Spark Streaming + Flume整合官网文档阅读及运行示例

1,基于Flume的Push模式(Flume-style Push-based Approach)

Flume被用于在Flume agents之间推送数据.在这种方式下,Spark Streaming可以很方便的建立一个receiver,起到一个Avro agent的作用.Flume可以将数据推送到改receiver.

1),需求

从集群中选择一台机器,

当Flume+Spark Streaming程序运行时,需要保证Spark的一个worker运行在同一台机器上.
Flume可以通过配置文件指定推送到该台机器的一个端口.

因为在push模式中,streaming程序在运行时,Flume需要通过配置的端口号监听该机器上的receiver,这样Flume才能进行数据推送

2)配置Flume

配置Flume agent用于发送数据到一个Avro sink,使用如下配置文件:

agent.sinks = avroSink agent.sinks.avroSink.type = avro agent.sinks.avroSink.channel = memoryChannel agent.sinks.avroSink.hostname = <选择机器的ip地址> agent.sinks.avroSink.port = <选择机器的端口号>

可以参考 Flume’s
documentation 获得更多的配置信息
3)配置Spark Streaming程序
a)连接:在SBT/Maven项目中定义,通过以下配置来配置运行环境

groupId = org.apache.spark artifactId = spark-streaming-flume_2.10 version = 1.5.0

b)编程:在程序中import FlumeUtils,创建DStream

import org.apache.spark.streaming.flume._ val flumeStream = FlumeUtils.createStream(streamingContext, [选择的机器ip], [选择的机器端口号])

更多可以参考API
docs 和 example.

需要注意的是:选择机器的ip必须与集群中resource manager中某一台机器一致.这样资源分配(resource allocation)才能匹配到这台机器,并且启动receiver.

c)发布(运行)程序:将spark-streaming-flume_2.10以及相关依赖(除spark-core_2.10和spark-streaming_2.10之外)打包到应用程序的jar包中.然后使用spark-submit来启动应用程序(具体可以参考Deploying
section)

4)示例

(a)Spark程序编写

完整程序如下:

object FlumeLog { def main(args: Array[String]) { StreamingExamples.setStreamingLogLevels() val host = "localhost" val port = 19999 val batchInterval = Milliseconds(2000) // Create the context and set the batch size val sparkConf = new SparkConf().setAppName("FlumeEventCount") val ssc = new StreamingContext(sparkConf, batchInterval) // Create a flume stream val stream = FlumeUtils.createStream(ssc, host, port, StorageLevel.MEMORY_ONLY_SER_2) // Print out the count of events received from this server in each batch stream.count().map(cnt => "Received " + cnt + " flume events." ).print() ssc.start() ssc.awaitTermination() } }

用于统计flume的event事件.
(b)编写flume的配置文件

a1.channels = c1 a1.sinks = k1 a1.sources = r1 a1.sinks.k1.type = avro a1.sinks.k1.channel = c1 a1.sinks.k1.hostname = localhost a1.sinks.k1.port = 19999 a1.sources.r1.type = exec a1.sources.r1.command = tail -F /home/file/bigdatatest/datalake/SougouQ.data a1.sources.r1.bind = localhost a1.sources.r1.port = 44444 a1.sources.r1.channels = c1 a1.channels.c1.type = memory a1.channels.c1.capacity = 1000 a1.channels.c1.transactionCapacity = 100

其中的sinks按照文档中的进行配置.sources用于从日志文件SougouQ.data中读取数据.SougouQ.data中的数据动态生成.
(c)运行
首先启动Spark Streaming程序,可以看到如下输出信息:

启动flume:/usr/local/flume140/jobconf# ../bin/flume-ng agent --conf ../conf/ --conf-file ./spark_avro.conf --name a1 -Dflume.root.logger=INFO,console

(d)结果,此时观察Spark程序的输出结果如下:



可以看到,每隔一定的间隔,就有一个Received 100 flume events的信息输出,表示spark streaming程序已从flume获取到数据.可以将Spark Streaming应用程序按照实际业务需求进行修改.

2,基于Custom Sink的Pull模式(Pull-based
Approach using a Custom Sink)

不同于Flume直接将数据推送到Spark Streaming中,第二种模式通过以下条件运行一个正常的Flume sink:

Flume将数据推送到sink中,并且数据保持buffered状态
Spark Streaming使用一个可靠的Flume接收器(reliable Flume receiver )和转换器(transaction)从sink拉取数据.只要当数据被接收并且被Spark
Streaming备份后,转换器才运行成功.

这样,与第一种模式相比,保证了很好的健壮性和容错能力( fault-tolerance
guarantees ).然而,这种模式需要为Flume配置一个正常的sink.以下为配置步骤

1)基本要求

选择一台运行在一个Flume agent中的普通sink节点的机器.Flume其他的pipeline配置成向该agent发送数据.Spark集群中的机器应该可以访问到选为sink节点的那台机器

2)配置Flume

Flume中sink节点的配置

(a)Sink JARS:将如下JAR包添加到Flume中被选为普通sink节点的classpath中,修改FLUME_HOME/conf/flune_env.sh文件中的FLUME_CLASSPATH=""配置信息,将以下三个jar包路径写入,或者将以下三个jar包复制到FLUME_HOME/lib/目录下

包括三种类型的JAR包

(i) 普通的sink JAR包(下载地址 direct
link):

groupId = org.apache.spark artifactId = spark-streaming-flume-sink_2.10 version = 1.5.0

(ii) Scala运行相关JAR包(下载地址 direct
link).:

groupId = org.scala-lang artifactId = scala-library version = 2.10.4

(iii)Lang 3JAR包(下载地址 direct
link):

groupId = org.apache.commons artifactId = commons-lang3 version = 3.3.2

(b)配置Flume conf文件

使用如下配置参数配置Flume agent用于发送数据到一个Avro sink

agent.sinks = spark agent.sinks.spark.type = org.apache.spark.streaming.flume.sink.SparkSink agent.sinks.spark.hostname = <hostname of the local machine> agent.sinks.spark.port = <port to listen on for connection from Spark> agent.sinks.spark.channel = memoryChannel

同样需要保证Flume upstream pipline已经配置好向运行上述sink的agent发送数据.更多Flume的配置信息可以在Flume’s
documentation 中查看.

3)编写Spark Streaming应用程序

(a)连接

在SBT/Maven项目定义中,需要引入spark-streaming-flume_2.10相关jar包.可以参考之前的配置

(b)编程

导入FLumeUtils创建一个flumeStream

import org.apache.spark.streaming.flume._ val flumeStream = FlumeUtils.createPollingStream(streamingContext, [sink machine hostname], [sink port])

可以查看示例程序FlumePollingEventCount.
注意:每个input Dstream可以接收从多个sinks中的数据

(c)发布(运行)

将spark-streaming-flume_2.10以及相关依赖(除spark-core_2.10和spark-streaming_2.10之外)打包到应用程序的jar包中.然后使用spark-submit来启动应用程序(具体可以参考Deploying
section)

4)示例

(a)Spark程序:

object FlumeLogPull { def main(args: Array[String]) { StreamingExamples.setStreamingLogLevels() val host = "localhost" val port = 19999 val batchInterval = Milliseconds(2000) // Create the context and set the batch size val sparkConf = new SparkConf().setAppName("FlumeEventCount") val ssc = new StreamingContext(sparkConf, batchInterval) // Create a flume stream val stream = FlumeUtils.createPollingStream(ssc, host, port, StorageLevel.MEMORY_ONLY_SER_2) // Print out the count of events received from this server in each batch stream.count().map(cnt => "Received " + cnt + " flume events." ).print() ssc.start() ssc.awaitTermination() } }

可以看到与之前一个相比,只是把val stream = FlumeUtils.createStream(ssc, host, port, StorageLevel.MEMORY_ONLY_SER_2)改成了val stream = FlumeUtils.createPollingStream(ssc,
host, port, StorageLevel.MEMORY_ONLY_SER_2)

(b)配置Flume

a1.channels = c1 a1.sinks = k1 a1.sources = r1 a1.sinks.k1.type = org.apache.spark.streaming.flume.sink.SparkSink a1.sinks.k1.channel = c1 a1.sinks.k1.hostname = localhost a1.sinks.k1.port = 19999 a1.sources.r1.type = exec a1.sources.r1.command = tail -F /home/file/bigdatatest/datalake/SougouQ.data a1.sources.r1.bind = localhost a1.sources.r1.port = 44444 a1.sources.r1.channels = c1 a1.channels.c1.type = memory a1.channels.c1.capacity = 100000 a1.channels.c1.transactionCapacity = 100000

(c)运行

需要先启动flume,再启动spark程序

启动flume:/usr/local/flume140/jobconf# ../bin/flume-ng agent --conf ../conf/ --conf-file ./spark_flume_log_pull.conf --name a1 -Dflume.root.logger=INFO,console

(d)结果



比较启动顺序:

在push模式中,先启动spark application,进入等待状态,等待flume push数据,此时启动flume进行数据的传递.

在pull模式中,spark application会从配置的端口pull数据,此时若flume还未启动,spark application会提示端口连接失败.所以需要先启动flume后启动spark application