公司業(yè)務(wù)中有一些實(shí)時(shí)流計(jì)算業(yè)務(wù)需要在線更新配置文件的內(nèi)容，因此需要流式廣播來實(shí)現(xiàn)，測試demo如下：

import org.apache.flink.api.scala._
import com.xuehai.utils.Constants
import org.apache.flink.api.common.state.{BroadcastState, MapStateDescriptor}
import org.apache.flink.api.common.typeinfo.BasicTypeInfo
import org.apache.flink.runtime.state.filesystem.FsStateBackend
import org.apache.flink.streaming.api.CheckpointingMode
import org.apache.flink.streaming.api.functions.co.BroadcastProcessFunction
import org.apache.flink.streaming.api.scala.StreamExecutionEnvironment
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer010
import org.apache.flink.streaming.util.serialization.SimpleStringSchema
import org.apache.flink.util.Collector

object StreamBroadCastDemo extends Constants{
    def main(args: Array[String]) {
        val env = StreamExecutionEnvironment.getExecutionEnvironment

        //基礎(chǔ)設(shè)置
        env.setStateBackend(new FsStateBackend("file:///D:\\checkpoint"))
        env.enableCheckpointing(60000)//開啟checkPoint，并且每分鐘做一次checkPoint保存
        env.getCheckpointConfig.setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE)
        env.getCheckpointConfig.setFailOnCheckpointingErrors(false)//當(dāng)checkpoint出錯后，task是否停止，默認(rèn)為true
        env.setParallelism(1)

        //配置廣播狀態(tài)kafka消費(fèi)實(shí)例
        val configKafkaConsumer = new FlinkKafkaConsumer010[String]("PK-Rank", new SimpleStringSchema(), props)
        configKafkaConsumer.setStartFromLatest()

        //讀取配置文件并生成廣播狀態(tài)
        val mapStateDescriptor = new MapStateDescriptor[String, String]("codeConfig", BasicTypeInfo.STRING_TYPE_INFO, BasicTypeInfo.STRING_TYPE_INFO)
        val init = env.readTextFile("D:\\code.txt")
        val broadStream = env.addSource(configKafkaConsumer).union(init).broadcast(mapStateDescriptor)

        //配置數(shù)據(jù)源kafka消費(fèi)實(shí)例
        val kafkaConsumer = new FlinkKafkaConsumer010[String](topic, new SimpleStringSchema(), props)
        kafkaConsumer.setStartFromLatest()

        //讀取實(shí)時(shí)流數(shù)據(jù)，并結(jié)合配置文件
        val streamData = env.addSource(kafkaConsumer).connect(broadStream).process(new BroadcastProcessFunction[String, String, String] {
            override def processBroadcastElement(value: String, ctx: BroadcastProcessFunction[String, String, String]#Context, out: Collector[String]): Unit = {
                val configMap: BroadcastState[String, String] = ctx.getBroadcastState(mapStateDescriptor)
                configMap.put(value.split(",")(0), value.split(",")(1))
            }

            override def processElement(value: String, ctx: BroadcastProcessFunction[String, String, String]#ReadOnlyContext, out: Collector[String]): Unit = {
                val configMap = ctx.getBroadcastState(mapStateDescriptor)
                val name: String = configMap.get(value)

                //配置文件里面沒有的需要判斷處理一下，否則就會重新加載配置文件
                //只要輸出是null，就會重新加載配置文件，之前讀取的kafka廣播內(nèi)容也會被覆蓋掉
                if(name==null)out.collect(null)
                else out.collect(name)
            }
        }).print()
        env.execute("stream broadCast demo")
    }
}