本來認為部署kafka很簡單，沒想到也折騰了2-3天，這水平沒治了～
kafka從3.4.0版本之后，可以不依賴zookeeper直接使用KRaft模式部署，也就是說部署kafka可以不安裝zookeeper直接部署。
在官網上沒有找到如何使用yaml文件在k8s上部署，搗鼓了2-3天終于穩定部署了，把步驟記錄下來以備后查。
yaml文件內容：
---
apiVersion: v1
kind: Service
metadata:name: kafka-hsnamespace: kafka
spec:clusterIP: Noneselector:app: kafkaports:- port: 9092targetPort: 9092name: kafka-server
---
apiVersion: v1
kind: Service
metadata:name: kafka-svcnamespace: kafka
spec:type: ClusterIPselector:app: kafkaports:- port: 9092targetPort: 9092name: server
---
apiVersion: apps/v1
kind: StatefulSet
metadata:name: kafkanamespace: kafkalabels:app: kafka
spec:serviceName: "kafka-hs" # 指向內部無頭服務replicas: 3selector:matchLabels:app: kafkatemplate:metadata:labels:app: kafkaspec:initContainers:- name: create-data-dirimage: docker.m.daocloud.io/library/busybox:latestimagePullPolicy: IfNotPresentcommand: ['sh', '-c', 'mkdir -p /host-data/$(POD_NAME) && chmod 755 /host-data/$(POD_NAME) && chown -R 1000:1000 /host-data']env:- name: POD_NAMEvalueFrom:fieldRef:fieldPath: metadata.namevolumeMounts:- name: datamountPath: /host-datacontainers:- name: kafkaimage: docker.m.daocloud.io/apache/kafka:4.1.0imagePullPolicy: IfNotPresentports:- containerPort: 9092protocol: TCP- containerPort: 9093protocol: TCPsecurityContext:runAsUser: 1000runAsGroup: 1000env:- name: KAFKA_ENABLE_KRAFTvalue: "yes"- name: KAFKA_PROCESS_ROLESvalue: "broker,controller" # 節點同時擔任 broker 和 controller- name: KAFKA_NODE_IDvalueFrom:fieldRef:fieldPath: metadata.labels['apps.kubernetes.io/pod-index']- name: POD_NAMEvalueFrom:fieldRef:fieldPath: metadata.name- name: KAFKA_CONTROLLER_QUORUM_VOTERSvalue: "0@kafka-0.kafka-hs.kafka.svc.cluster.local:9093,1@kafka-1.kafka-hs.kafka.svc.cluster.local:9093,2@kafka-2.kafka-hs.kafka.svc.cluster.local:9093"- name: KAFKA_LISTENERSvalue: "PLAINTEXT://:9092,CONTROLLER://:9093"- name: KAFKA_ADVERTISED_LISTENERSvalue: "PLAINTEXT://$(POD_NAME).kafka-hs.kafka.svc.cluster.local:9092"- name: KAFKA_LOG_DIRSvalue: "/bitnami/kafka/log"- name: KAFKA_CLUSTER_IDvalue: "oUp8pYCCRTKwXIc8KiQ2Uw"- name: ALLOW_PLAINTEXT_LISTENERvalue: "yes"# 添加存儲初始化命令command: ["sh", "-c"]args:- |# 首次啟動時格式化存儲目錄# KAFKA_CLUSTER_ID="$(/opt/kafka/bin/kafka-storage.sh random-uuid)"if [ ! -f /bitnami/kafka/log ]; thenecho "創建log目錄"mkdir -p /bitnami/kafka/logecho "初始化配置文件"sed -i 's/^log.dirs=.*//g' /opt/kafka/config/server.propertiessed -i 's/^node.id=.*/node.id=$(KAFKA_NODE_ID)/' /opt/kafka/config/server.propertiessed -i 's/localhost/$(POD_NAME).kafka-hs.kafka.svc.cluster.local/g' /opt/kafka/config/server.propertiesecho 'controller.quorum.voters=$(KAFKA_CONTROLLER_QUORUM_VOTERS)' >> /opt/kafka/config/server.propertiesecho 'log.dirs=$(KAFKA_LOG_DIRS)' >> /opt/kafka/config/server.propertiessleep 1echo "配置文件初始化完畢..."echo "cluster.id=$(KAFKA_CLUSTER_ID)" > /bitnami/kafka/cluster.idcat /opt/kafka/config/server.properties/opt/kafka/bin/kafka-storage.sh format \-c /opt/kafka/config/server.properties \-t $(KAFKA_CLUSTER_ID) \--no-initial-controllersecho "格式化log存儲目錄"fisleep 1 # 啟動 Kafkaecho "啟動 kafka"/opt/kafka/bin/kafka-server-start.sh /opt/kafka/config/server.propertiesvolumeMounts:- name: datamountPath: /bitnami/kafkasubPathExpr: $(POD_NAME)volumes:- name: datahostPath:path: /data/juicefs-mnt/kafkatype: DirectoryOrCreate
這里面有幾個關鍵知識點是以前我所沒有涉及的：
1. 通過metadata.labels['apps.kubernetes.io/pod-index']? 可以獲取使用statefulset部署pod的索引;
2. 在volumeMounts:
??????? - name: data
????????? mountPath: /bitnami/kafka
????????? subPathExpr: $(POD_NAME) 中，可以使用subPathExpr獲取環境變量
3. kafka部署是先要進行格式化存儲目錄的，并且在部署過程中設置傳入pod中的環境變量對初始化命令沒有直接影響，必須通過更改配置文件server.properties來修改格式化過程。官網只提供了單機模式部署，初始化過程中的kafka-storage format 命令提供了三種模式：[--standalone |
???????????????????? --no-initial-controllers |
???????????????????? --initial-controllers INITIAL_CONTROLLERS] 其幫助文檔如下：
usage: kafka-storage format [-h] --config CONFIG --cluster-id CLUSTER_ID[--add-scram ADD_SCRAM] [--ignore-formatted][--release-version RELEASE_VERSION][--feature FEATURE] [--standalone |--no-initial-controllers |--initial-controllers INITIAL_CONTROLLERS]optional arguments:-h, --help show this help message and exit--config CONFIG, -c CONFIGThe Kafka configuration file to use.--cluster-id CLUSTER_ID, -t CLUSTER_IDThe cluster ID to use.--add-scram ADD_SCRAM, -S ADD_SCRAMA SCRAM_CREDENTIAL to add to the__cluster_metadata log e.g.'SCRAM-SHA-256=[name=alice,password=alice-secret]''SCRAM-SHA-512=[name=alice,iterations=8192,salt="N3E=",saltedpassword="YCE="]'--ignore-formatted, -gWhen this option is passed, the format commandwill skip over already formatted directoriesrather than failing.--release-version RELEASE_VERSION, -r RELEASE_VERSIONThe release version to use for the initialfeature settings. The minimum is 3.3-IV3; thedefault is 4.1-IV1--feature FEATURE, -f FEATUREThe setting to use for a specific feature, infeature=level format. For example: `kraft.version=1`.--standalone, -s Used to initialize a controller as a single-nodedynamic quorum.--no-initial-controllers, -NUsed to initialize a server without a dynamicquorum topology.--initial-controllers INITIAL_CONTROLLERS, -I INITIAL_CONTROLLERSUsed to initialize a server with a specificdynamic quorum topology. The argument is a comma-separated list of id@hostname:port:directory. Thesame values must be used to format all nodes. Forexample:0@example.com:8082:JEXY6aqzQY-32P5TStzaFg,1@example.com:8083:MvDxzVmcRsaTz33bUuRU6A,2@example.com:8084:07R5amHmR32VDA6jHkGbTA你若用initial-controllers模式需要先轉換directory.id 比較費勁，所以我選擇了在部署過程中直接修改/opt/kafka/config/server.properties ，按照設置傳遞進來的環境變量修改配置文件后，再啟動kafka就可以三節點穩定運行了。
