes.io/component: kafka
      app.kubernetes.io/instance: kafka
      app.kubernetes.io/name: kafka
  serviceName: kafka-headless
  template:
    metadata:
      creationTimestamp: null
      labels:
        app.kubernetes.io/component: kafka
        app.kubernetes.io/instance: kafka
        app.kubernetes.io/managed-by: Tiller
        app.kubernetes.io/name: kafka
        helm.sh/chart: kafka-6.0.1
      name: kafka
    spec:
      containers:
      - env:
        - name: MY_POD_IP
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: status.podIP
        - name: MY_POD_NAME
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: metadata.name
        - name: KAFKA_CFG_ZOOKEEPER_CONNECT
          value: kafka-zookeeper
        - name: KAFKA_PORT_NUMBER
          value: "9092"
        - name: KAFKA_CFG_LISTENERS
          value: PLAINTEXT://:$(KAFKA_PORT_NUMBER)
        - name: KAFKA_CFG_ADVERTISED_LISTENERS
          value: PLAINTEXT://10.21.0.191:9092
        - name: ALLOW_PLAINTEXT_LISTENER
          value: "yes"
        - name: KAFKA_CFG_BROKER_ID
          value: "-1"
        - name: KAFKA_CFG_DELETE_TOPIC_ENABLE
          value: "false"
        - name: KAFKA_HEAP_OPTS
          value: -Xmx1024m -Xms1024m
        - name: KAFKA_CFG_LOG_FLUSH_INTERVAL_MESSAGES
          value: "10000"
        - name: KAFKA_CFG_LOG_FLUSH_INTERVAL_MS
          value: "1000"
        - name: KAFKA_CFG_LOG_RETENTION_BYTES
          value: "1073741824"
        - name: KAFKA_CFG_LOG_RETENTION_CHECK_INTERVALS_MS
          value: "300000"
        - name: KAFKA_CFG_LOG_RETENTION_HOURS
          value: "168"
        - name: KAFKA_CFG_LOG_MESSAGE_FORMAT_VERSION
        - name: KAFKA_CFG_MESSAGE_MAX_BYTES
          value: "1000012"
        - name: KAFKA_CFG_LOG_SEGMENT_BYTES
          value: "1073741824"
        - name: KAFKA_CFG_LOG_DIRS
          value: /bitnami/kafka/data
        - name: KAFKA_CFG_DEFAULT_REPLICATION_FACTOR
          value: "1"
        - name: KAFKA_CFG_OFFSETS_TOPIC_REPLICATION_FACTOR
          value: "1"
        - name: KAFKA_CFG_TRANSACTION_STATE_LOG_REPLICATION_FACTOR
          value: "1"
        - name: KAFKA_CFG_SSL_ENDPOINT_IDENTIFICATION_ALGORITHM
          value: https
        - name: KAFKA_CFG_TRANSACTION_STATE_LOG_MIN_ISR
          value: "1"
        - name: KAFKA_CFG_NUM_IO_THREADS
          value: "8"
        - name: KAFKA_CFG_NUM_NETWORK_THREADS
          value: "3"
        - name: KAFKA_CFG_NUM_PARTITIONS
          value: "1"
        - name: KAFKA_CFG_NUM_RECOVERY_THREADS_PER_DATA_DIR
          value: "1"
        - name: KAFKA_CFG_SOCKET_RECEIVE_BUFFER_BYTES
          value: "102400"
        - name: KAFKA_CFG_SOCKET_REQUEST_MAX_BYTES
          value: "104857600"
        - name: KAFKA_CFG_SOCKET_SEND_BUFFER_BYTES
          value: "102400"
        - name: KAFKA_CFG_ZOOKEEPER_CONNECTION_TIMEOUT_MS
          value: "6000"
        image: docker.io/bitnami/kafka:2.3.0-debian-9-r88
        imagePullPolicy: IfNotPresent
        livenessProbe:
          failureThreshold: 2
          initialDelaySeconds: 10
          periodSeconds: 10
          successThreshold: 1
          tcpSocket:
            port: kafka
          timeoutSeconds: 5
        name: kafka
        ports:
        - containerPort: 9092
          name: kafka
          protocol: TCP
        readinessProbe:
          failureThreshold: 6
          initialDelaySeconds: 5
          periodSeconds: 10
          successThreshold: 1
          tcpSocket:
            port: kafka
          timeoutSeconds: 5
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /bitnami/kafka
          name: data
      dnsPolicy: ClusterFirst
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext:
        fsGroup: 1001
        runAsUser: 1001
      terminationGracePeriodSeconds: 30
  updateStrategy:
    type: RollingUpdate
  volumeClaimTemplates:
  - metadata:
      creationTimestamp: null
      name: data
    spec:
      accessModes:
      - ReadWriteOnce
      resources:
        requests:
          storage: 8Gi
      volumeMode: Filesystem
    status:
      phase: Pending
status:
  collisionCount: 0
  currentReplicas: 3
  currentRevision: kafka-56ff499d74
  observedGeneration: 1
  readyReplicas: 1
  replicas: 3
  updateRevision: kafka-56ff499d74
  updatedReplicas: 3

Answer 1

我发现您在为 StatefulSet 中的不同 pods 传递不同的环境变量时遇到了一些麻烦。

您正在尝试使用 helm 模板实现此目的：

- name: KAFKA_CFG_ADVERTISED_LISTENERS
  {{- if $MY_POD_NAME := "kafka-0" }}
  value: 'PLAINTEXT://{{ .Values.advertisedListeners1 }}:9092'
  {{- else if $MY_POD_NAME := "kafka-1" }}
  value: 'PLAINTEXT://{{ .Values.advertisedListeners2 }}:9092'
  {{- else if $MY_POD_NAME := "kafka-2" }}
  value: 'PLAINTEXT://{{ .Values.advertisedListeners3 }}:9092'
  {{- end }}

在helm template guide documentation中你可以找到这样的解释：

In Helm templates, a variable is a named reference to another object. It follows the form $name. Variables are assigned with a special assignment operator: :=.

现在让我们看看您的代码：

{{- if $MY_POD_NAME := "kafka-0" }}

这是变量赋值，不是比较在这个赋值之后，if 语句将这个表达式计算为 true 这就是为什么在你的 staefulset yaml 清单你将其视为输出：

- name: KAFKA_CFG_ADVERTISED_LISTENERS
    value: PLAINTEXT://10.21.0.191:9092

要使其按预期工作，您不应使用 helm 模板。它不会起作用。

一种方法是为每个 kafka 节点创建单独的环境变量并将所有这些变量传递给所有 pods，如下所示：

- env:
  - name: MY_POD_NAME
    valueFrom:
      fieldRef:
        apiVersion: v1
        fieldPath: metadata.name
  - name: KAFKA_0
      value: 10.21.0.191
  - name: KAFKA_1
      value: 10.21.0.192
  - name: KAFKA_2
      value: 10.21.0.193
#  - name: KAFKA_CFG_ADVERTISED_LISTENERS
#      value: PLAINTEXT://$MY_POD_NAME:9092

并使用修改后的启动脚本创建您自己的 docker 图像，该脚本将导出 KAFKA_CFG_ADVERTISED_LISTENERS 变量具有适当的值取决于 MY_POD_NAME.

如果您不想创建自己的映像，您可以创建一个 ConfigMap 并修改 entrypoint.sh 并挂载它代替旧的entrypoint.sh（你也可以使用任何其他文件，看看here 有关如何构建 kafka 映像的更多信息）。

安装 ConfigMap 看起来像这样：

apiVersion: v1
kind: Pod
metadata:
  name: test
spec:
  containers:
    - name: test-container
      image: docker.io/bitnami/kafka:2.3.0-debian-9-r88
      volumeMounts:
      - name: config-volume
        mountPath: /entrypoint.sh
        subPath: entrypoint.sh
  volumes:
    - name: config-volume
      configMap:
        # Provide the name of the ConfigMap containing the files you want
        # to add to the container
        name: kafka-entrypoint-config
        defaultMode: 0744 # remember to add proper (executable) permissions

apiVersion: v1
kind: ConfigMap
metadata:
  name: kafka-entrypoint-config
  namespace: default
data:
  entrypoint.sh: |
    #!/bin/bash
    # Here add modified entrypoint script

如果有帮助，请告诉我。

Answer 2

我认为 The helm chart 没有将您的外部（到 kubernetes）网络列入 advertised.listeners 的白名单。我通过像这样重新配置 helm values.yaml 解决了类似的问题。在我的例子中，127.0.0.1 网络是 mac，你的可能不同：

externalAccess:
  enabled: true
  autoDiscovery:
    enabled: false
    image:
      registry: docker.io
      repository: bitnami/kubectl
      tag: 1.23.4-debian-10-r17
      pullPolicy: IfNotPresent
      pullSecrets: []
    resources:
      limits: {}
      requests: {}
  service:
    type: NodePort
    port: 9094
    loadBalancerIPs: []
    loadBalancerSourceRanges: []
    nodePorts:
      - 30000
      - 30001
      - 30002
    useHostIPs: false
    annotations: {}
    domain: 127.0.0.1

在 kubernetes 集群外访问 bitnami/kafka

Accessing bitnami/kafka outside the kubernetes cluster

bitnami

apache-kafka

kubernetes

kubernetes-helm

kubernetes-statefulset