◢ <實作九> | <回課程大綱> ▲ | <實作十一> ◣

請連線至 hadoop.3du.me 作練習

實作十 Lab10

HDFS、MapReduce 與 Hadoop Streaming 觀念驗證
Running Hadoop Streaming with HDFS and MapReduce

準備輸入資料集 Input Dataset

• 首先準備輸入，包含兩部份：(1) 供 Velvet 計算的 *.fa 檔案，這裡為了方便示範起見，採用 test_long.fa 當範本，並複製 99 份不同檔名，當輸入 (2) 供 Mapper 運算的輸入檔案（內含 HDFS 的檔名路徑）

  ~$ cp /usr/share/doc/velvet-example/examples/data/test_long.fa.gz .
  ~$ gunzip test_long.fa.gz
  ~$ for ((i=1;i<100;i++)); do hadoop fs -put test_long.fa input-$i.fa; done
  ~$ for ((i=1;i<20;i++)); do echo /user/$(whoami)/input-$i.fa; done > split-01.txt
  ~$ for ((i=20;i<40;i++)); do echo /user/$(whoami)/input-$i.fa; done > split-02.txt
  ~$ for ((i=40;i<60;i++)); do echo /user/$(whoami)/input-$i.fa; done > split-03.txt
  ~$ for ((i=60;i<80;i++)); do echo /user/$(whoami)/input-$i.fa; done > split-04.txt
  ~$ for ((i=80;i<100;i++)); do echo /user/$(whoami)/input-$i.fa; done > split-05.txt
  ~$ hadoop fs -mkdir lab10_input
  ~$ hadoop fs -put split-0* lab10_input
  

• 檢查輸入檔案

  ~$ hadoop fs -ls
  ~$ hadoop fs -ls lab10_input
  

觀察 Hadoop Streaming 執行身份與工作目錄

• 撰寫 testmapper.sh

  #!/bin/bash
  
  id="h998"
  mkdir -p /tmp/$id
  host=`hostname`
  pwd=`pwd`
  uid=`whoami`
  
  while read line; do 
    input=$line
    filename=`basename $input`
    echo "$uid@$host:$pwd> hadoop fs -get $input /tmp/$id/$filename"
    echo "$uid@$host:$pwd> velveth output-$filename 17 -fasta -short /tmp/$id/$filename"
    echo "$uid@$host:$pwd> hadoop fs -put output-$filename ."
  done
  rm -rf /tmp/$id
  

• 接著，讓我們在本地端先驗證一下 testmapper.sh 的運作

  ~$ head -n 2 split-01.txt > split-00.txt
  ~$ cat > testmapper.sh << EOF
  #!/bin/bash
  
  id="`whoami`"
  mkdir -p /tmp/\$id
  host=\`hostname\`
  pwd=\`pwd\`
  uid=\`whoami\`
  
  while read line; do 
    input=\$line
    filename=\`basename \$input\`
    echo "\$uid@\$host:\$pwd> hadoop fs -get \$input /tmp/\$id/\$filename"
    echo "\$uid@\$host:\$pwd> velveth output-\$filename 17 -fasta -short /tmp/\$id/\$filename"
    echo "\$uid@\$host:\$pwd> hadoop fs -put output-\$filename ."
  done
  rm -rf /tmp/\$id
  EOF
  ~$ chmod a+x testmapper.sh
  ~$ cat sample-00.txt | ./testmapper.sh 
  h998@hadoop:/home/h998> hadoop fs -get /user/h998/input-1.fa /tmp/h998/input-1.fa
  h998@hadoop:/home/h998> velveth output-input-1.fa 17 -fasta -short /tmp/h998/input-1.fa
  h998@hadoop:/home/h998> hadoop fs -put output-input-1.fa .
  h998@hadoop:/home/h998> hadoop fs -get /user/h998/input-2.fa /tmp/h998/input-2.fa
  

• 讓我們用 Hadoop Streaming 的方式來執行 testmapper.sh

  ~$ hadoop jar hadoop-streaming.jar -input lab10_input -output lab10_out1 -mapper testmapper.sh -file testmapper.sh 
  

• 觀察 lab10_out1 的結果，看與本機執行有何不同呢？

  ~$ hadoop fs -cat /user/$(whoami)/lab10_out1/part-00000 | head
  hadoop@hadoop104:/var/lib/hadoop/cache/hadoop/mapred/local/taskTracker/jobcache/job_201106041247_1820/attempt_201106041247_1820_m_000002_0/work> hadoop fs -get /user/h998/input-60.fa /tmp/h998/input-60.fa	
  hadoop@hadoop104:/var/lib/hadoop/cache/hadoop/mapred/local/taskTracker/jobcache/job_201106041247_1820/attempt_201106041247_1820_m_000002_0/work> hadoop fs -get /user/h998/input-61.fa /tmp/h998/input-61.fa	
  hadoop@hadoop104:/var/lib/hadoop/cache/hadoop/mapred/local/taskTracker/jobcache/job_201106041247_1820/attempt_201106041247_1820_m_000002_0/work> hadoop fs -get /user/h998/input-62.fa /tmp/h998/input-62.fa	
  

實作透過 Hadoop Streaming 執行 99 組 velvet 運算

• 撰寫 mapper.sh

  #!/bin/bash
  
  id="h998"
  mkdir -p /tmp/$id
  host=`hostname`
  pwd=`pwd`
  uid=`whoami`
  
  while read line; do 
    input=$line
    filename=`basename $input`
    echo "$uid@$host> hadoop fs -get $input /tmp/$id/$filename"
    hadoop fs -get $input /tmp/$id/$filename
    echo "$uid@$host> velveth output-$filename 17 -fasta -short /tmp/$id/$filename"
    velveth output-$filename 17 -fasta -short /tmp/$id/$filename
    echo "$uid@$host> hadoop fs -put output-$filename /user/$id/."
    hadoop fs -put output-$filename /user/$id/.
    hadoop fs -chown $id /user/$id/output-$filename
  done
  rm -rf /tmp/$id
  

• 於本機測試 mapper.sh

  ~$ cat > mapper.sh << EOF
  #!/bin/bash
  
  id="`whoami`"
  mkdir -p /tmp/\$id
  host=\`hostname\`
  pwd=\`pwd\`
  uid=\`whoami\`
  
  while read line; do 
    input=\$line
    filename=\`basename \$input\`
    echo "\$uid@\$host> hadoop fs -get \$input /tmp/\$id/\$filename"
    hadoop fs -get \$input /tmp/\$id/\$filename
    echo "\$uid@\$host> velveth output-\$filename 17 -fasta -short /tmp/\$id/\$filename"
    velveth output-\$filename 17 -fasta -short /tmp/\$id/\$filename
    echo "\$uid@\$host> hadoop fs -put output-\$filename /user/\$id/."
    hadoop fs -put output-\$filename /user/\$id/.
    hadoop fs -chown \$id /user/\$id/output-\$filename
  done
  rm -rf /tmp/\$id
  EOF
  ~$ chmod a+x mapper.sh
  ~$ cat sample-00.txt | ./mapper.sh 
  ~$ hadoop fs -rmr output-*
  ~$ rm -rf output-sample-*
  

• 接著用 hadoop streaming 來執行

  ~$ hadoop jar hadoop-streaming.jar -input lab10_input -output lab10_out2 -mapper mapper.sh -file mapper.sh