本文介绍如何在 CentOS7 中以伪分布式模式 ( Pseudo-Distributed Mode ) 运行 Hadoop-1.2.1。
前期准备
安装 Linux 系统
建议在虚拟机下安装 Linux 系统,并善用虚拟机的 snapshot 功能,可以节省很多时间。
安装 JDK1.8 并设置环境变量
Hadoop 的安装及配置
Hadoop 版本有很多,建议安装 hadoop.1.2.1 或 hadoop 的最新版本。本文以 hadoop-1.2.1 为例。另外,Hadoop Wiki 上有更系统更权威的资料。
step1:下载安装 Hadoop
# 切换到 root 身份
[zodas@localhost ~]$ su
# 将下载到桌面上的包 hadoop-1.2.1.tar.gz 拷贝到 /opt/installpackages 目录下
[root@localhost zodas]# cp ./Desktop/hadoop-1.2.1.tar.gz /opt/installpackages
# 切换工作目录到 /opt
[root@localhost zodas]# cd /opt
# 将 hadoop-1.2.1.tar.gz 解压到 /opt 目录下
[root@localhost opt]# tar -xvzf /opt/installpackages/hadoop-1.2.1.tar.gz
step2:修改环境变量 /etc/profile
打开终端输入vi /etc/profile,增加以下内容,配置 Hadoop 环境变量:# Hadoop
export HADOOP_HOME=/opt/hadoop-1.2.1
export PATH=$PATH:$HADOOP_HOME/bin:$HADOOP_HOME/sbin
打开终端输入source /etc/profile,使新添加的环境变量立即生效
打开终端输入hadoop version,验证环境变量是否配置成功
注:由于是在 /etc/profile 中配置的环境变量,所以建议配置完后重启一次系统,让环境变量永久生效
step3:配置hadoop
Hadoop-1.2.1 的配置文件默认是放在其 conf/ 目录下,我们只需要配置其中的4个文件即可,分别是:hadoop-env.sh, core-site.xml, hdfs-site.xml, mapred-site.xml。
hadoop-env.sh
在该文件中我们只需要修改 JAVA_HOME 这一变量的值为 JDK 安装的真实路径即可,如果按照笔者的教程操作的话,JAVA_HOME=/opt/jdk1.8.0,修改内容如下:
Change hadoop-env.sh
# The java implementation to use. Required.
# export JAVA_HOME=/usr/lib/j2sdk1.5-sun
to
# The java implementation to use. Required.
export JAVA_HOME=/opt/jdk1.8.0_131
conf/*-site.xml
conf/core-site.xom 修改为:
hadoop.tmp.dir
/usr/workspace/hadoop/hdfs/tmp
fs.default.name
hdfs://localhost:9000
conf/mapred-site.xml 修改为:
mapred.job.tracker
localhost:9001
conf/hdfs-site.xml 修改为:
dfs.replication
1
dfs.data.dir
/usr/workspace/hadoop/hdfs/data
dfs.name.dir
/usr/workspace/hadoop/hdfs/name
step4:修改 Hadoop 工作目录的 owner 和 group
建议创建一个新的群组,并将用户添加到该新群组中,否则,在 hadoop namenode -format 可能会报 Cannot create directory /usr/workspace/hadoop/hdfs/name/current 的错误,这是因为新创建的 workspace 目录其owner=root, group=root,所以 Hadoop 在 -format 无法自动创建需要的目录,解决办法是修改 workspace/hadoop/ 及其子目录的 owner 和 group。
创建一个新的群组hadoop(名字随意)# 创建一个新的群组 hadoop
[zodas@localhost ~]$ sudo groupadd hadoop
# 将用户 zodas 添加到新的群组中
[zodas@localhost ~]$ sudo usermod -G hadoop zodas
修改工作目录的权限[zodas@localhost ~]$ cd /usr
[zodas@localhost usr]$ sudo mkdir -p /usr/workspace/hadoop/hdfs
# 修改 workspace 目录及其子目录的 owner=zodas, group=hadoop
[zodas@localhost usr]$ sudo chown -R zodas:hadoop workspace/
# 修改 workspace 目录及其子目录的 权限为 rwxrwxr-x(即775)
[zodas@localhost usr]$ sudo chmod -R g+w workspace/
同上,修改 /opt/hadoop-1.2.1 的拥有者及群组为 zodas:hadoop[zodas@localhost ~]$ sudo chown -R zodas:hadoop /opt/hadoop-1.2.1
[zodas@localhost ~]$ sudo chmod -R g+x /opt/hadoop-1.2.1
step5:配置 SSH 免密码登录
不了解 SSH 的,可参考阮一峰的这篇博文"SSH原理与运用(一):远程登录"
CentOS 默认是已经安装了 OpenSSH的,可在终端中输入ssh -V,查看已安装的OpenSSH版本
使用 ssh-keygen 生成公私钥:# 生成公私钥。会在 ~/.ssh 目录下生成两个文件:id_rsa.pub 和 id_rsa
[zodas@localhost ~]$ ssh-keygen -t rsa -P ""
tips:-P 表示密码,-P "" 就表示空密码,也可以不用 -P 参数,这样就要三次回车,用 -P 就一次回车。
再输入下面的命令,将公钥传送到远程主机host上面:[zodas@localhost ~]$ ssh-copy-id zodas@localhost
好了,从此你再登录,就不需要输入密码了。
如果还是不行,就打开 /etc/ssh/sshd_config 这个文件,检查下面几行前面"#"注释是否取掉RSAAuthentication yes
PubkeyAuthentication yes
AuthorizedKeysFile .ssh/authorized_keys
重启远程主机的ssh服务。# ubuntu系统
service ssh restart
# debian系统
/etc/init.d/ssh restart
打开终端输入ssh zodas@localhost,若能不输入密码即可登录主机 localhost,则表示 SSH 免密码登录配置成功
另外,
Ubuntu 默认是没有安装 openssh-server,这也是为什么在/etc/ssh 目录下找不到 sshd_config 文件。解决方案执行sudo apt-get install openssh-server 安装 openssh-server
启动 Hadoop
Formatting the HDFS filesystem via the namenode
运行:
[zodas@localhost ~]hadoop namenode -format
输出类似于以下:
[zodas@localhost ~]$ hadoop namenode -format
Warning: $HADOOP_HOME is deprecated.
17/07/10 11:13:31 INFO namenode.NameNode: STARTUP_MSG:
/************************************************************
STARTUP_MSG: Starting NameNode
STARTUP_MSG: host = ./192.168.12.184
STARTUP_MSG: args = [-format]
STARTUP_MSG: version = 1.2.1
STARTUP_MSG: build = https://svn.apache.org/repos/asf/hadoop/common/branches/branch-1.2 -r 1503152; compiled by 'mattf' on Mon Jul 22 15:23:09 PDT 2013
STARTUP_MSG: java = 1.8.0_131
************************************************************/
17/07/10 11:13:31 INFO util.GSet: Computing capacity for map BlocksMap
17/07/10 11:13:31 INFO util.GSet: VM type = 64-bit
17/07/10 11:13:31 INFO util.GSet: 2.0% max memory = 1013645312
17/07/10 11:13:31 INFO util.GSet: capacity = 2^21 = 2097152 entries
17/07/10 11:13:31 INFO util.GSet: recommended=2097152, actual=2097152
17/07/10 11:13:31 INFO namenode.FSNamesystem: fsOwner=zodas
17/07/10 11:13:31 INFO namenode.FSNamesystem: supergroup=supergroup
17/07/10 11:13:31 INFO namenode.FSNamesystem: isPermissionEnabled=true
17/07/10 11:13:31 INFO namenode.FSNamesystem: dfs.block.invalidate.limit=100
17/07/10 11:13:31 INFO namenode.FSNamesystem: isAccessTokenEnabled=false accessKeyUpdateInterval=0 min(s), accessTokenLifetime=0 min(s)
17/07/10 11:13:31 INFO namenode.FSEditLog: dfs.namenode.edits.toleration.length = 0
17/07/10 11:13:31 INFO namenode.NameNode: Caching file names occuring more than 10 times
17/07/10 11:13:32 ERROR namenode.NameNode: java.io.IOException: Cannot create directory /usr/workspace/hadoop/hdfs/name/current
at org.apache.hadoop.hdfs.server.common.Storage$StorageDirectory.clearDirectory(Storage.java:294)
at org.apache.hadoop.hdfs.server.namenode.FSImage.format(FSImage.java:1337)
at org.apache.hadoop.hdfs.server.namenode.FSImage.format(FSImage.java:1356)
at org.apache.hadoop.hdfs.server.namenode.NameNode.format(NameNode.java:1261)
at org.apache.hadoop.hdfs.server.namenode.NameNode.createNameNode(NameNode.java:1467)
at org.apache.hadoop.hdfs.server.namenode.NameNode.main(NameNode.java:1488)
17/07/10 11:13:32 INFO namenode.NameNode: SHUTDOWN_MSG:
/************************************************************
SHUTDOWN_MSG: Shutting down NameNode at ./192.168.12.184
************************************************************/
Starting single-node cluster
运行:
[zodas@localhost ~]$ start-all.sh
输出:
[zodas@ ~]$ start-all.sh
Warning: $HADOOP_HOME is deprecated.
starting namenode, logging to /opt/hadoop-1.2.1/libexec/../logs/hadoop-zodas-namenode-..out
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: starting datanode, logging to /opt/hadoop-1.2.1/libexec/../logs/hadoop-zodas-datanode-..out
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: starting secondarynamenode, logging to /opt/hadoop-1.2.1/libexec/../logs/hadoop-zodas-secondarynamenode-..out
starting jobtracker, logging to /opt/hadoop-1.2.1/libexec/../logs/hadoop-zodas-jobtracker-..out
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: starting tasktracker, logging to /opt/hadoop-1.2.1/libexec/../logs/hadoop-zodas-tasktracker-..out
用 jps 命令查看 java 进程,验证伪分布式集群是否安装成功
运行:
[zodas@localhost ~]$ jps
输出:
[zodas@localhost ~]$ jps
7431 Jps
7192 TaskTracker
6858 DataNode
6989 SecondaryNameNode
7069 JobTracker
6735 NameNode
Stoping single-node cluster
运行:
[zodas@localhost ~]$ stop-all.sh
输出:
[zodas@localhost ~]$ stop-all.sh
Warning: $HADOOP_HOME is deprecated.
stopping jobtracker
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: stopping tasktracker
stopping namenode
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: stopping datanode
localhost: Warning: $HADOOP_HOME is deprecated.
localhost:
localhost: stopping secondarynamenode
参考