Fedora-GlusterFS-Hadoop Installation Script

== Overview ==

The install.sh script (and the companion prep_node.sh script) sets up

GlusterFS on Fedora for Hadoop workloads. The storage (brick) partition

(usually /dev/sdb) can be configured as RAID 6 or as JBODs.

A tarball named "glusterfs-cluster-install-<version>.tar.gz" is downloaded to

one of the cluster nodes or to the user's localhost. The download directory

is arbitrary. install.sh requires password-less ssh from the node hosting the

rhs install tarball (the "install-from" node) to all nodes in the cluster.

GlusterFS does not require password-less SSH.

The tarball contains the following:

- hosts.example: sample "hosts" config file.

- install.sh: the main install script, executed by the root user.

- prep_node.sh: companion script, not to be executed directly.

- README.txt: this file.

- devutils/: utility scripts, eg. password-less SSH setup.

install.sh is the main script and should be run as the root user. It copies

prep_node.sh and other files to each node contained in the "hosts" file, and

ensures that each node is setup to be a data (tasktracker) node.

== Before you begin ==

The "hosts" file must be created by the user. It is not part of the tarball

but an example hosts file is provided. The "hosts" file is expected to be

created in the same directory where the tarball has been downloaded. If a

different location is required the "--hosts" option can be used to specify

the "hosts" file path. The "hosts" file contains a list of IP adress followed

by hostname (same format as /etc/hosts), one pair per line. Each line

represents one node in the storage cluster (gluster trusted pool). Example:

ip-for-node-1 hostname-for-node-1

ip-for-node-3 hostname-for-node-3

ip-for-node-2 hostname-for-node-2

ip-for-node-4 hostname-for-node-4

IMPORTANT: the node order in the hosts file is important. Assuming the Gluster

volume is created with replica 2 then each pair of lines in hosts represents

replica pairs. For example, the first 2 lines in hosts are replica pairs, as

are the next two lines, etc. Hostnames are expected to be lower-case.

Also:

- passwordless SSH is required between the installation node and each storage

node. See the Addendum at the end of this document if you would like to see

instructions on how to do this.

Essentially, Fedora needs to be installed with the

hostname configured with a static IP address. Do not create a gluster

volume.

- the order of the nodes in the "hosts" file is in replica order

== Installation ==

Instructions:

0) upload glusterfs-cluster-install-<version> tarball to the deployment

directory on the "install-from" node.

1) extract tarball to the local directory:

$ tar xvzf glusterfs-cluster-install-<version>.tar.gz

2) cd to the extracted glusterfs-cluster-install directory:

$ cd glusterfs-cluster-install-<version>

3) execute "install.sh" from the install directory:

$ ./install.sh [options (see --help)] <brick-dev> (note: brick_dev is

required)

For example: ./install.sh /dev/sdb

Output is displayed on STDOUT and is also written to:

/var/log/glusterfs-cluster-install.log

on both the delpoyment node.

4) The script should complete at which ...

5) Validate the Installation

Open a terminal and navigate to the Hadoop Directory

cd /usr/lib/hadoop

Change user to the mapred user

su mapred

Submit a TeraGen Hadoop job test

bin/hadoop jar hadoop-examples-1.2.0.1.3.2.0-111.jar teragen 1000 in-dir

Submit a TeraSort Hadoop job test

bin/hadoop jar hadoop-examples-1.2.0.1.3.2.0-111.jar terasort in-dir out-dir

== Addendum ==

1) Setting up password-less SSH

There is a utility script (devutils/passwordless-ssh.sh) which will set up

password-less SSH from localhost (or wherever you run the script from) to

all hosts defined in the local "hosts" file. Use --help for more info.