【转载】hadoop中遇到的坑

i1zhuxian

环境说明：
CentOS-6.4-x86_64-bin-DVD1.iso
hadoop-2.4.1.tar.gz
hbase-0.98.3-hadoop2-bin.tar.gz
jdk-7u79-linux-x64.tar.gz
scala-2.10.4.tgz
spark-1.2.0-bin-hadoop2.4.tgz.tar
zookeeper-3.4.6.tar.gz

附上下载地址：

hadoop2.4.1:http://archive.apache.org/dist/hadoop/common/hadoop-2.4.1/hadoop-2.4.1.tar.gz
hbase0.98.3:http://archive.apache.org/dist/hbase/hbase-0.98.3/hbase-0.98.3-hadoop2-bin.tar.gz
spark1.2.0:http://archive.apache.org/dist/spark/spark-1.2.0/spark-1.2.0-bin-hadoop2.4.tgz
zookeeper3.4.6:http://apache.fayea.com/zookeeper/zookeeper-3.4.6/zookeeper-3.4.6.tar.gz
scala2.10.4:http://www.scala-lang.org/files/archive/scala-2.10.4.tgz

---

1、hadoop与hbase依赖关系【没选好，后面就等着推倒重来吧】

见：Apache HBase Reference Guide
https://hbase.apache.org/book.html#configuration
----4.1.Hadoop章节

Paste_Image.png

2、namenode无法启动

日志报错“ulimit -a for user root”
解决：
重新格式化namenode，然后启动hadoop，jps存在namenode。

3、然也有可能datanode无法启动

日志报错“ulimit -a for user root”
原因：datanamenode运行时打开文件数，达到系统最大限制
当前最大限制
[iyunv@centos-FI hadoop-2.4.1]# ulimit -n
1024
解决：
调整最大打开文件数
[iyunv@centos-FI hadoop-2.4.1]# ulimit -n 65536
[iyunv@centos-FI hadoop-2.4.1]# ulimit -n
65536
再次启动hadoop
[iyunv@centos-FI hadoop-2.4.1]# jps
6330 WebAppProxyServer
6097 NameNode
6214 ResourceManager
6148 DataNode
6441 Jps
6271 NodeManager
6390 JobHistoryServer
ps：ulimit命令只是临时修改，重启又恢复默认，可在/etc/security/limits.conf 里修改 nofile 的限制。
参考：http://labs.chinamobile.com/mblog/225_17546

4、hadoop fs -ls报“WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable”

还是中招了64位操作系统，32位的jdk，不过还好该warn无影响可忽视。

5、hadoop fs -ls报“ls: `.': No such file or directory”

hadoop2.x不同于hadoop1.x的地方，
1.x是可以执行的，
而2.x的执行命令为：hadoop fs -ls /
【坑啊】

6、启动hbase时，hmaster启动后又消失了，且hin/base shell后，list执行报错"ERROR:can't get master address from ZooKeeper; znode data == null"

[iyunv@centos-FI hbase-0.98.3-hadoop2]# jps
28406 NameNode
28576 ResourceManager
32196 HRegionServer
32079 HMaster
28464 DataNode
32253 Jps
28748 JobHistoryServer
28635 NodeManager
24789 QuorumPeerMain
[iyunv@centos-FI hbase-0.98.3-hadoop2]# jps
28406 NameNode
28576 ResourceManager
32196 HRegionServer
28464 DataNode
32293 Jps
28748 JobHistoryServer
28635 NodeManager
24789 QuorumPeerMain
[iyunv@centos-FI hbase-0.98.3-hadoop2]#

日志提示：

2015-07-13 13:25:53,904 DEBUG [main-EventThread] master.ActiveMasterManager: A master is now available
2015-07-13 13:25:53,912 INFO [master:centos-FI:60000] master.ActiveMasterManager: Registered Active Master=centos-FI,60000,1436819147113
2015-07-13 13:25:53,921 INFO [master:centos-FI:60000] Configuration.deprecation: fs.default.name is deprecated. Instead, use fs.defaultFS
2015-07-13 13:25:54,250 FATAL [master:centos-FI:60000] master.HMaster: Unhandled exception. Starting shutdown.
java.net.ConnectException: Call From centos-FI/127.0.0.1 to centos-FI:9000 failed on connection exception: java.net.ConnectException: Connection refused; For more details see:http://wiki.apache.org/hadoop/ConnectionRefused

解决方法：
根据 http://wiki.apache.org/hadoop/ConnectionRefused 的这条提示信息"Check that there isn't an entry for your hostname mapped to 127.0.0.1 or 127.0.1.1 in /etc/hosts (Ubuntu is notorious for this)"检查：

[iyunv@centos-FI ~]# hostname -i
127.0.0.1 192.168.128.120
[iyunv@centos-FI hbase-0.98.3-hadoop2]# cat /etc/hosts
127.0.0.1 centos-FI localhost #localhost localhost.localdomain localhost4 localhost4.localdomain4

::1 centos-FI #localhost localhost.localdomain localhost6 localhost6.localdomain6

192.168.128.120 master
192.168.128.120 slave
192.168.128.120 centos-FI

改为：

[iyunv@centos-FI hbase-0.98.3-hadoop2]# cat /etc/hosts
127.0.0.1 localhost #localhost localhost.localdomain localhost4 localhost4.localdomain4

::1 centos-FI #localhost localhost.localdomain localhost6 localhost6.localdomain6

192.168.128.120 master
192.168.128.120 slave
192.168.128.120 centos-FI

即确保

[iyunv@centos-FI hbase-0.98.3-hadoop2]# hostname -i
192.168.128.120

hmaster进程正常后，hin/base shell再次执行list也就正常了
当然hin/base shell后，list执行报错"ERROR:can't get master address from ZooKeeper; znode data == null"还有可能是其他原因，详见：

http://blog.iyunv.com/u010022051/article/details/44176931

7、启动hbase时报错“localhost: ssh: Could not resolve hostname localhost: Temporary failure in name resolution”

[iyunv@centos-FI hbase-0.98.3-hadoop2]# bin/start-hbase.sh
starting master, logging to /opt/program/hbase-0.98.3-hadoop2/bin/../logs/hbase-root-master-centos-FI.out
localhost: ssh: Could not resolve hostname localhost: Temporary failure in name resolution

以上报错，/etc/hosts中必须存在127.0.0.1 localhost

8、CRT进入hbase shell后无法退格

在secureCRT中，点击【选项】【会话选项】【终端】【仿真】，右边的终端选择linux，在hbase shell中如输入出错，按住Ctrl+删除键(backspace) 即可删除！

9、执行spark-example：HBaseTest.scala 如下报错：

WARN TableInputFormatBase: Cannot resolve the host name for master/192.168.128.120 because of javax.naming.CommunicationException: DNS error [Root exception is java.net.PortUnreachableException: ICMP Port Unreachable]; remaining name '120.128.168.192.in-addr.arpa'

这是由于DNS服务器没有当前节点的记录（都没使用到DNS服务器），手动在/etc/hosts中添加一条记录"master 192.168.128.120"

10、使用flume拦截器时，报错"java.lang.NullPointerException: Expected timestamp in the Flume event headers, but it was null"

sink是hdfs，然后使用目录自动生成功能。出现如题的错误，看官网文档说的是需要在每个文件记录行的开头需要有时间戳，但是时间戳的格式可能比较难调节，所以亦可设置hdfs.useLocalTimeStamp这个参数，比如以每个小时作为一个文件夹，那么配置应该是这样：

a1.sinks.k1.hdfs.path = hdfs://ubuntu:9000/flume/events/%y-%m-%d/%H
a1.sinks.k1.hdfs.filePrefix = events-
a1.sinks.k1.hdfs.round = true
a1.sinks.k1.hdfs.roundValue = 1
a1.sinks.k1.hdfs.roundUnit = hour
a1.sinks.k1.hdfs.useLocalTimeStamp = true

修改之后再次执行flume确实是自动生成了hdfs目录，
flume日志:

2015-07-23 15:16:28,405 (SinkRunner-PollingRunner-DefaultSinkProcessor) [INFO - org.apache.flume.sink.hdfs.BucketWriter.open(BucketWriter.java:261)] Creating hdfs://master:9000/flume_test/15-07-23/events-.1437689788240.tmp

---

备忘

查看RM：http://ip:8088/
查看hdfs：http://ip:50070/
查看jobtrack；http://ip:50030/
hbase master web ui:http://ip:60010/
hbase region web ui:http://ip:60030/

• spark监控之web UI:
  每个SparkContext启动一个UI，默认端口4040，多个SparkContext端口4040以此累加，显示应用信息：

  高度器stage和task列表
  RDD大小和内存占用概况
  环境信息
  正在运行的executors信息

  
  

此信息之在应用运行时才能显示，若想运行后查看web UI，须在启动应用之前将spark-default.conf的spark.eventLog.enabled设置为true：spark将UI中显示的信息记录为spark事件，并记录到持久化存储中

spark.eventLog.enabled true
spark.eventLog.dir hdfs://centos-FI:9000/spark_eventLog

---

hadoop2.4.1参考资料：

• http://hadoop.apache.org/docs/r2.4.1/hadoop-project-dist/hadoop-common/ClusterSetup.html
• http://hadoop.apache.org/docs/r2.4.1/hadoop-yarn/hadoop-yarn-common/yarn-default.xml
• http://hadoop.apache.org/docs/r2.4.1/hadoop-project-dist/hadoop-hdfs/hdfs-default.xml