flink on k8s 官网
https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-release-1.1/docs/try-flink-kubernetes-operator/quick-start/
我的部署脚本和官网不一样，有些地方官网不够详细
部署k8s集群
注意，按照默认配置至少有两台worker
安装helm
https://helm.sh/zh/docs/intro/install/
安装flink opreator
kubectl create -f https://github.com/jetstack/cert-manager/releases/download/v1.8.2/cert-manager.yaml
helm repo add flink-operator-repo https://downloads.apache.org/flink/flink-kubernetes-operator-1.1.0/
helm install flink-kubernetes-operator flink-operator-repo/flink-kubernetes-operator -n flink-
安装完成后，资源如下
[root@k8s1 flinkinstall]# helm list -n flink
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
flink-kubernetes-operator flink 1 2024-03-07 16:57:48.374299701 +0800 CST deployed flink-kubernetes-operator-1.7.0 1.7.0
[root@k8s1 flinkinstall]# kubectl get all -A
NAMESPACE NAME READY STATUS RESTARTS AGE
cert-manager pod/cert-manager-66b646d76-gkw55 1/1 Running 1 2d2h
cert-manager pod/cert-manager-cainjector-59dc9659c7-pkgrm 1/1 Running 1 2d2h
cert-manager pod/cert-manager-webhook-7f7787f7fd-wd5vv 1/1 Running 1 2d2h
flink pod/flink-kubernetes-operator-857d48ff65-45mg2 2/2 Running 6 5d20hNAMESPACE NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
cert-manager service/cert-manager ClusterIP 192.178.138.19 <none> 9402/TCP 2d2h
cert-manager service/cert-manager-webhook ClusterIP 192.178.130.219 <none> 443/TCP 2d2h
flink service/flink-operator-webhook-service ClusterIP 192.178.139.67 <none> 443/TCP 5d20hNAMESPACE NAME READY UP-TO-DATE AVAILABLE AGE
cert-manager deployment.apps/cert-manager 1/1 1 1 2d2h
cert-manager deployment.apps/cert-manager-cainjector 1/1 1 1 2d2h
cert-manager deployment.apps/cert-manager-webhook 1/1 1 1 2d2h
flink deployment.apps/flink-kubernetes-operator 1/1 1 1 5d20hNAMESPACE NAME DESIRED CURRENT READY AGE
cert-manager replicaset.apps/cert-manager-66b646d76 1 1 1 2d2h
cert-manager replicaset.apps/cert-manager-cainjector-59dc9659c7 1 1 1 2d2h
cert-manager replicaset.apps/cert-manager-webhook-7f7787f7fd 1 1 1 2d2h
flink replicaset.apps/flink-kubernetes-operator-857d48ff65 1 1 1 5d20h -
此时k8s集群就可以支持我们按照flink-opreator的指定格式提交flink任务了
提交flink任务
session模式与application模式区别在于资源隔离度
- session模式： jobmanager预先启动，随时准备接收flink jar，启动taskmanager，flink任务结束后jobmanager不退出，所有flink任务共享同一个jobmanager，资源隔离差，某个flink任务导致jobmanager异常，会影响到其他flink任务，小任务，不在乎异常情况可以用
- application模式：每次提交flink任务才会启动一个jobmanger，flink任务结束后，jobmanager也退出，隔离效果好，生产常用
- per-job模式：这个模式与application模式类似， 区别在于client的运行位置，但是新版的flink已经删除了这种提交方式
这里是flink on yarn的运行模式
https://blog.csdn.net/java_creatMylief/article/details/126172793
application模式
apiVersion: flink.apache.org/v1beta1
kind: FlinkDeployment
metadata:
name: pod-template-example
spec:
image: flink:1.15
flinkVersion: v1_15
flinkConfiguration:
taskmanager.numberOfTaskSlots: "2"
serviceAccount: flink
podTemplate:
apiVersion: v1
kind: Pod
metadata:
name: pod-template
spec:
serviceAccount: flink
containers:
# Do not change the main container name
- name: flink-main-container
env:
volumeMounts:
- mountPath: /flink-logs
name: flink-logs
initContainers:
- name: init-nginx
image: busybox
command: [ 'sh','-c','wget http://192.168.33.2/phoenix-client-1.0-SNAPSHOT-jar-with-dependencies.jar -O /flink-logs/StateMachineExample1.jar' ]
volumeMounts:
- mountPath: /flink-logs
name: flink-logs
volumes:
- name: flink-logs
emptyDir: { }
jobManager:
resource:
memory: "1024m"
cpu: 1
taskManager:
resource:
memory: "1024m"
cpu: 1
job:
jarURI: local:///flink-logs/StateMachineExample1.jar
parallelism: 1
entryClass: org.examplexxx.test
args: [/path/from/data,/path/to/data]
initialSavepointPath: hdfs://flink/ckpath/xxxxx
kubectl apply -f ${name}.yaml
kubectl port-forward svc/basic-example-rest 8081 --address 192.168.33.81
访问 http://192.168.33.81:8081jarURI: local:///flink-logs/StateMachineExample1.jar
此处jarURL只得是docker内部路径，且不支持远程路径（http/s3/hdfs），因此需要将jar包放到docker内部。
1、可以将flink版本和jar包打到一个镜像中。
2、可以使用pvc挂载进去。
3、使用initContainers和 containers使用相同的挂载路径，然后使用远程文件下载放到挂载路径中，containers就能获取到该jar包
此处使用第三种情况，使用initContainers变相支持远程文件地址，使用起来比较方便。
yarn-application 对比
| yarn-application | k8s-application |
|---|---|
| -p (并行度) | spec.job.parallelism |
| -yjm (jobmanager内存) | spec.jobManager.resource.memory |
| -ytm (taskmanager内存) | spec.taskManager.resource.memory |
| -ys (taskmanger的slot槽数) | spec.flinkConfiguration.taskmanager.numberOfTaskSlots |
| -c (主类) | spec.job.entryClass |
| jar (jar包) | spec.job.jarURI |
| -s (恢复点启动) | spec.job.initialSavepointPath |
session模式
部署session cluster
apiVersion: flink.apache.org/v1beta1
kind: FlinkDeployment
metadata:
namespace: flink
name: session-deployment-only
spec:
image: flink:1.13.6
flinkVersion: v1_13
imagePullPolicy: IfNotPresent # 镜像拉去策略，优先本地，没有，仓库拉去
ingress:
template: "flink.k8s.io/{{namespace)}/{{name}}(/|$)(.*)"
className: "Nginx"
annotations:
taskmanager.numberOfTaskSlots: "2"
serviceAccount: flink
jobManager:
replicas: 1
resource:
memory: "1024m"
cpu: 1
taskManager:
replicas: 1
resource:
memory: "1024m"
cpu: 1
kubectl apply -f ${name}.yaml
部署cluster完成，配置svcType 后即可访问，flink web ui，此时jobManager是启动着的 taskmanager随着flink jar进行启动和停止
部署flink jar
apiVersion: flink.apache.org/v1beta1
kind: FlinkSessionJob
metadata:
namespace: flink
name: session-job-only
job:
jarUrl: sasa
entryClass: aa
parallelism: 1
upgradeMode: stateless
kubectl apply -f ${name}.yaml