hadoop 配置机架感知

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

原文地址：hadoop 配置机架感知, 感谢原作者分享。