Flink 侧输出流拆分流应用

╰+攻爆jí腚メ 2021-09-24 08:12 267阅读 0赞

> 业务场景：使用Flink同步Kafka数据近实时写入MySQL，需要将登录数据拆分为登录，日活，新增分别入三个MySQL表。 采用侧输出流将流拆分为多个流，分别进行处理。

主类

/**
      * Flink 读取 Kafka，每秒聚合一次数据，批量写入 MySQL
      *
      * create by LiuJinHe 2020/5/26
      */
    object CpDataKafkaToMySQL {
      private val logger = LoggerFactory.getLogger(this.getClass)
    
      def main(args: Array[String]): Unit = {
    
        val env = StreamExecutionEnvironment.getExecutionEnvironment
        // 本地测试线程 1
        //    env.setParallelism(1)
    
        // 失败重启,固定间隔,每隔3秒重启1次,总尝试重启10次
        env.setRestartStrategy(RestartStrategies.fixedDelayRestart(10, 3))
    
        // 事件处理的时间，由系统时间决定
        env.setStreamTimeCharacteristic(TimeCharacteristic.ProcessingTime)
    
        // 设置checkpoint周期
        env.enableCheckpointing(60000)
        val config = env.getCheckpointConfig
        // 设置模式 exactly_one
        config.setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE)
        // checkpoint 之间最小间隔
        config.setMinPauseBetweenCheckpoints(1000)
        // checkpoint 超时时间60秒,超时将被丢弃
        config.setCheckpointTimeout(60000)
        // 同一时间只允许进行一次 checkpoint
        config.setMaxConcurrentCheckpoints(1)
    
        val kafkaProp = ConfigManager.load(Constants.CP_KAFKA_PROP)
        val kafkaConsumer = KafkaUtils.getConsumer(kafkaProp).setStartFromGroupOffsets()
    
        val dataKafkaStream = env.addSource(kafkaConsumer).name("cp kafka source")
    
        val dataStream = dataKafkaStream.map(
          json => {
            val sdkData =
              try
                JSON.parseObject(json, classOf[CpData])
              catch {
                case ex: Throwable =>
                  val str =
                    s"""
                       |${DateUtils.timestampToTime(System.currentTimeMillis())} Exception: ${ex.getMessage}
                       |============ illegal json:
                       |$json
                    """.stripMargin
                  logger.info(str)
                  CpData("none", 0, 0, "")
              }
            sdkData
          })
          .filter(line =>
            line.key != "none" && line.cp_game_id != 0 &&
              (line.key == "role_create" || line.key == "cp_role_login" || line.key == "role_pay")
          )
    
        // 定义侧输出流
        val roleActiveSide = new OutputTag[CpData]("role_active")
        val roleNewUserSide = new OutputTag[CpData]("role_new_user")
    
        val mainDataStream = dataStream
          .keyBy(data => data.key)
          .process(new CpProcess(roleActiveSide, roleNewUserSide))
    
        // 侧输出流,日活
        val roleActiveSideStream = mainDataStream.getSideOutput(roleActiveSide)
        roleActiveSideStream
          .timeWindowAll(Time.seconds(1))
          .apply(new CpWindow)
          .addSink(new CpMySQLSink).name("active side stream mysql sink")
    
        // 侧输出流,新增
        val roleNewUserSideStream = mainDataStream.getSideOutput(roleNewUserSide)
        roleNewUserSideStream
          .timeWindowAll(Time.seconds(1))
          .apply(new CpWindow)
          .addSink(new CpMySQLSink).name("new user side stream mysql sink")
    
        // 主输出流,登录
        mainDataStream
          .timeWindowAll(Time.seconds(1))
          .apply(new CpWindow)
          .addSink(new CpMySQLSink).name("cp main stream mysql sink")
    
        env.execute("cp K to M stream job")
      }
    }

Kafka 数据为 json 格式，定义样例类。Flink 读取 Kafka 之前有记录过。

/**
      * 输入类
      */
    case class CpData(key: String, cp_game_id: Int, time: Long, data: String)

因为还有其他事件，所以 keyBy 以后 process 为 KeyedProcessFunction。

/**
      * processFunction 拆分流发送到不同流中
      *
      * create by LiuJinHe 2020/5/26
      */
    class CpProcess(roleActiveSide:OutputTag[CpData], roleNewUserSide:OutputTag[CpData]) extends KeyedProcessFunction[String, CpData, CpData]{
      override def processElement(value: CpData, ctx: KeyedProcessFunction[String, CpData, CpData]#Context, out: Collector[CpData]): Unit = {
        // 全量发送数据到流中
        out.collect(value)
    
        val key = ctx.getCurrentKey
    
        if (key == "cp_role_login") {
          // 发送数据到侧输出流
          ctx.output(roleActiveSide, value.copy("role_active",value.cp_game_id,value.time,value.data))
          ctx.output(roleNewUserSide, value.copy("role_new_user",value.cp_game_id,value.time,value.data))
        }
      }
    }

WindowFunction

/**
      * 自定义 cp window function
      *
      * create by LiuJinHe 2020/5/26
      */
    class CpWindow extends AllWindowFunction[CpData, Iterable[CpData], TimeWindow] {
      override def apply(window: TimeWindow, input: Iterable[CpData], out: Collector[Iterable[CpData]]): Unit = {
        if (input.nonEmpty) {
          println("1 秒数内据条数: " + input.size)
          out.collect(input)
        }
      }
    }

sink 为自定义MySQL Sink，之前有写过，逻辑基本不变。