hadoop-hdfs-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Forrest Aldrich <for...@gmail.com>
Subject Re: Setting the HADOOP_HOME_DIR
Date Sun, 08 Dec 2013 22:28:30 GMT
Adam, my apologies, you are correct.  I meant to refer to the 
HADOOP_CONF_DIR -- this is what I get for posting before I've had enough 
coffee :-)

I want to put the CONF and LOGS directories outside of the /usr/hadoop 
structure -- this will make it easier to manage with Puppet, and prevent 
accidental overwrites during upgrades etc.


On 12/8/13, 4:56 PM, Adam Kawa wrote:
> Hi,
> If am not sure, if I understood your issue correctly. Would you like 
> to specify somehow where the configuration directory for your Hadoop 
> cluster is located (e.g. /etc/hadoop/conf)?
> If you use init scripts from CDH, they assume that config directory is 
> CONF_DIR="/etc/hadoop/conf".
> Afaik, when you use HDP or Apache distribution, then you can specify 
> where your configuration directory is when you start a script e.g. 
> "sudo -u hdfs /usr/lib/hadoop/sbin/hadoop-daemon.sh --config 
> <config_directory> start datanode"
> PS.
> I grepped my configuration directory, and installation directory 
> (/usr/lib/hadoop), but I can not see variable called: HADOOP_HOME_DIR 
> anywhere.
> I see that /usr/lib/hadoop/libexec/hadoop-layout.sh contains variable 
> HADOOP_CONF_DIR=${HADOOP_CONF_DIR:-"/etc/hadoop/conf"}.
> 2013/12/8 Forrest Aldrich <forrie@gmail.com <mailto:forrie@gmail.com>>
>     We have a home-grown Hadoop installation that I am primarily
>     responsible for.   I've managed to get new nodes rolled out, it
>     takes time as we have other dependencies.   One item I've not been
>     able figure out is where to set the HADOOP_HOME_DIR variable, so I
>     can store the actual configuration for each node separate from the
>     binary tree.
>     Can anyone point me to where this gets set properly? We have an
>     init.d script that starts the services on the master, which calls
>     out to the slaves (as user "hadoop") -- but I'm guessing the
>     variable can be started there, exported and inherited -- but
>     perhaps it may be more proper to set in ~hadoop/conf/hadoop-env.sh.
>     The idea is to enable me to more easily roll out slaves, perhaps
>     using Puppet, so that the CONF and LOGS directories are separate
>     -- it's easier to manage that way.
>     Thanks.

View raw message