当前位置：首页 > news >正文

Flink on K8S集群搭建及StreamPark平台安装

news 2026/3/27 18:45:51

1.环境准备

1.1 介绍

在使用 Flink&Spark 时发现从编程模型, 启动配置到运维管理都有很多可以抽象共用的地方, 目前streampark提供了一个flink一站式的流处理作业开发管理平台, 从流处理作业开发到上线全生命周期都做了支持, 是一个一站式的流出来计算平台。
未来spark开发也在规划范围内，目前还不支持

1.2 下载

StreamPark安装包下载：https://streampark.apache.org/download
StreamPark官网：https://streampark.apache.org/docs/intro
最新版为2.1.2，本次安装为2.1.2版本

1.3 已有组件及版本

序号	名称	配置
1	K8S(TEK)	CPU:316,Me:361.57G,Storage:349.1G,Pods:3253,IP:3*273 (v1.26.1)
2	NFS(CFS)	40G磁盘（目前，可扩充）
3	Harbor(容器仓库）	未知
4	flink	flink-(1.13.0/1.14.4/1.16.2)
5	StreamPark	2.1.2
6	Mysql	5.7

2.挂载与权限

2.1 k8s挂载

找运维开通权限，主要是kubectl的权限

下载kubectl
权限文件：/root/.kube/config

2.2 nfs挂载

找运维要服务ip

#挂载前，请确保系统中已经安装了nfs-utils或nfs-common
sudo yum install nfs-utils
#创建待挂载目标目录
mkdir <待挂载目标目录>   mkdir /localfolder/
#挂载文件系统
#挂载 CFS 根目录
#以下命令可以到 CFS 控制台-文件系统详情-挂载点详情中获取，由于部分旧版本文件系统不支持 noresvport 参数，具体挂载命令请以控制台建议命令为主。配置 norevsport 参数后，在网络重连时使用新的 TCP 端口，可以保障在网络异常到恢复期间、客户端和文件系统的连接不会中断，建议启用该参数。
#另，部分旧版本 Linux 内核需要使用 vers=4 挂载，若使用 vers=4.0 挂载有异常，可以尝试修改为 vers=4。
sudo mount -t nfs -o vers=4.0,noresvport <挂载点 IP>:/ <待挂载目录>
sudo mount -t nfs -o vers=4.0,noresvport 10.0.24.4:/ /localfolder
#挂载 CFS 子目录
#以下命令可以到 CFS 控制台-文件系统详情-挂载点详情中获取，由于部分旧版本文件系统不支持 noresvport 参数，具体挂载命令请以控制台建议命令为主。配置 norevsport 参数后，在网络重连时使用新的 TCP 端口，可以保障在网络异常到恢复期间、客户端和文件系统的连接不会中断，建议启用该参数。
#另，部分旧版本 Linux 内核需要使用 vers=4 挂载，若使用 vers=4.0 挂载有异常，可以尝试修改为 vers=4。
sudo mount -t nfs -o vers=4.0,noresvport 10.0.24.4:/subfolder /localfolder

腾讯nfs(CFS)使用文档：https://cloud.tencent.com/document/product/582/11523

2.3 Harbor

需要找运维开通权限

此处注意：项目库要公开才可以使用

3.测试Flink

在k8s为flink单独开通命名空间，并创建相应账户

kubectl create clusterrolebinding flink-role-bind --clusterrole=edit --serviceaccount=flink:flink

在有k8s操作权限的节点运行flink-session

bin/flink run \
-e kubernetes-session \
-Dkubernetes.namespace=flink \
-Dkubernetes.rest-service.exposed.type=NodePort \
-Dkubernetes.cluster-id=flink-cluster \
-c WordCount1 \
/data/package/jar/flink_test-1.0-SNAPSHOT.jar#参考配置
bin/kubernetes-session.sh \
-Dkubernetes.namespace=flink \
-Dkubernetes.jobmanager.service-account=flink \
-Dkubernetes.rest-service.exposed.type=NodePort \
-Dkubernetes.cluster-id=flink-cluster \
-Dkubernetes.jobmanager.cpu=0.2 \
-Djobmanager.memory.process.size=1024m \
-Dresourcemanager.taskmanager-timeout=3600000 \
-Dkubernetes.taskmanager.cpu=0.2 \
-Dtaskmanager.memory.process.size=1024m \
-Dtaskmanager.numberOfTaskSlots=1

4.安装StreamPark

4.1 streampark（k8s)镜像打包


#编译
./build #注意maven 镜像配置，要不找不到依赖包同时要安装npm
#测试npm是否安装
npm -v#streampark安装包添加mysql连接包
cp /data/module/streampart_2.12-2.1.2/lib/mysql-connector-java-8.0.30.jar lib/
#配置maven配置拷贝
cp /data/module/maven-3.6.3/conf/settings.xml /data/module/docker/streampark-docker/
#修改application.yml
profiles.active: mysql #[h2,pgsql,mysql]
lark-url: https://open.feishu.cn
workspace:local: /opt/streampark_workspace
#配置application-mysql.yml
tee /data/module/docker/streampark-docker/streampark-2.1.2/conf/application-mysql.yml <<-'EOF'
spring.datasource.driver-class-name: com.mysql.cj.jdbc.Driver
streampark.docker.http-client.docker-host: ${DOCKER_HOST:}
streampark.maven.settings: ${MAVEN_SETTINGS_PATH:/root/.m2/settings.xml}
streampark.workspace.local: ${WORKSPACE_PATH:/opt/streampark_workspace}
EOF# 编写Dockerfile
#需要提前准备kubectl、settings.xml 、config（kubectl的密钥）
tee docker<<-'EOF'
FROM flink:1.17.1-scala_2.12-java8
WORKDIR /opt/streampark/
ADD ./streampark-2.1.2/ /opt/streampark/
ADD ./kubectl /opt/streampark/
ADD ./settings.xml /root/.m2/
USER root
RUN sed -i -e 's/eval $NOHUP/eval/' bin/streampark.sh \&& sed -i -e 's/>> "$APP_OUT" 2>&1 "&"//' bin/streampark.sh \&& install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl \&& mkdir -p ~/.kube
WORKDIR /opt/streampark/
ADD ./config /root/.kube/
RUN chown -R flink /opt/streampark/EXPOSE 10000
EOF
#构建镜像
docker build -f Dockerfile -t apache/streampark-flink:2.1.2 .
#推送镜像到仓库
docker tag apache/streampark-flink:2.1.2 storage/bigdata/streampark-flink:2.1.2
docker push  storage/bigdata/streampark-flink:2.1.2docker tag apache/streampark-flink:2.1.2-rc4 storage/bigdata/streampark-flink:2.1.2-rc4
docker push  storage/bigdata/streampark-flink:2.1.2-rc4

4.2部署MySQL的pod


#k8s上创建mysql的namespace
#含义：kubectl create clusterrolebinding ClusterRoleBinding名 --clusterrole=绑定的Role serviceaccount=被绑定的SA -n 命名空间
kubectl create namespace mysql
kubectl create serviceaccount mysql
kubectl create clusterrolebinding mysql-role-bind --clusterrole=edit --serviceaccount=mysql:mysql -n mysql
clusterrolebinding.rbac.authorization.k8s.io/mysql-role-bind created
#查看角色绑定
kubectl get clusterrolebinding flink-role-bind -n flink -o yaml
kubectl get clusterrolebinding mysql-role-bind -n mysql -o yaml
kubectl get clusterrolebinding mysql-role-bind -n flink -o yaml
#配置pvc和pv和nfs指定(腾讯nfs可直接使用）
apiVersion: v1
kind: PersistentVolume
metadata:name: data-mysql
spec:accessModes:- ReadWriteManycapacity:storage: 10Gicsi:driver: com.tencent.cloud.csi.cfsvolumeAttributes:host: x x xpath: /data_mysqlvers: "4"volumeHandle: cfs   #此处需要每个pv都不相同，否则挂载两个pvc会报错persistentVolumeReclaimPolicy: RetainstorageClassName: data-mysqlvolumeMode: Filesystem
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: data-mysqlnamespace: mysql
spec:accessModes:- ReadWriteManyresources:requests:storage: 10GistorageClassName: data-mysql
#docker拉去mysql镜像并上传至镜像库
docker tag mysql:5.7 storage/bigdata/mysql:5.7
docker push storage/bigdata/mysql:5.7
#查看集群的node ip
kubectl get node
# 编写mysql的yaml文件，并做配置
sudo mkdir -p /data/module/docker/mysql/{conf,data}/
sudo tee /data/module/docker/mysql/conf/pod-db-mysql.yaml <<-'EOF'
apiVersion: v1
kind: ConfigMap
metadata:name: conf-mysqlnamespace: mysql
data:mysql.cnf: |[mysqld]#Mysql服务的唯一编号 每个mysql服务Id需唯一server-id=1# 允许访问的IP网段bind-address=0.0.0.0#设置时区default-time_zone='+8:00'#数据库默认字符集,主流字符集支持一些特殊表情符号（特殊表情符占用4个字节）character-set-server=utf8mb4#数据库字符集对应一些排序等规则，注意要和character-set-server对应collation-server=utf8mb4_general_ci#设置client连接mysql时的字符集,防止乱码init_connect='SET NAMES utf8mb4'#是否对sql语句大小写敏感，1表示不敏感lower_case_table_names=1#最大连接数max_connections=400#最大错误连接数max_connect_errors=1000#TIMESTAMP如果没有显示声明NOT NULL，允许NULL值explicit_defaults_for_timestamp=true#SQL数据包发送的大小，如果有BLOB对象建议修改成1Gmax_allowed_packet=128M#MySQL连接闲置超过一定时间后(单位：秒)将会被强行关闭#MySQL默认的wait_timeout  值为8个小时, interactive_timeout参数需要同时配置才能生效interactive_timeout=3600wait_timeout=3600
---
apiVersion: v1
kind: Pod
metadata:name: pod-db-mysqlnamespace: mysql
spec:#serviceAccount: mysqlnodeName: xxxhostNetwork: true #主机网络可见(会占用node端口)containers:- name: mysql-k8simage: storage/bigdata/mysql:5.7env:- name: TZvalue: "Asia/Shanghai"- name: LANGvalue: "zh_CN.UTF-8"- name: MYSQL_ROOT_PASSWORDvalue: "xxxx"ports:#- containerPort: 3306volumeMounts:- mountPath: /var/lib/mysqlsubPath: mysqlname: data-mysql- mountPath: /etc/mysql/conf.dname: conf-volumereadOnly: truevolumes:- name: data-mysqlpersistentVolumeClaim:claimName: data-mysql- name: conf-volumeconfigMap:name: conf-mysql
EOF#kubectl delete -f /opt/module/k3s/conf/pod-db-mysql-k3s100.yaml
#启动pod
kubectl apply -f /data/module/docker/mysql/conf/pod-db-mysql.yaml
kubectl delete -f /data/module/docker/mysql/conf/pod-db-mysql.yaml
# 稍等片刻
kubectl get pod -A -owide
kubectl describe pod pod-db-mysql
kubectl logs --tail=100 -f pod-db-mysql -n mysql

4.3 mysql 数据库初始化

# 复制数据库文件
cp -r /data/software/incubator-streampark-2.1.2-rc3/streampark-console/streampark-console-service/src/main/assembly/script/ /localnfs/data_mysql/mysql/streampark-sql/
cp -r /data/module/docker/streampark-docker/streampark-2.2.0/script/ /localnfs/data_mysql/mysql/streampark-sql/
# 创建用户及数据库
# 进入mysql容器
kubectl exec -n mysql -it pod-db-mysql  -- bash
#------------------------进入mysql容器----------------------------
mysql -uroot -proot
create database if not exists `streampark` character set utf8mb4 collate utf8mb4_general_ci;
create user 'xxxx'@'%' IDENTIFIED WITH mysql_native_password by 'xxx';
grant ALL PRIVILEGES ON streampark.* to 'xxxx'@'%';
flush privileges;
-- 导入数据文件
use streampark;
source /var/lib/mysql/streampark-sql/schema/mysql-schema.sql
source /var/lib/mysql/streampark-sql/data/mysql-data.sql
-- 退出mysql
quit
#------------------------退出mysql容器------------------------
exit#查看mysql 的建表
vim /data/module/docker/streampark-docker/streampark-2.2.0/script/schema/mysql-schema.sql
vim /data/module/docker/streampark-docker/streampark-2.2.0/script/data/mysql-data.sql

4.4 创建StreamPark的pod


#k8s上创建mysql的namespace
#含义：kubectl create clusterrolebinding ClusterRoleBinding名 --clusterrole=绑定的Role serviceaccount=被绑定的SA -n 命名空间
kubectl create namespace streampark
kubectl create serviceaccount streampark
kubectl create clusterrolebinding streampark-role-bind --clusterrole=edit --serviceaccount=streampark:streampark -n streampark
clusterrolebinding.rbac.authorization.k8s.io/mysql-role-bind created#配置pvc和pv和nfs指定
vim pv-pvc-streampark.yamlapiVersion: v1
kind: PersistentVolume
metadata:name: data-streampark
spec:accessModes:- ReadWriteManycapacity:storage: 10Gicsi:driver: com.tencent.cloud.csi.cfsvolumeAttributes:host: xxxpath: /data_streamparkvers: "4"volumeHandle: cfs    #此处需要每个pv都不相同，否则挂载两个pvc会报错persistentVolumeReclaimPolicy: RetainstorageClassName: data-streamparkvolumeMode: Filesystem
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: data-streamparknamespace: streampark
spec:accessModes:- ReadWriteManyresources:requests:storage: 10GistorageClassName: data-streampark#执行
kubectl apply -f /data/module/docker/k8s/pv-pvc-streampark.yaml
kubectl delete -f /data/module/docker/k8s/pv-pvc-streampark.yaml
# 在xxx 上执行,在指定 节点 xxx 安装StreamPark
sudo tee /data/pod-app-streampark.yaml <<-'EOF'
apiVersion: apps/v1
kind: Deployment
metadata:labels:app: pod-app-streamparkname: pod-app-streamparknamespace: streampark
spec:replicas: 1selector:matchLabels:app: pod-app-streamparktemplate:metadata:labels:app: pod-app-streamparkspec:nodeName: xxxhostNetwork: true #主机网络可见(会占用node端口)containers:- name: streamparkimage: storage/bigdata/streampark-flink:2.1.2-rc4imagePullPolicy: Alwaysenv:- name: TZvalue: "Asia/Shanghai"- name: LANGvalue: "zh_CN.UTF-8"- name: SPRING_PROFILES_ACTIVEvalue: "mysql"- name: SPRING_DATASOURCE_URLvalue: "jdbc:mysql://xxx:3306/streampark?useSSL=false&useUnicode=true&characterEncoding=UTF-8&allowPublicKeyRetrieval=false&useJDBCCompliantTimezoneShift=true&useLegacyDatetimeCode=false&serverTimezone=GMT%2B8"- name: SPRING_DATASOURCE_USERNAMEvalue: "xxxx"- name: SPRING_DATASOURCE_PASSWORDvalue: "xxxx"- name: DOCKER_HOSTvalue: "tcp://xxx:2375"- name: DEBUG_OPTS #调试端口参数value: "-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=*:10001"ports:#- containerPort: 10000volumeMounts:- mountPath: /root/.kubesubPath: .kubename: conf-volume- mountPath: /opt/streampark_workspacesubPath: streampark_workspacename: data-volumecommand: ["sh","-c","bash bin/startup.sh debug"]volumes:- name: conf-volumehostPath:path: /roottype: DirectoryOrCreate- name: data-volumenfs:path: /data_streamparkserver: xxxEOF#kubectl delete -f /opt/module/k3s/conf/pod-app-streampark-k3s100.yaml
#启动pod
kubectl apply -f /data/pod-app-streampark.yaml
kubectl delete -f /data/pod-app-streampark.yaml
# 稍等片刻
kubectl get pod -A -o wide -n mysql
kubectl describe pod pod-app-streampark-k3s100
kubectl logs --tail=1000 -f pod-app-streampark -n mysql
# 进入mysql容器中创建用户及数据库
kubectl exec -n mysql -it pod-app-streampark  -- bash
#-c streampark-k3s100# 为默认命名空间添加权限 kubectl create clusterrolebinding flink-role-binding-default --clusterrole=edit --serviceaccount=flink_dev:default
kubectl create clusterrolebinding flink-role-binding-default --clusterrole=edit --serviceaccount=default:default