a2005147 发表于 2015-9-17 07:22:44

cdh环境下,spark streaming与flume的集成问题总结

  文章发自:http://www.cnblogs.com/hark0623/p/4170156.html转发请注明
  
  如何做集成,其实特别简单,网上其实就是教程。


http://blog.iyunv.com/fighting_one_piece/article/details/40667035看这里就成。 我用的是第一种集成。。



做的时候,出现了各种问题。    大概从从2014.12.17 早晨5点搞到2014.12.17晚上18点30



总结起来其实很简单,但做的时候搞了许久啊啊啊!!!!   这样的事情,吃一堑长一智吧

问题1、需要引用各种包,这些包要打入你的JAR中, 因为用的是spark on yarn模式,所以如果不打进去,在集群中是找不到依赖包的!!!去哪找呢?直接去search.maven.org找。。







问题2:因为搭建的spark on yarn集群,所以监听时只能监听localhost,不然如果你指定了ip,那么非该IP下的结点,就会因为监听不到而出现了问题



问题3:cdh中的flume的启动,你要去find / -name flume.conf ,找一下,然后找到最新的,与cloudera manager配置文件一样的那么,flume启动时就用这个配置文件



问题4:不要直接用集群,先用单点测试一下。。因为单点测试一下后会发现各种问题。 解决后再去集群测试



问题5:一定要注意版本!cdh5.2中spark的版本是1.1.0,而我用的插件一直是1.1.1版本的!!! 啊, 为这事儿,我从中午搞到现在。   这个要吃一堑长一智啦!!!









spark代码如下:




package com.hark
import java.io.File
import org.apache.spark.SparkConf
import org.apache.spark.storage.StorageLevel
import org.apache.spark.streaming.flume.FlumeUtils
import org.apache.spark.streaming.{Seconds, StreamingContext}
import org.apache.spark.streaming.StreamingContext._
/**
* Created by Administrator on 2014-12-16.
*/
object SparkStreamingFlumeTest {
def main(args: Array) {
//println("harkhark")

val path = new File(".").getCanonicalPath()
//File workaround = new File(".");
System.getProperties().put("hadoop.home.dir", path);
new File("./bin").mkdirs();
new File("./bin/winutils.exe").createNewFile();
//val sparkConf = new SparkConf().setAppName("HdfsWordCount").setMaster("local")
val sparkConf = new SparkConf().setAppName("HdfsWordCount")
// Create the context
val ssc = new StreamingContext(sparkConf, Seconds(20))

//val hostname = "127.0.0.1"
val hostname = "localhost"
val port = 2345
val storageLevel = StorageLevel.MEMORY_ONLY
val flumeStream = FlumeUtils.createStream(ssc, hostname, port, storageLevel)
flumeStream.count().map(cnt => "Received " + cnt + " flume events." ).print()

ssc.start()
ssc.awaitTermination()

}
}

  


flume配置文件如下:






# Please paste flume.conf here. Example:
# Sources, channels, and sinks are defined per
# agent name, in this case 'tier1'.
tier1.sources= source1
tier1.channels = channel1
tier1.sinks    = sink1
# For each source, channel, and sink, set
# standard properties.
tier1.sources.source1.type   = exec
tier1.sources.source1.command   = tail -F /opt/data/test3/123
tier1.sources.source1.channels = channel1
tier1.channels.channel1.type   = memory
#tier1.sinks.sink1.type         = logger
tier1.sinks.sink1.type         = avro
tier1.sinks.sink1.hostname      = localhost
tier1.sinks.sink1.port      = 2345
tier1.sinks.sink1.channel      = channel1
# Other properties are specific to each type of yhx.hadoop.dn01
# source, channel, or sink. In this case, we
# specify the capacity of the memory channel.
tier1.channels.channel1.capacity = 100
  






spark启动命令如下:




spark-submit --driver-memory 512m --executor-memory 512m --executor-cores 1--num-executors 3 --class com.hark.SparkStreamingFlumeTest --deploy-mode cluster --master yarn /opt/spark/SparkTest.jar
  






flume启动命令如下:




flume-ng agent --conf /opt/cloudera-manager/run/cloudera-scm-agent/process/585-flume-AGENT --conf-file /opt/cloudera-manager/run/cloudera-scm-agent/process/585-flume-AGENT/flume.conf --name tier1 -Dflume.root.logger=INFO,console
  
页: [1]
查看完整版本: cdh环境下,spark streaming与flume的集成问题总结