……

这几天由于项目的需要,需要将Flume收集到的日志插入到Hbase中,有人说,这不很简单么?Flume里面自带了Hbase sink,可以直接调用啊,还用说么?是的,我在本博客的《Flume-1.4.0和Hbase-0.96.0整合》文章中就提到如何用Flume和Hbase整合,从文章中就看出整个过程不太复杂,直接做相应的配置就行了。那么为什么今天还要特意提一下Flume-0.9.4和Hbase-0.96整合?这是因为Flume-0.9.4和Hbase-0.96整合比Flume-1.4.0和Hbase-0.96整合麻烦多了!不是随便几个配置就能搞定的,里面涉及到修改Flume和Hadoop的源码!
  先看下我公司的Hadoop、Hbase、Flume等的配置吧。2013年10月末,公司的Hadoop升级到2.2.0,Hbase升级到0.96,Zookeeper升级到3.4.5,但是Flume版本由于各种原因没有升级,还是用Flume-0.9.4,而Flume-0.9.4源码是基于Hadoop-0.20.2-CDH3B4、Hbase-0.90.1-cdh3u0开发的,Hadoop-0.20.2-CDH3B4和现在的Hadoop-2.2.0完全不一样的设计,而且直接用Hadoop-0.20.2-CDH3B4会使得Flume-0.9.4不能和Hbase-0.96.0通信,都不能通信了,何谈整合!但是经过几天的奋战,我们终于通过修改Flume和Hadoop的部分源码达到了Flume-0.9.4和Hbase-0.96整合,今天就分享一下我们是怎么修改的。

  1、修改Flume-src根目录下的pom.xml文件中的部分依赖版本
  (1)、Hadoop2x里面已经没有hadoop-core jar包,所以修改Hadoop的依赖包的版本:

<dependency>
        <groupId>org.apache.hadoop</groupId>
        <artifactId>hadoop-core</artifactId>
        <version>${cdh.hadoop.version}</version>
</dependency>
 
修改为
 
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-core</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-common</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-common</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-jobclient</artifactId>
            <version>2.2.0</version>
</dependency>

  (2)、修改Guava的版本

<dependency>
        <groupId>com.google.guava</groupId>
        <artifactId>guava</artifactId>
        <version>r07</version>
</dependency>
 
修改为
 
<dependency>
        <groupId>com.google.guava</groupId>
        <artifactId>guava</artifactId>
        <version>10.0.1</version>
</dependency>

  (3)、修改flume-src\flume-core\pom.xml里面的以下配置

<dependency>
      <groupId>org.apache.hadoop</groupId>
      <artifactId>hadoop-core</artifactId>
</dependency>
 
修改为
 
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-core</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-common</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-common</artifactId>
            <version>2.2.0</version>
</dependency>
<dependency>
            <groupId>org.apache.hadoop</groupId>
            <artifactId>hadoop-mapreduce-client-jobclient</artifactId>
            <version>2.2.0</version>
</dependency>

  (4)、修改flume-src\plugins\flume-plugin-hbasesink\pom.xml里面的以下配置

<dependency>
      <groupId>org.apache.hbase</groupId>
      <artifactId>hbase</artifactId>
      <version>${cdh.hbase.version}</version>
</dependency>
 
<dependency>
      <groupId>org.apache.hbase</groupId>
      <artifactId>hbase</artifactId>
      <version>${cdh.hbase.version}</version>
      <classifier>tests</classifier>
      <scope>test</scope>
</dependency>
 
<dependency>
      <groupId>org.apache.hadoop</groupId>
      <artifactId>hadoop-test</artifactId>
      <version>${cdh.hadoop.version}</version>
      <scope>test</scope>
</dependency>
 
修改为
 
<dependency>
          <groupId>org.apache.hbase</groupId>
          <artifactId>hbase-it</artifactId>
          <version>0.96.0-hadoop2</version>
</dependency>

 2,修改flume-core \ src \ main \ java \ org \ apache \ hadoop \ io \ FlushingSequenceFileWriter.java和RawSequenceFileWriter.java两个java类,
  因为步骤一中我们用新版本的Hadoop替换了旧版本的Hadoop,而新版本Hadoop中的org.apache.hadoop.io.SequenceFile.Writer类和旧版本的org.apache.hadoop.io.SequenceFile.Writer类有些不一样。所以导致了FlushingSequenceFileWriter.java和RawSequenceFileWriter.java两个java类出现了部分的错误,解决方法如下:
  (1),需要修改Hadoop-2.2.0源码中的hadoop-2.2.0-src \ hadoop-common-project \ hadoop-common \ src \ main \ java \ org \ apache \ hadoop \ io \ SequenceFile.java类,在Writer类里面添加替换的构造函数:

Writer(){
    this.compress = CompressionType.NONE;
}

然后重新编译hadoop-common-project工程,将编译后的hadoop-common-2.2.0.jar替换为hadoop-common-2.2.0.jar
  (2),修改FlushingSequenceFileWriter.java和RawSequenceFileWriter.java
  这两个类中有错误,请使用新版本的Hadoop的相应API替换掉旧版本的Hadoop的API,具体如何修改,由此不不说了,如有需要的同学,可以邮件联系我(wyphao.2007@163.com )
  (3),修改com.cloudera.flume.handlers.seqfile中的SequenceFileOutputFormat类修改如下:

this(SequenceFile.getCompressionType(FlumeConfiguration.get()),
        new DefaultCodec());
 
修改为
 
this(SequenceFile.getDefaultCompressionType(FlumeConfiguration.get()),
              new DefaultCodec());
 
CompressionType compressionType = SequenceFile.getCompressionType(conf);
 
修改为
 
CompressionType compressionType = SequenceFile.getDefaultCompressionType(conf);

  3,重新编译Flume源码
  重新编译Flume源码(如何编译Flume源码?请参见本博客的《 Flume-0.9.4内核编译及一些编译错误解决方法》),并用编译之后的flume-core-0.9.4- cdh3u3.jar替换$ {FLUME_HOME} / lib中的flume-core-0.9.4-cdh3u3.jar类。删除掉$ {FLUME_HOME} /lib/hadoop-core-0.20.2-cdh3u3.jar等有关Hadoop旧版本的包。
  4,修改$ {} FLUME_HOME /斌/水槽脚本启动
仔细分析$ {} FLUME_HOME /斌/水槽脚本,你会发现如下代码:

# put hadoop conf dir in classpath to include Hadoop
 # core-site.xml/hdfs-site.xml
 if [ -n "${HADOOP_CONF_DIR}" ]; then
     CLASSPATH="${CLASSPATH}:${HADOOP_CONF_DIR}"
 elif [ -n "${HADOOP_HOME}" ] ; then
     CLASSPATH="${CLASSPATH}:${HADOOP_HOME}/conf"
 elif [ -e "/usr/lib/hadoop/conf" ] ; then
     # if neither is present see if the CDH dir exists
     CLASSPATH="${CLASSPATH}:/usr/lib/hadoop/conf";
     HADOOP_HOME="/usr/lib/hadoop"
 fi  # otherwise give up
 
 # try to load the hadoop core jars
 HADOOP_CORE_FOUND=false
 while true; do
     if [ -n "$HADOOP_HOME" ]; then
         HADCOREJARS=`find ${HADOOP_HOME}/hadoop-core*.jar ||  \
               find ${HADOOP_HOME}/lib/hadoop-core*.jar ||  true`
         if [ -n "$HADCOREJARS" ]; then
             HADOOP_CORE_FOUND=true
             CLASSPATH="$CLASSPATH:${HADCOREJARS}"
             break;
         fi
     fi
 
     HADCOREJARS=`find ./lib/hadoop-core*.jar 2> /dev/null || true`
     if [ -n "$HADCOREJARS" ]; then
         # if this is the dev environment then hadoop jar will
         # get added as part of ./lib (below)
         break
     fi
 
     # core jars may be missing, we'll check for this below
     break
 done

  你会发现,这是Flume加载Hadoop旧版本的依赖包,在新版本的Hadoop根本就没有$ {HADOOP_HOME} / conf等文件夹,所以会出现Flume不能加载对新版本Hadoop的依赖。这里教你用最简单的方法来实现对新版本的Hbase和Hadoop的依赖,在$ {FLUME_HOME} / bin / flume脚本里面加入下面的CLASSPATH依赖:

CLASSPATH="/home/q/hbase/hbase-0.96.0-hadoop2/lib/*"
请注意,hbase-0.96.0-hadoop2里面对hadoop的依赖,hbase-0.96.0-hadoop2里面对Hadoop的依赖包是2.1.0,用上面编译好的hadoop-common-2.2.0.jar替换$ { HBASE_HOME} / lib里面的hadoop-common-2.1.0.jar

  5,如何和Hbase-0.96整合
  在flume- src \ plugins \ flume-plugin-hbasesink \ src \ main \ java里面的添加自己的类(当然你完全可以自己创建一个新的maven工程)。如果需要和Hbase整合,必须继承EventSink.Base类,改写里面的方法(可以参照flume-src \ plugins \ flume-plugin-hbasesink \ src \ main \ java \ com \ cloudera \ flume \ hbase \ Attr2 HBase EventSink.java),写完之后需要重新编译flume-src \ plugins \ flume-plugin-hbasesink底下的类,打包成jar文件。然后将您写好的HBase接收器注册到Flume中,关于如何注册,请参见本博客的《 Flume -0.9.4配置Hbase sink》。。
  6,结束
  经过上面几步的配置,你的水槽-0.9.4就可以和Hbase-0.96整合了,祝你成功。

 posted on 2020-06-04 10:08  大码王  阅读(297)  评论(0编辑  收藏  举报
复制代码