zoukankan      html  css  js  c++  java
  • 基于docker构建flink大数据处理平台

    由于公司业务需求,需要搭建一套实时处理数据平台,基于多方面调研选择了Flink.

    • 初始化Swarm环境(也可以选择k8s)

      部署zookeeper集群 基于docker-compose ,使用 docker stack 部署在容器中,由于zookeeper存在数据持久化存储,这块后面可以考虑共享存储方案.

    services:
      zoo1:
        image: zookeeper
        restart: always
        hostname: zoo1
        ports:
          - 2181:2181
        environment:
          ZOO_MY_ID: 1
          ZOO_SERVERS: server.1=0.0.0.0:2888:3888 server.2=zoo2:2888:3888 server.3=zoo3:2888:3888
    
      zoo2:
        image: zookeeper
        restart: always
        hostname: zoo2
        ports:
          - 2182:2181
        environment:
          ZOO_MY_ID: 2
          ZOO_SERVERS: server.1=zoo1:2888:3888 server.2=0.0.0.0:2888:3888 server.3=zoo3:2888:3888
    
      zoo3:
        image: zookeeper
        restart: always
        hostname: zoo3
        ports:
          - 2183:2181
        environment:
          ZOO_MY_ID: 3
          ZOO_SERVERS: server.1=zoo1:2888:3888 server.2=zoo2:2888:3888 server.3=0.0.0.0:2888:3888
    • 部署flink镜像
    version: "3"
    
    services:
      jobmanager:
        image: flink:1.7.2-scala_2.12-alpine
        ports:
          - "8081:8081"
        command: jobmanager
        environment:
          - JOB_MANAGER_RPC_ADDRESS=jobmanager
    
      taskmanager:
        image: flink:1.7.2-scala_2.12-alpine
        command: taskmanager
        environment:
          - JOB_MANAGER_RPC_ADDRESS=jobmanager

    此时只是一个jobmanager 存在单机问题,可以考虑将容器内部的 fluentd.conf 挂载出来,配置zookeeper HA。

    • 对于扩充 TaskManager直接 docker service scala  TaskManager-NAME=3即可

    Flink案例demo,采用读取kafka中数据实时处理,然后将结果存储到influxDb中展示

    // 实时流main
    public
    class SportRealTimeJob { public static void main(String[] args) throws Exception { StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); env.setStreamTimeCharacteristic(TimeCharacteristic.ProcessingTime); KafkaConnector connector = new KafkaConnector("192.168.30.60:9092","big-data"); env .addSource(connector.getConsumerConnector(Lists.newArrayList("test0"))) .<MessageBody>flatMap((sentence,out)->{ MessageBody body=JSON.parseObject(sentence, MessageBody.class); out.collect(body); }) .shuffle() .keyBy(messageBody -> messageBody.getPhone()+messageBody.getUserId()) .timeWindow(Time.seconds(10)) .reduce((t0, t1) -> new MessageBody(t0.getUserId(),t0.getPhone(),t0.getValue()+t1.getValue())) .addSink(new InfluxWriter()) .setParallelism(1); env.execute("Window WordCount"); } }
    // 数据处理实体类demo
    @Data @Measurement(name
    = "sport") public class MessageBody { @Column(name = "userId",tag = true) private String userId; @Column(name = "phone",tag = true) private String phone; @Column(name = "value") private int value; public MessageBody() { } public MessageBody(String userId, String phone, int value) { this.userId = userId; this.phone = phone; this.value = value; } }
    // 自定义数据输出源
    public
    class InfluxWriter extends RichSinkFunction<MessageBody> { private InfluxTemplate template; @Override public void open(Configuration parameters) throws Exception { InfluxBean bean= InfluxBean.builder().dbName("game") .url("http://localhost:8086") .username("admin") .password("admin") .build(); template = new SimpleInfluxTemplate(bean); } @Override public void close() throws Exception { template.close(); } @Override public void invoke(MessageBody value, Context context) throws Exception { template.write(Point.measurement("sport") .addField("value",value.getValue()) .tag("userId",String.valueOf(value.getUserId())) .tag("phone",value.getPhone()) .time(context.currentProcessingTime(), TimeUnit.MILLISECONDS).build()); } }
    // influxDb操作类
    public class SimpleInfluxTemplate implements InfluxTemplate {
    
        private final InfluxDB db;
    
        public SimpleInfluxTemplate(InfluxBean bean){
            this.db= InfluxDBFactory.connect(bean.getUrl(), bean.getUsername(), bean.getPassword());
            db.setDatabase(bean.getDbName());
            db.enableBatch(BatchOptions.DEFAULTS.exceptionHandler(
                    (failedPoints, throwable) -> {
                        /* custom error handling here */ })
                    .consistency(InfluxDB.ConsistencyLevel.ALL)
                    .bufferLimit(100)
            );
        }
    
        @Override
        public void write(Point point) {
            db.write(point);
        }
    
        @Override
        public void bentchWrite(BatchPoints points) {
            db.write(points);
        }
    
        @Override
        public <T> List<T> query(Query query, Class<T> tClass) {
            QueryResult result=db.query(query);
            InfluxDBResultMapper resultMapper = new InfluxDBResultMapper(); // thread-safe - can be reused
            return resultMapper.toPOJO(result, tClass);
        }
    
        @Override
        public void close() {
            db.close();
        }
    
    
    
    public interface InfluxTemplate {
    
        void write(Point point);
    
        void bentchWrite(BatchPoints points);
    
        <T> List<T> query(Query query, Class<T> tClass);
    
        void close();
    }
    
    
    @ToString
    @Getter
    @Setter
    @Builder
    public class InfluxBean {
    
        private String url;
    
        private String username;
    
        private String password;
    
        private String dbName;
    
    
    
    }
  • 相关阅读:
    HADOOP security
    apache sentry
    spark-deployment-modes-cluster-or-client
    Hadoop,Spark,Flink 相关KB
    OSGi类加载流程
    why-the-default-authentication-hadoop-is-unsecured ?
    Spring源码情操陶冶-PathMatchingResourcePatternResolver路径资源匹配溶解器
    maven跳过单元测试-maven.test.skip和skipTests的区别
    Maven+eclipse+jetty配置
    分布式事务资料
  • 原文地址:https://www.cnblogs.com/1ssqq1lxr/p/10417005.html
Copyright © 2011-2022 走看看