刚开始配置hadoop难免出错,学习一下启动的脚本对分析错误还是很有帮助的。而且可以顺便学习shell。
我自己对shell命令还算比较熟,shell脚本基本看得懂,不过没具体去深究。所以下面提到的一些shell要点高手莫笑。
Hadoop 0.20.203
hadoop | 主要命令集散地 |
hadoop-config.sh | |
hadoop-daemon.sh | Runs a Hadoop command as a daemon. |
hadoop-daemons.sh | Run a Hadoop command on all slave hosts. |
jsvc | application to launch java daemon |
rcc | The Hadoop record compiler 不懂 |
slaves.sh | Run a shell command on all slave hosts. |
start-all.sh | start-all = start-dfs + start+mapred # Start all hadoop daemons. Run this on master node. |
start-balancer.sh | |
start-dfs.sh | |
start-jobhistoryserver.sh | |
start-mapred.sh | |
stop-all.sh | |
stop-balancer.sh | |
stop-dfs.sh | |
stop-jobhistoryserver.sh | |
stop-mapred.sh |
至于bash的参考资料,首选这里 : http://www.gnu.org/software/bash/manual/bashref.html
我们从最常用的命令开始
start-all.sh
bin=`dirname "$0"`
bin=`cd "$bin"; pwd`
. "$bin"/hadoop-config.sh
# start dfs daemons
"$bin"/start-dfs.sh --config $HADOOP_CONF_DIR
# start mapred daemons
"$bin"/start-mapred.sh --config $HADOOP_CONF_DIR
第一部分是取得bin目录,方便调用其他sh。因为你无法知道是从什么目录运行脚本的,所以无法使用相对路径。
注意hadoop-config.sh 前面有一个点。
参看资料
. (a period)
. filename [arguments]
Read and execute commands from the filename argument in the current shell context.
我们获得一个很好的信息,all = dfs + mapred
之前都把namenode和jobtracker放到一起,datanode和tasktracker放到一起。其实dfs和mapred这两部分还是现对比较独立的。最新的hadoop就把这两部分分开了。
start-dfs.sh / start-mapred.sh
注意daemon一个后面有s,一个没。带s的是启动slaves的。刚开始没看清还蛋疼了许久。
"$bin"/hadoop-daemon.sh --config $HADOOP_CONF_DIR start namenode $nameStartOpt
"$bin"/hadoop-daemons.sh --config $HADOOP_CONF_DIR start datanode $dataStartOpt
"$bin"/hadoop-daemons.sh --config $HADOOP_CONF_DIR --hosts masters start secondarynamenode
这里能解开我之前的一个疑惑:master和slaves的怎样配置master和slaves文件。
对于master:master文件只用来启动secondarynamenode,slaves只用来启动slaves
对于slaves:bin下的脚本没用到这两个文件 但其他部分可能用到
hadoop-daemon.sh
log rotate,看得出日志的流动方向么? 新 1->2->3->4->5 旧hadoop_rotate_log ()
{
log=$1;
num=5;
if [ -n "$2" ]; then
num=$2
fi
if [ -f "$log" ]; then # rotate logs
while [ $num -gt 1 ]; do
prev=`expr $num - 1`
[ -f "$log.$prev" ] && mv "$log.$prev" "$log.$num"
num=$prev
done
mv "$log" "$log.$num";
fi
}
# Determine if we're starting a secure datanode, and if so, redefine appropriate variables
if [ "$command" == "datanode" ] && [ "$EUID" -eq 0 ] && [ -n "$HADOOP_SECURE_DN_USER" ]; then
至于$EUID是系统变量。在hadoop还用到IFS变量。
nohup nice -n $HADOOP_NICENESS "$HADOOP_HOME"/bin/hadoop --config $HADOOP_CONF_DIR $command "$@" > "$log" 2>&1 < /dev/null &
echo $! > $pid
nohup 是个好东西,ssh登录上主机,如果连接断了,当前执行的命令也会挂掉(sighup),有兴趣自己再去google。
nice用来调优先级,没用过。
最终还是通过hadoop来执行程序。
hadoop-daemons.sh
exec "$bin/slaves.sh" --config $HADOOP_CONF_DIR cd "$HADOOP_HOME" \; "$bin/hadoop-daemon.sh" --config $HADOOP_CONF_DIR "$@"
cd "$HADOOP_HOME" \; 似乎没啥用。。。暂时没发现有关联pwd的配置
实际调用了slaves.sh,"$bin/hadoop-daemon.sh" 也传了过去。
这里能解开我之前的一个疑惑:master怎么知道slaves的的hadoop安装在哪。
"$bin/hadoop-daemon.sh" 这坑爹的路径是master的,也就是说,各个机器的hadoop目录要一致。hadoop-daemon.sh 里面可以rsync from $HADOOP_MASTER
slaves.sh
if [ "$HOSTLIST" = "" ]; then
if [ "$HADOOP_SLAVES" = "" ]; then
export HOSTLIST="${HADOOP_CONF_DIR}/slaves"
else
export HOSTLIST="${HADOOP_SLAVES}"
fi
fi
for slave in `cat "$HOSTLIST"|sed "s/#.*$//;/^$/d"`; do
ssh $HADOOP_SSH_OPTS $slave $"${@// /\\ }" \
2>&1 | sed "s/^/$slave: /" &
if [ "$HADOOP_SLAVE_SLEEP" != "" ]; then
sleep $HADOOP_SLAVE_SLEEP
fi
done
一般我们都用slaves文件来配置slaves机器。
sed "s/#.*$//;/^$/d"`;
根据正则表达式的经验,看来是去掉#注释行和空行。
$"${@// /\\ }"
$@ = $1 $2 ...
${@// /\\ }为带有空格的参数加上空格转义。
$"..." 本土化?不知有啥意义,不都是英文么?
资料:
${parameter/pattern/string}
The pattern is expanded to produce a pattern just as in filename expansion. Parameter is expanded and the longest match of
pattern against its value is replaced with string. If pattern begins with ‘/’, all matches of pattern are replaced with string.
Normally only the first match is replaced. If pattern begins with ‘#’, it must match at the beginning of the expanded value of
parameter. If pattern begins with ‘%’, it must match at the end of the expanded value of parameter. If string is null, matches of
pattern are deleted and the / following pattern may be omitted. If parameter is ‘@’ or ‘*’, the substitution operation is
applied to each positional parameter in turn, and the expansion is the resultant list. If parameter is an array variable
subscripted with ‘@’ or ‘*’, the substitution operation is applied to each member of the array in turn, and the expansion is
the resultant list.
Bash supports the $"..." quoting syntax to do locale-specific translation of the characters between the double quotes.
这里能解开我之前的一个疑惑:master怎么去启动slaves的。
原来是通过ssh命令
hadoop-config.sh
# resolve links - $0 may be a softlink
this="$0"
while [ -h "$this" ]; do
ls=`ls -ld "$this"`
link=`expr "$ls" : '.*-> \(.*\)$'`
if expr "$link" : '.*/.*' > /dev/null; then
this="$link"
else
this=`dirname "$this"`/"$link"
fi
done
取个绝对地址也搞得这么蛋疼。取出链接 -> 后面的部分。如果带有/就是绝对路径链接?不对吧??

果然是bug。https://issues.apache.org/jira/browse/HADOOP-7089 我用的是0.20.203版本。可以看看fix好的版本
# Resolve links ($0 may be a softlink) and convert a relative path
# to an absolute path. NB: The -P option requires bash built-ins
# or POSIX:2001 compliant cd and pwd.
this="${BASH_SOURCE-$0}"
common_bin=$(cd -P -- "$(dirname -- "$this")" && pwd -P)
script="$(basename -- "$this")"
this="$common_bin/$script"
奇怪,在gnu的manual没到${BASH_SOURCE-$0} 这种 Shell Parameter Expansion,这里才有 http://pubs.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_06_02 坑爹啊

-- 的作用是表示参数接收已结束 http://pubs.opengroup.org/onlinepubs/009695399/utilities/xcu_chap01.html#tag_01_11
也就是说,比如
grep -- -v file
-v是内容,不是grep的参数了。
man pwd ,找到
-P, --physical
avoid all symlinks
hadoop
exec "$JAVA" -Dproc_$COMMAND $JAVA_HEAP_MAX $HADOOP_OPTS -classpath "$CLASSPATH" $CLASS "$@"
最终启动java class