基于Headless构建高可用spark+pyspark集群

本文主要是介绍基于Headless构建高可用spark+pyspark集群，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

1、创建Headless Service服务

Headless 服务类型并不分配容器云虚拟 IP，而是直接暴露所属 Pod 的 DNS 记录。没有默认负载均衡器，可直接访问 Pod IP 地址。因此，当我们需要与集群内真实的 Pod IP 地址进行直接交互时，Headless 服务就很有用。
其中Service的关键配置如下：clusterIP: None，不让其获取clusterIP ， DNS解析的时候直接走pod。

---
kind: Service
apiVersion: v1
metadata:name: ecc-spark-servicenamespace: ecc-spark-cluster
spec:clusterIP: Noneports:- port: 7077protocol: TCPtargetPort: 7077name: spark- port: 10000protocol: TCPtargetPort: 10000name: thrift-server-tcp- port: 8080targetPort: 8080name: http- port: 45970protocol: TCPtargetPort: 45970name: thrift-server-driver-tcp  - port: 45980protocol: TCPtargetPort: 45980name: thrift-server-blockmanager-tcp    - port: 4040protocol: TCPtargetPort: 4040name: thrift-server-tasks-tcp              selector:app: ecc-spark-serviceEOF

Service的完全域名： ecc-spark-service.ecc-spark-cluster.svc.cluster.local
headless service的完全域名: headless-service.ecc-spark-cluster.svc.cluster.local
在容器里面ping 完全域名， service解析出的地址是clusterIP，headless service 解析出来的地址是 pod IP。

2、构建spark集群

2.1 、创建spark master

spark master分为两个部分，一个是类型为ReplicationController的主体，命名为ecc-spark-master.yaml，另一部分为一个service，暴露master的7077端口给slave使用。

#如下是把thriftserver部署在master节点，则需要暴露thriftserver端口、driver端口、
#blockmanager端口服务,以提供worker节点executor与driver交互.
cat >ecc-spark-master.yaml <<EOF
kind: Deployment
apiVersion: apps/v1
metadata:name: ecc-spark-masternamespace: ecc-spark-clusterlabels:app: ecc-spark-master
spec:replicas: 1selector:matchLabels:app: ecc-spark-mastertemplate:metadata:labels:app: ecc-spark-masterspec:serviceAccountName: spark-cdpsecurityContext: {}dnsPolicy: ClusterFirsthostname: ecc-spark-mastercontainers:- name: ecc-spark-masterimage: spark:3.4.1imagePullPolicy: IfNotPresentcommand: ["/bin/sh"]args: ["-c","sh /opt/spark/sbin/start-master.sh && tail -f /opt/spark/logs/spark--org.apache.spark.deploy.master.Master-1-*"]ports:- containerPort: 7077- containerPort: 8080volumeMounts:- mountPath: /opt/usrjars/name: ecc-spark-pvclivenessProbe:failureThreshold: 9initialDelaySeconds: 2periodSeconds: 15successThreshold: 1tcpSocket:port: 8080timeoutSeconds: 10resources:requests:cpu: "2"memory: "6Gi"limits:cpu: "2"memory: "6Gi"- env:- SPARK_LOCAL_DIRSvalue: "/odsdata/sparkdirs/"             volumes:- name: ecc-spark-pvcpersistentVolumeClaim:claimName: ecc-spark-pvc-static

2.2、创建spark worker

在启动spark worker脚本中需要传入master的地址，在容器云kubernetes dns且设置了service的缘故，可以通过ecc-spark-master.ecc-spark-cluster.svc.cluster.local:7077访问。

cat >ecc-spark-worker.yaml <<EOF
kind: Deployment
apiVersion: apps/v1
metadata:name: ecc-spark-workernamespace: ecc-spark-clusterlabels:app: ecc-spark-worker
spec:replicas: 1selector:matchLabels:app: ecc-spark-workertemplate:metadata:labels:app: ecc-spark-workerspec:serviceAccountName: spark-cdpsecurityContext: {}dnsPolicy: ClusterFirsthostname: ecc-spark-workercontainers:- name: ecc-spark-workerimage: spark:3.4.1imagePullPolicy: IfNotPresentcommand: ["/bin/sh"]args: ["-c","sh /opt/spark/sbin/start-worker.sh spark://ecc-spark-master.ecc-spark-cluster.svc.cluster.local:7077;tail -f /opt/spark/logs/spark--org.apache.spark.deploy.worker.Worker*"]ports:- containerPort: 8081volumeMounts:- mountPath: /opt/usrjars/name: ecc-spark-pvcresources:requests:cpu: "2"memory: "2Gi"limits:cpu: "2"memory: "4Gi"- env:- SPARK_LOCAL_DIRSvalue: "/odsdata/sparkdirs/"              volumes:- name: ecc-spark-pvcpersistentVolumeClaim:claimName: ecc-spark-pvc-staticEOF

2.3 构建pyspark提交环境

import json
import flask
from flask import Flask
from concurrent.futures import ThreadPoolExecutorapp = Flask(__name__)
pool = ThreadPoolExecutor(max_workers=8)@app.route('/')
def hello_world():  # put application's code herereturn 'Hello World!'@app.route('/downloadCode', methods=['post'])
def download_file():model_id = flask.request.json.get('modelId')print(model_id)"""异步提交任务：pool.submit()"""return json.dumps(0, ensure_ascii=False)@app.route('/modelRun', methods=['post'])
def model_run():"""异步提交任务：pool.submit()"""return json.dumps(0, ensure_ascii=False)if __name__ == '__main__':app.run()

spark@c67e6477b2f1:/opt/spark$ python3
Python 3.8.10 (default, May 26 2023, 14:05:08) 
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> 
>>>