224356461.jpg





流量比较大的日志要是直接写入Hadoop对Namenode负载过大,所以入库前合并,可以把各个节点的日志凑并成一个文件写入HDFS。 根据情况定期合成,写入到hdfs里面。

咱们看看日志的大小,200G的dns日志文件,我压缩到了18G,要是用awk perl当然也可以,但是处理速度肯定没有分布式那样的给力。

230102727.jpg


Hadoop Streaming原理

mapper和reducer会从标准输入中读取用户数据,一行一行处理后发送给标准输出。Streaming工具会创建MapReduce作业,发送给各个tasktracker,同时监控整个作业的执行过程。


任何语言,只要是方便接收标准输入输出就可以做mapreduce~

再搞之前我们先简单测试下shell模拟mapreduce的性能速度~

234955396.jpg


看下他的结果,350M的文件用时35秒左右。

235045406.jpg


这是2G的日志文件,居然用了3分钟。 当然和我写的脚本也有问题,我们是模拟mapreduce的方式,而不是调用shell下牛逼的awk,gawk处理。


001056805.jpg

awk的速度 !果然很霸道,处理日志的时候,我也很喜欢用awk,只是学习的难度有点大,不像别的shell组件那么灵活简单。

000946258.jpg


这是官方的提供的两个demo ~

map.py

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
#!/usr/bin/env python
"" "A more advanced Mapper, using Python iterators and generators." ""
import  sys
def read_input(file):
     for  line  in  file:
         # split the line into words
         yield line.split()
def main(separator= '\t' ):
     # input comes from STDIN (standard input)
     data = read_input(sys.stdin)
     for  words  in  data:
         # write the results to STDOUT (standard output);
         # what we output here will be the input  for  the
         # Reduce step, i.e. the input  for  reducer.py
         #
         # tab-delimited; the trivial word count  is  1
         for  word  in  words:
             print  '%s%s%d'  % (word, separator,  1 )
if  __name__ ==  "__main__" :
     main()


reduce.py的修改方式

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
#!/usr/bin/env python
"" "A more advanced Reducer, using Python iterators and generators." ""
from itertools  import  groupby
from operator  import  itemgetter
import  sys
def read_mapper_output(file, separator= '\t' ):
     for  line  in  file:
         yield line.rstrip().split(separator,  1 )
def main(separator= '\t' ):
     # input comes from STDIN (standard input)
     data = read_mapper_output(sys.stdin, separator=separator)
     # groupby groups multiple word-count pairs by word,
     # and creates an iterator that returns consecutive keys and their group:
     #   current_word - string containing a word (the key)
     #   group - iterator yielding all [ "<current_word>" "<count>" ] items
     for  current_word, group  in  groupby(data, itemgetter( 0 )):
         try :
             total_count = sum( int (count)  for  current_word, count  in  group)
             print  "%s%s%d"  % (current_word, separator, total_count)
         except ValueError:
             # count was not a number, so silently discard  this  item
             pass
if  __name__ ==  "__main__" :
     main()


咱们再简单点:

1
2
3
4
5
6
7
#!/usr/bin/env python
import  sys
for  line  in  sys.stdin:
     line = line.strip()
     words = line.split()
     for  word  in  words:
         print  '%s\t%s'  % (word,  1 )


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
#!/usr/bin/env python
                                                                                                                                                                                                                                                                                                                                                               
from operator  import  itemgetter
import  sys
                                                                                                                                                                                                                                                                                                                                                               
current_word = None
current_count =  0
word = None
                                                                                                                                                                                                                                                                                                                                                               
for  line  in  sys.stdin:
     line = line.strip()
     word, count = line.split( '\t' 1 )
     try :
         count =  int (count)
     except ValueError:
         continue
     if  current_word == word:
         current_count += count
     else :
         if  current_word:
             print  '%s\t%s'  % (current_word, current_count)
         current_count = count
         current_word = word
                                                                                                                                                                                                                                                                                                                                                              
if  current_word == word:
     print  '%s\t%s'  % (current_word, current_count)


咱们就简单模拟下数据,跑个测试


084336884.jpg



剩下就没啥了,在hadoop集群环境下,运行hadoop的steaming.jar组件,加入mapreduce的脚本,指定输出就行了.  下面的例子我用的是shell的成分。


1
2
3
4
5
[root@ 101  cron]#$HADOOP_HOME/bin/hadoop  jar $HADOOP_HOME/contrib/streaming/hadoop-*-streaming.jar \
-input myInputDirs \
-output myOutputDir \
-mapper cat \
-reducer wc


详细的参数,对于咱们来说提供性能可以把tasks的任务数增加下,根据情况自己测试下,也别太高了,增加负担。

(1)-input:输入文件路径

(2)-output:输出文件路径

(3)-mapper:用户自己写的mapper程序,可以是可执行文件或者脚本

(4)-reducer:用户自己写的reducer程序,可以是可执行文件或者脚本

(5)-file:打包文件到提交的作业中,可以是mapper或者reducer要用的输入文件,如配置文件,字典等。

(6)-partitioner:用户自定义的partitioner程序

(7)-combiner:用户自定义的combiner程序(必须用java实现)

(8)-D:作业的一些属性(以前用的是-jonconf),具体有:
           1)mapred.map.tasks:map task数目
           2)mapred.reduce.tasks:reduce task数目
           3)stream.map.input.field.separator/stream.map.output.field.separator: map task输入/输出数
据的分隔符,默认均为\t。
            4)stream.num.map.output.key.fields:指定map task输出记录中key所占的域数目
            5)stream.reduce.input.field.separator/stream.reduce.output.field.separator:reduce task输入/输出数据的分隔符,默认均为\t。
            6)stream.num.reduce.output.key.fields:指定reduce task输出记录中key所占的域数目


这里是统计dns的日志文件有多少行 ~

114512821.jpg

在mapreduce作为参数的时候,不能用太多太复杂的shell语言,他不懂的~

可以写成shell文件的模式;

1
2
3
4
5
6
7
8
#! /bin/bash
while  read LINE;  do
#   for  word  in  $LINE
#   do
#    echo  "$word 1"
         awk  '{print $5}'                                                                                                       
   done
done


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
#! /bin/bash
count= 0
started= 0
word= ""
while  read LINE; do
   goodk=`echo $LINE | cut -d  ' '   -f  1 `
   if  "x"  == x "$goodk"  ];then
      continue
   fi
   if  "$word"  !=  "$goodk"  ];then
     [ $started -ne  0  ] && echo -e  "$word\t$count"
     word=$goodk                                                                                                                
     count= 1
     started= 1
   else
     count=$(( $count +  1  ))
   fi
done


有时候会出现这样的问题,好好看看自己写的mapreduce程序 ~

13/12/14 13:26:52 INFO streaming.StreamJob: Tracking URL: http://101.rui.com:50030/jobdetails.jsp?jobid=job_201312131904_0030

13/12/14 13:26:53 INFO streaming.StreamJob:  map 0%  reduce 0%

13/12/14 13:27:16 INFO streaming.StreamJob:  map 100%  reduce 100%

13/12/14 13:27:16 INFO streaming.StreamJob: To kill this job, run:

13/12/14 13:27:16 INFO streaming.StreamJob: /usr/local/hadoop/libexec/../bin/hadoop job  -Dmapred.job.tracker=localhost:9001 -kill job_201312131904_0030

13/12/14 13:27:16 INFO streaming.StreamJob: Tracking URL: http://101.rui.com:50030/jobdetails.jsp?jobid=job_201312131904_0030

13/12/14 13:27:16 ERROR streaming.StreamJob: Job not successful. Error: # of failed Map Tasks exceeded allowed limit. FailedCount: 1. LastFailedTask: task_201312131904_0030_m_000000

13/12/14 13:27:16 INFO streaming.StreamJob: killJob...

Streaming Command Failed!


python做为mapreduce执行成功后,结果和日志一般是放在你指定的目录下的,结果是在part-00000文件里面~


172550144.jpg


下面咱们谈下,如何入库和后台的执行









 本文转自 rfyiamcool 51CTO博客,原文链接:http://blog.51cto.com/rfyiamcool/1340057 ,如需转载请自行联系原作者