spark streaming从指定offset处消费Kafka数据

原文地址:http://blog.csdn.net/high2011/article/details/53706446

      首先很感谢原文作者，看到这篇文章我少走了很多弯路，转载此文章是为了保留一份供复习用，请大家支持原作者，移步到上面的连接去看，谢谢

一、情景：当Spark streaming程序意外退出时，数据仍然再往Kafka中推送，然而由于Kafka默认是从latest的offset读取，这会导致数据丢失。为了避免数据丢失，那么我们需要记录每次消费的offset，以便下次检查并且从指定的offset开始读取

二、环境：kafka-0.9.0、spark-1.6.0、jdk-1.7、Scala-2.10.5、idea16

三、实现代码：

      1、引入spark和kafka的相关依赖包

[html]  view plain  copy

1. <?xml version="1.0" encoding="UTF-8"?>  
2. <project xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"  
3.          xmlns="http://maven.apache.org/POM/4.0.0"  
4.          xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">  
5.     <modelVersion>4.0.0</modelVersion>  
6.   
7.     <groupId>com.ngaa</groupId>  
8.     <artifactId>test-my</artifactId>  
9.     <version>1.0-SNAPSHOT</version>  
10.     <inceptionYear>2008</inceptionYear>  
11.     <properties>  
12.         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>  
13.         <project.reporting.outputEncoding>UTF-8</project.reporting.outputEncoding>  
14.         <!--add  maven release-->  
15.         <maven.compiler.source>1.7</maven.compiler.source>  
16.         <maven.compiler.target>1.7</maven.compiler.target>  
17.         <encoding>UTF-8</encoding>  
18.         <!--scala版本-->  
19.         <scala.version>2.10.5</scala.version>  
20.         <!--测试机器上的scala版本-->  
21.         <test.scala.version>2.11.7</test.scala.version>  
22.   
23.         <jackson.version>2.3.0</jackson.version>  
24.         <!--slf4j版本-->  
25.         <slf4j-version>1.7.20</slf4j-version>  
26.         <!--cdh-spark-->  
27.         <spark.cdh.version>1.6.0-cdh5.8.0</spark.cdh.version>  
28.         <spark.streaming.cdh.version>1.6.0-cdh5.8.0</spark.streaming.cdh.version>  
29.         <kafka.spark.cdh.version>1.6.0-cdh5.8.0</kafka.spark.cdh.version>  
30.         <!--cdh-hadoop-->  
31.         <hadoop.cdh.version>2.6.0-cdh5.8.0</hadoop.cdh.version>  
32.         <!--http client必需要兼容CDH中的hadoop版本（cd /opt/cloudera/parcels/CDH/lib/hadoop/lib）-->  
33.         <httpclient.version>4.2.5</httpclient.version>  
34.   
35.         <!--http copre-->  
36.         <httpcore.version>4.2.5</httpcore.version>  
37.         <!--fastjson-->  
38.         <fastjson.version>1.1.39</fastjson.version>  
39.   
40.     </properties>  
41.   
42.     <repositories>  
43.         <repository>  
44.             <id>scala-tools.org</id>  
45.             <name>Scala-Tools Maven2 Repository</name>  
46.             <url>http://scala-tools.org/repo-releases</url>  
47.         </repository>  
48.         <!--配置依赖库地址（用于加载CDH依赖的jar包） -->  
49.         <repository>  
50.             <id>cloudera</id>  
51.             <url>https://repository.cloudera.com/artifactory/cloudera-repos/</url>  
52.         </repository>  
53.     </repositories>  
54.   
55.     <pluginRepositories>  
56.         <pluginRepository>  
57.             <id>scala-tools.org</id>  
58.             <name>Scala-Tools Maven2 Repository</name>  
59.             <url>http://scala-tools.org/repo-releases</url>  
60.         </pluginRepository>  
61.     </pluginRepositories>  
62.   
63.     <dependencies>  
64.   
65.         <!--fastjson-->  
66.         <dependency>  
67.             <groupId>com.alibaba</groupId>  
68.             <artifactId>fastjson</artifactId>  
69.             <version>${fastjson.version}</version>  
70.         </dependency>  
71.         <!--httpclient-->  
72.         <dependency>  
73.             <groupId>org.apache.httpcomponents</groupId>  
74.             <artifactId>httpclient</artifactId>  
75.             <version>${httpclient.version}</version>  
76.         </dependency>  
77.   
78.         <!--http core-->  
79.         <dependency>  
80.             <groupId>org.apache.httpcomponents</groupId>  
81.             <artifactId>httpcore</artifactId>  
82.             <version>${httpcore.version}</version>  
83.         </dependency>  
84.   
85.         <!--slf4j-->  
86.         <dependency>  
87.             <groupId>org.slf4j</groupId>  
88.             <artifactId>slf4j-log4j12</artifactId>  
89.             <version>${slf4j-version}</version>  
90.         </dependency>  
91.         <!--hadoop-->  
92.         <dependency>  
93.             <groupId>org.apache.hadoop</groupId>  
94.             <artifactId>hadoop-client</artifactId>  
95.             <version>${hadoop.cdh.version}</version>  
96.             <exclusions>  
97.                 <exclusion>  
98.                     <groupId>javax.servlet</groupId>  
99.                     <artifactId>*</artifactId>  
100.                 </exclusion>  
101.             </exclusions>  
102.         </dependency>  
103.         <dependency>  
104.             <groupId>org.apache.hadoop</groupId>  
105.             <artifactId>hadoop-common</artifactId>  
106.             <version>${hadoop.cdh.version}</version>  
107.             <exclusions>  
108.                 <exclusion>  
109.                     <groupId>javax.servlet</groupId>  
110.                     <artifactId>*</artifactId>  
111.                 </exclusion>  
112.             </exclusions>  
113.         </dependency>  
114.         <dependency>  
115.             <groupId>org.apache.hadoop</groupId>  
116.             <artifactId>hadoop-hdfs</artifactId>  
117.             <version>${hadoop.cdh.version}</version>  
118.             <exclusions>  
119.                 <exclusion>  
120.                     <groupId>javax.servlet</groupId>  
121.                     <artifactId>*</artifactId>  
122.                 </exclusion>  
123.             </exclusions>  
124.         </dependency>  
125.         <!--spark scala-->  
126.         <dependency>  
127.             <groupId>org.scala-lang</groupId>  
128.             <artifactId>scala-library</artifactId>  
129.             <version>${scala.version}</version>  
130.         </dependency>  
131.         <dependency>  
132.             <groupId>com.fasterxml.jackson.core</groupId>  
133.             <artifactId>jackson-databind</artifactId>  
134.             <version>${jackson.version}</version>  
135.         </dependency>  
136.   
137.         <!--spark streaming和kafka的相关包-->  
138.         <dependency>  
139.             <groupId>org.apache.spark</groupId>  
140.             <artifactId>spark-streaming_2.10</artifactId>  
141.             <version>${spark.streaming.cdh.version}</version>  
142.         </dependency>  
143.         <dependency>  
144.             <groupId>org.apache.spark</groupId>  
145.             <artifactId>spark-streaming-kafka_2.10</artifactId>  
146.             <version>${kafka.spark.cdh.version}</version>  
147.         </dependency>  
148.         <dependency>  
149.             <groupId>junit</groupId>  
150.             <artifactId>junit</artifactId>  
151.             <version>4.12</version>  
152.             <scope>test</scope>  
153.         </dependency>  
154.   
155.         <!--引入windows本地库的spark包-->  
156.         <dependency>  
157.         <groupId>org.apache.spark</groupId>  
158.         <artifactId>spark-assembly_2.10</artifactId>  
159.         <version>${spark.cdh.version}</version>  
160.         <scope>system</scope>  
161.         <systemPath>D:/crt_send_document/spark-assembly-1.6.0-cdh5.8.0-hadoop2.6.0-cdh5.8.0.jar</systemPath>  
162.         </dependency>  
163.   
164.         <!--引入测试环境linux本地库的spark包-->  
165.         <!--<dependency>-->  
166.             <!--<groupId>org.apache.spark</groupId>-->  
167.             <!--<artifactId>spark-assembly_2.10</artifactId>-->  
168.             <!--<version>${spark.cdh.version}</version>-->  
169.             <!--<scope>system</scope>-->  
170.             <!--<systemPath>/opt/cloudera/parcels/CDH/lib/spark/lib/spark-examples-1.6.0-cdh5.8.0-hadoop2.6.0-cdh5.8.0.jar-->  
171.             <!--</systemPath>-->  
172.         <!--</dependency>-->  
173.   
174.         <!--引入中央仓库的spark包-->  
175.         <!--<dependency>-->  
176.         <!--<groupId>org.apache.spark</groupId>-->  
177.         <!--<artifactId>spark-assembly_2.10</artifactId>-->  
178.         <!--<version>${spark.cdh.version}</version>-->  
179.         <!--</dependency>-->  
180.   
181.         <!-- https://mvnrepository.com/artifact/org.apache.hadoop/hadoop-yarn-server-web-proxy -->  
182.         <dependency>  
183.             <groupId>org.apache.hadoop</groupId>  
184.             <artifactId>hadoop-yarn-server-web-proxy</artifactId>  
185.             <version>2.6.0-cdh5.8.0</version>  
186.         </dependency>  
187.   
188.     </dependencies>  
189.   
190.     <!--maven打包-->  
191.     <build>  
192.         <finalName>test-my</finalName>  
193.         <sourceDirectory>src/main/scala</sourceDirectory>  
194.         <testSourceDirectory>src/test/scala</testSourceDirectory>  
195.         <plugins>  
196.             <plugin>  
197.                 <groupId>org.scala-tools</groupId>  
198.                 <artifactId>maven-scala-plugin</artifactId>  
199.                 <version>2.15.2</version>  
200.                 <executions>  
201.                     <execution>  
202.                         <goals>  
203.                             <goal>compile</goal>  
204.                             <goal>testCompile</goal>  
205.                         </goals>  
206.                     </execution>  
207.                 </executions>  
208.                 <configuration>  
209.                     <scalaVersion>${scala.version}</scalaVersion>  
210.                     <args>  
211.                         <arg>-target:jvm-1.7</arg>  
212.                     </args>  
213.                 </configuration>  
214.             </plugin>  
215.             <plugin>  
216.                 <groupId>org.apache.maven.plugins</groupId>  
217.                 <artifactId>maven-eclipse-plugin</artifactId>  
218.                 <configuration>  
219.                     <downloadSources>true</downloadSources>  
220.                     <buildcommands>  
221.                         <buildcommand>ch.epfl.lamp.sdt.core.scalabuilder</buildcommand>  
222.                     </buildcommands>  
223.                     <additionalProjectnatures>  
224.                         <projectnature>ch.epfl.lamp.sdt.core.scalanature</projectnature>  
225.                     </additionalProjectnatures>  
226.                     <classpathContainers>  
227.                         <classpathContainer>org.eclipse.jdt.launching.JRE_CONTAINER</classpathContainer>  
228.                         <classpathContainer>ch.epfl.lamp.sdt.launching.SCALA_CONTAINER</classpathContainer>  
229.                     </classpathContainers>  
230.                 </configuration>  
231.             </plugin>  
232.             <plugin>  
233.                 <artifactId>maven-assembly-plugin</artifactId>  
234.                 <configuration>  
235.                     <descriptorRefs>  
236.                         <descriptorRef>jar-with-dependencies</descriptorRef>  
237.                     </descriptorRefs>  
238.                     <archive>  
239.                         <manifest>  
240.                             <mainClass></mainClass>  
241.                         </manifest>  
242.                     </archive>  
243.                 </configuration>  
244.                 <executions>  
245.                     <execution>  
246.                         <id>make-assembly</id>  
247.                         <phase>package</phase>  
248.                         <goals>  
249.                             <goal>single</goal>  
250.                         </goals>  
251.                     </execution>  
252.                 </executions>  
253.             </plugin>  
254.         </plugins>  
255.     </build>  
256.     <reporting>  
257.         <plugins>  
258.             <plugin>  
259.                 <groupId>org.scala-tools</groupId>  
260.                 <artifactId>maven-scala-plugin</artifactId>  
261.                 <configuration>  
262.                     <scalaVersion>${scala.version}</scalaVersion>  
263.                 </configuration>  
264.             </plugin>  
265.         </plugins>  
266.     </reporting>  
267.   
268. </project>  

 2、新建 测试 类

[java]  view plain  copy

1. import kafka.common.TopicAndPartition  
2. import kafka.message.MessageAndMetadata  
3. import kafka.serializer.StringDecoder  
4. import org.apache.log4j.{Level, Logger}  
5. import org.apache.spark.{SparkConf, TaskContext}  
6. import org.apache.spark.streaming.dstream.InputDStream  
7. import org.apache.spark.streaming.kafka.{HasOffsetRanges, KafkaUtils, OffsetRange}  
8. import org.apache.spark.streaming.{Seconds, StreamingContext}  
9. import org.slf4j.LoggerFactory  
10.   
11. /** 
12.   * Created by yangjf on 2016/12/18 
13.   * Update date: 
14.   * Time: 11:10 
15.   * Describle :从指定偏移量读取kafka数据 
16.   * Result of Test: 
17.   * Command: 
18.   * Email: jifei.yang@ngaa.com.cn 
19.   */  
20. object ReadBySureOffsetTest {  
21.   val logger = LoggerFactory.getLogger(ReadBySureOffsetTest.getClass)  
22.   
23.   def main(args: Array[String]) {  
24.     //设置打印日志级别  
25.     Logger.getLogger("org.apache.kafka").setLevel(Level.ERROR)  
26.     Logger.getLogger("org.apache.zookeeper").setLevel(Level.ERROR)  
27.     Logger.getLogger("org.apache.spark").setLevel(Level.ERROR)  
28.     logger.info("测试从指定offset消费kafka的主程序开始")  
29.     if (args.length < 1) {  
30.       System.err.println("Your arguments were " + args.mkString(","))  
31.       System.exit(1)  
32.       logger.info("主程序意外退出")  
33.     }  
34.     //hdfs://hadoop1:8020/user/root/spark/checkpoint  
35.     val Array(checkpointDirectory) = args  
36.     logger.info("checkpoint检查：" + checkpointDirectory)  
37.     val ssc = StreamingContext.getOrCreate(checkpointDirectory,  
38.       () => {  
39.         createContext(checkpointDirectory)  
40.       })  
41.     logger.info("streaming开始启动")  
42.     ssc.start()  
43.     ssc.awaitTermination()  
44.   }  
45.   
46.   def createContext(checkpointDirectory: String): StreamingContext = {  
47.     //获取配置  
48.     val brokers = "hadoop3:9092,hadoop4:9092"  
49.     val topics = "20161218a"  
50.   
51.     //默认为5秒  
52.     val split_rdd_time = 8  
53.     // 创建上下文  
54.     val sparkConf = new SparkConf()  
55.       .setAppName("SendSampleKafkaDataToApple").setMaster("local[2]")  
56.       .set("spark.app.id", "streaming_kafka")  
57.   
58.     val ssc = new StreamingContext(sparkConf, Seconds(split_rdd_time))  
59.   
60.     ssc.checkpoint(checkpointDirectory)  
61.   
62.     // 创建包含brokers和topic的直接kafka流  
63.     val topicsSet: Set[String] = topics.split(",").toSet  
64.     //kafka配置参数  
65.     val kafkaParams: Map[String, String] = Map[String, String](  
66.       "metadata.broker.list" -> brokers,  
67.       "group.id" -> "apple_sample",  
68.       "serializer.class" -> "kafka.serializer.StringEncoder"  
69. //      "auto.offset.reset" -> "largest"   //自动将偏移重置为最新偏移（默认）  
70. //      "auto.offset.reset" -> "earliest"  //自动将偏移重置为最早的偏移  
71. //      "auto.offset.reset" -> "none"      //如果没有为消费者组找到以前的偏移，则向消费者抛出异常  
72.     )  
73.     /** 
74.       * 从指定位置开始读取kakfa数据 
75.       * 注意：由于Exactly  Once的机制，所以任何情况下，数据只会被消费一次！ 
76.       *      指定了开始的offset后，将会从上一次Streaming程序停止处，开始读取kafka数据 
77.       */  
78.     val offsetList = List((topics, 0, 22753623L),(topics, 1, 327041L))                          //指定topic，partition_no，offset  
79.     val fromOffsets = setFromOffsets(offsetList)     //构建参数  
80.     val messageHandler = (mam: MessageAndMetadata[String, String]) => (mam.topic, mam.message()) //构建MessageAndMetadata  
81.    //使用高级API从指定的offset开始消费，欲了解详情，  
82.    //请进入"http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.streaming.kafka.KafkaUtils$"查看  
83.     val messages: InputDStream[(String, String)] = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder, (String, String)](ssc, kafkaParams, fromOffsets, messageHandler)  
84.   
85.     //数据操作  
86.     messages.foreachRDD(mess => {  
87.       //获取offset集合  
88.       val offsetsList = mess.asInstanceOf[HasOffsetRanges].offsetRanges  
89.       mess.foreachPartition(lines => {  
90.         lines.foreach(line => {  
91.           val o: OffsetRange = offsetsList(TaskContext.get.partitionId)  
92.           logger.info("++++++++++++++++++++++++++++++此处记录offset+++++++++++++++++++++++++++++++++++++++")  
93.           logger.info(s"${o.topic}  ${o.partition}  ${o.fromOffset}  ${o.untilOffset}")  
94.           logger.info("+++++++++++++++++++++++++++++++此处消费数据操作++++++++++++++++++++++++++++++++++++++")  
95.           logger.info("The kafka  line is " + line)  
96.         })  
97.       })  
98.     })  
99.     ssc  
100.   }  
101.   
102.   //构建Map  
103.   def setFromOffsets(list: List[(String, Int, Long)]): Map[TopicAndPartition, Long] = {  
104.     var fromOffsets: Map[TopicAndPartition, Long] = Map()  
105.     for (offset <- list) {  
106.       val tp = TopicAndPartition(offset._1, offset._2)//topic和分区数  
107.       fromOffsets += (tp -> offset._3)           // offset位置  
108.     }  
109.     fromOffsets  
110.   }  
111. }  

四、参考文档：

    1、spark API  http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.streaming.kafka.KafkaUtils$

    2、Kafka官方配置说明：http://kafka.apache.org/documentation.html#configuration

    3、Kafka SampleConsumer：https://cwiki.apache.org/confluence/display/KAFKA/0.8.0+SimpleConsumer+Example

    4、Spark streaming 消费遍历offset说明：http://spark.apache.org/docs/1.6.0/streaming-kafka-integration.html

    5、Kafka官方API说明：http://kafka.apache.org/090/javadoc/index.html?org/apache/kafka/clients/consumer/KafkaConsumer.html

注：以上测试通过，可以根据需要修改。如有疑问，请留言！