Hadoop 2.2.0 fails running start-dfs.sh with Error: JAVA_HOME is not set and could not be found

Question

I have a work in progress installation of Hadoop in Ubuntu 12.x. I already had a deploy user which I plan to use to run hadoop in a cluster of machines. The following code demonstrate my problem basically I can ssh olympus no problems but start-dfs.sh fails doing exactly that:

deploy@olympus:~$ ssh olympus
Welcome to Ubuntu 12.04.4 LTS (GNU/Linux 3.5.0-45-generic x86_64)

    * Documentation:  https://help.ubuntu.com/

Last login: Mon Feb  3 18:22:27 2014 from olympus
deploy@olympus:~$ echo $JAVA_HOME
/opt/dev/java/1.7.0_51

deploy@olympus:~$ start-dfs.sh
Starting namenodes on [olympus]
olympus: Error: JAVA_HOME is not set and could not be found.

Answer 1

You can edit hadoop-env.sh file and set JAVA_HOME for Hadoop

Open the file and find the line as bellow

export JAVA_HOME=/usr/lib/j2sdk1.6-sun

Uncomment the line And update the java_home as per your environment

This will solve the problem with java_home.

Answer 2

Weird out of the box bug on Ubuntu. The current line

export JAVA_HOME=${JAVA_HOME}

in /etc/hadoop/hadoop-env.sh should pick up java home from host but it doesnt.

Just edit the file and hard code the java home for now.

Answer 3

Alternatively you can edit /etc/environment to include:

JAVA_HOME=/usr/lib/jvm/[YOURJAVADIRECTORY]

This makes JAVA_HOME available to all users on the system, and allows start-dfs.sh to see the value. My guess is that start-dfs.sh is kicking off a process as another user somewhere that does not pick up the variable unless explicitly set in hadoop-env.sh.

Using hadoop-env.sh is arguably clearer -- just adding this option for completeness.