欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页  >  数据库

hadoop 配置机架感知

程序员文章站 2022-03-31 22:21:30
...

周海汉?2013.7.24 http://abloz.com 假如设备链接层次分3层,第一层交换机d1下面连多个交换机rk1,rk2,rk3,rk4,. 每个交换机对应一个机架。 d1(rk1(hs11,hs12,),rk2(hs21,hs22,), rk3(hs31,hs32,),rk4(hs41,hs42,),) 可以用程序或脚本完成由host到设备的映射

周海汉?2013.7.24

http://abloz.com

假如设备链接层次分3层,第一层交换机d1下面连多个交换机rk1,rk2,rk3,rk4,…. 每个交换机对应一个机架。

d1(rk1(hs11,hs12,…),rk2(hs21,hs22,…), rk3(hs31,hs32,…),rk4(hs41,hs42,…),…)

可以用程序或脚本完成由host到设备的映射。比如,用python,生成一个topology.py:

然后在core-site.xml中配置

topology.script.file.name
/home/hadoop/hadoop-1.1.2/conf/topology.py
The script name that should be invoked to resolve DNS names to
NetworkTopology names. Example: the script would take host.foo.bar as an
argument, and return /rack1 as the output.

python机架脚本:

[hadoop@hs11 conf]$ cat topology.py
#!/usr/bin/env python

”’
This script used by hadoop to determine network/rack topology. It
should be specified in hadoop-site.xml via topology.script.file.name
Property.
topology.script.file.name
/home/hadoop/hadoop-1.1.2/conf/topology.py

To generate dict:
for i in range(xx):
#print “\”hs%d\”:\”/rk%d/hs%d\”,”%(i,(i-1)/10,i)

print “\”hs%d\”:\”/rk%d\”,”%(i,(i-1)/10)

Andy 2013.7.23
”’

import sys
from string import join

DEFAULT_RACK = ‘/rk0′;

RACK_MAP = {
“hs11″:”/rk1″,
“hs12″:”/rk1″,
“hs13″:”/rk1″,
“hs14″:”/rk1″,
“hs15″:”/rk1″,
“hs16″:”/rk1″,
“hs17″:”/rk1″,
“hs18″:”/rk1″,
“hs19″:”/rk1″,
“hs20″:”/rk1″,
“hs21″:”/rk2″,
“hs22″:”/rk2″,
“hs23″:”/rk2″,
“hs24″:”/rk2″,
“hs25″:”/rk2″,
“hs26″:”/rk2″,
“hs27″:”/rk2″,
“hs28″:”/rk2″,
“hs29″:”/rk2″,
“hs30″:”/rk2″,
“hs31″:”/rk3″,
“hs32″:”/rk3″,
“hs33″:”/rk3″,
“hs34″:”/rk3″,
“hs35″:”/rk3″,
“hs36″:”/rk3″,
“hs37″:”/rk3″,
“hs38″:”/rk3″,
“hs39″:”/rk3″,
“hs40″:”/rk3″,
“hs41″:”/rk4″,
“hs42″:”/rk4″,
“hs43″:”/rk4″,
“hs44″:”/rk4″,
“hs45″:”/rk4″,
“hs46″:”/rk4″,

“10.10.20.11″:”/rk1″,
“10.10.20.12″:”/rk1″,
“10.10.20.13″:”/rk1″,
“10.10.20.14″:”/rk1″,
“10.10.20.15″:”/rk1″,
“10.10.20.16″:”/rk1″,
“10.10.20.17″:”/rk1″,
“10.10.20.18″:”/rk1″,
“10.10.20.19″:”/rk1″,
“10.10.20.20″:”/rk1″,
“10.10.20.21″:”/rk2″,
“10.10.20.22″:”/rk2″,
“10.10.20.23″:”/rk2″,
“10.10.20.24″:”/rk2″,
“10.10.20.25″:”/rk2″,
“10.10.20.26″:”/rk2″,
“10.10.20.27″:”/rk2″,
“10.10.20.28″:”/rk2″,
“10.10.20.29″:”/rk2″,
“10.10.20.30″:”/rk2″,
“10.10.20.31″:”/rk3″,
“10.10.20.32″:”/rk3″,
“10.10.20.33″:”/rk3″,
“10.10.20.34″:”/rk3″,
“10.10.20.35″:”/rk3″,
“10.10.20.36″:”/rk3″,
“10.10.20.37″:”/rk3″,
“10.10.20.38″:”/rk3″,
“10.10.20.39″:”/rk3″,
“10.10.20.40″:”/rk3″,
“10.10.20.41″:”/rk4″,
“10.10.20.42″:”/rk4″,
“10.10.20.43″:”/rk4″,
“10.10.20.44″:”/rk4″,
“10.10.20.45″:”/rk4″,
“10.10.20.46″:”/rk4″,


}

if len(sys.argv)==1:
print DEFAULT_RACK
else:
print join([RACK_MAP.get(i, DEFAULT_RACK) for i in sys.argv[1:]],” “)

原来这个程序我返回的是

“hs11″:”/rk1/hs11″,

结果执行mapreduce程序时报如下错误:

Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there’s no reduce operator
Starting Job = job_201307241502_0003, Tracking URL = http://hs11:50030/jobdetails.jsp?jobid=job_201307241502_0003
Kill Command = /home/hadoop/hadoop-1.1.2/libexec/../bin/hadoop job? -kill job_201307241502_0003
Hadoop job information for Stage-1: number of mappers: 0; number of reducers: 0
2013-07-24 18:38:11,854 Stage-1 map = 100%,? reduce = 100%
Ended Job = job_201307241502_0003 with errors
Error during job, obtaining debugging information…
Job Tracking URL: http://hs11:50030/jobdetails.jsp?jobid=job_201307241502_0003
FAILED: Execution Error, return code 2 from org.apache.hadoop.hive.ql.exec.MapRedTask
MapReduce Jobs Launched:
Job 0:? HDFS Read: 0 HDFS Write: 0 FAIL
Total MapReduce CPU Time Spent: 0 msec

通过http://hs11:50030/jobdetails.jsp?jobid=job_201307241502_0002?可以看到:

Job initialization failed:

java.lang.NullPointerException

at?org.apache.hadoop.mapred.JobTracker.resolveAndAddToTopology(JobTracker.java:2751)
at?org.apache.hadoop.mapred.JobInProgress.createCache(JobInProgress.java:578)
at?org.apache.hadoop.mapred.JobInProgress.initTasks(JobInProgress.java:750)

at org.apache.hadoop.mapred.JobTracker.initJob(JobTracker.java:3775)

at?org.apache.hadoop.mapred.EagerTaskInitializationListener$InitJob.run(EagerTaskInitializationListener.java:90)
at?java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
at?java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
at java.lang.Thread.run(Thread.java:662)

原来系统在配置机架敏感时,并不需要在脚本中返回设备ns或hostname,系统会自动添加。改为上面的topology.py后,系统执行正确。

相关博文:

  1. hadoop 打印配置变量
  2. hadoop 中的 ClassNotFoundException
  3. hadoop ubuntu集群安装