SparkStreaming整合Flume

SparkStreaming整合Flume有两种方式，下面会一一列举这两个Demo

案例一、Flume-style Push-based Approach

首先来看一下官方文档，之前所介绍的socket或者fileSystem都属于基本数据源，在这里将主要介绍一下高级数据源。

官网给出三种高级数据源，今天主要来看一下Flume的相关部分。。。

可以将数据放入多个Flume Agent之间，串联或并联放入都可以，SparkStreaming作为一个avro的接收方，接收Flume采集过来的数据。。
配置方法是：

1. Flume和Worker在一台节点上启动
2. Flume配置之后将数据发送给一个端口之中
   此外SparkStreaming是接收数据的，因此要先启动并且监听一个Flume注入数据的端口。。

1. 先配置一下Flume

# Name the components on this agent
simple-agent.sources = netcat-source
simple-agent.sinks = avro-sink
simple-agent.channels = memory-channel

# Describe/configure the source
simple-agent.sources.netcat-source.type = netcat
simple-agent.sources.netcat-source.bind = hadoop1
simple-agent.sources.netcat-source.port = 44444

# Describe the sink
simple-agent.sinks.avro-sink.type = avro
simple-agent.sinks.avro-sink.hostname = hadoop1
simple-agent.sinks.avro-sink.port = 41414 

# Use a channel which buffers events in memory
simple-agent.channels.memory-channel.type = memory
simple-agent.channels.memory-channel.capacity = 1000
simple-agent.channels.memory-channel.transactionCapacity = 100

# Bind the source and sink to the channel
simple-agent.sources.netcat-source.channels = memory-channel
simple-agent.sinks.avro-sink.channel = memory-channel

2.写SparkStreaming应用程序，导入FlumeUtils创建DStream

首先导入相关依赖：

		<dependency>
            <groupId>org.apache.sparkgroupId>
            <artifactId>spark-streaming-flume_2.11artifactId>
            <version>2.1.1version>
        dependency>

然后写一个Push方式的wordcount demo

3.测试

本地测试中，需要将Flume的配置中Sink配置改成主机ip地址，不是服务器地址，然后启动SparkStreaming，之后启动Flume，使用talent输入数据，观察控制台的输出
simple-agent.conf

# Name the components on this agent
simple-agent.sources = netcat-source
simple-agent.sinks = avro-sink
simple-agent.channels = memory-channel

# Describe/configure the source
simple-agent.sources.netcat-source.type = netcat
simple-agent.sources.netcat-source.bind = hadoop1
simple-agent.sources.netcat-source.port = 44444

# Describe the sink
simple-agent.sinks.avro-sink.type = avro
simple-agent.sinks.avro-sink.hostname = 192.168.1.161
simple-agent.sinks.avro-sink.port = 41414

# Use a channel which buffers events in memory
simple-agent.channels.memory-channel.type = memory
simple-agent.channels.memory-channel.capacity = 1000
simple-agent.channels.memory-channel.transactionCapacity = 100

# Bind the source and sink to the channel
simple-agent.sources.netcat-source.channels = memory-channel
simple-agent.sinks.avro-sink.channel = memory-channel

启动Flume

flume-ng agent 
--name simple-agent 
--conf /home/hadoop1/modules/apache-flume-1.7.0-bin/conf/ 
--conf-file /home/hadoop1/modules/apache-flume-1.7.0-bin/conf/flume_push_streaming.conf  
-Dflume.root.logger=INFO,console

使用telnet，要先开放端口，然后再启动telnet-server才能连接上

4. spark-submit上线部署

测试成功之后进行线上部署，先把Flume的配置文件改成之前的hadoop1的hostname，然后用mvn clean package -DskipTests将SparkStreaming打成jar包，然后启动spark-submit

[1@hadoop1 spark-2.1.1-bin-hadoop2.7]$ spark-submit 
--name spark_flume 
--class com.fyj.spark.spark_flume 
--master local[*] 
--packages org.apache.spark:spark-streaming-flume_2.11:2.1.1 /home/hadoop1/modules/apache-flume-1.7.0-bin/test_dataSource/flume_spark/target/flume_spark-1.0-SNAPSHOT.jar 
hadoop1 41414

案例二、Push-based Approach using a Custom Sink

指SparkStreaming拉取过来信息，只需要让Flume将数据push到一个buffer区，SparkStreaming会使用一个合适的Flume Receiver，从sink内拉出来，并且这个操作只会在数据被SparkStreaming完成副本和接收成功之后才会完成。。
因此这种方式比第一种方式要安全可靠，支持容错很高，所以需要配置Flume到一个自定义的Sink上面。。

需求：使用一台机器运行flume agent ，然后用SparkStreaming去方位这台正在工作的自定义sink就ok了。

1. 首先需要配置sink的jar包到SparkStreaming的pom文件上

		<dependency>
            <groupId>org.apache.commonsgroupId>
            <artifactId>commons-lang3artifactId>
        dependency>
        <dependency>
            <groupId>org.apache.sparkgroupId>
            <artifactId>spark-streaming-flume-sink_2.11artifactId>
            <version>2.1.1version>
        dependency>

2. 配置Flume Agent Conf

# Name the components on this agent
simple-agent.sources = netcat-source
simple-agent.sinks = spark-sink
simple-agent.channels = memory-channel

# Describe/configure the source
simple-agent.sources.netcat-source.type = netcat
simple-agent.sources.netcat-source.bind = hadoop1
simple-agent.sources.netcat-source.port = 44444

# Describe the sink
simple-agent.sinks.spark-sink.type = org.apache.spark.streaming.flume.sink.SparkSink
simple-agent.sinks.spark-sink.hostname = hadoop1
simple-agent.sinks.spark-sink.port = 41414

# Use a channel which buffers events in memory
simple-agent.channels.memory-channel.type = memory
simple-agent.channels.memory-channel.capacity = 1000
simple-agent.channels.memory-channel.transactionCapacity = 100

# Bind the source and sink to the channel
simple-agent.sources.netcat-source.channels = memory-channel
simple-agent.sinks.spark-sink.channel = memory-channel

3. 具体代码如下：