How do I do a bare-metal remote installation?
Why do it?
The default bare-metal installation is pretty much a no-brainer, so a bare-metal remote installation isn't that common a need.
However, a bare-metal remote installation is useful when the person who can physically access the cluster hardware is expected to spend minimal time at its physical location, and the aim is to set it up to allow an administrator who is not physically there to configure the cluster. Some possible examples:
- When installing to a head node in a data center, where a third-party data center administrator who knows very little about the cluster is agreeable to taking about 5 minutes to get the cluster installation started on the hardware, and then letting the cluster administrator log into the head node and do the rest remotely.
- When the cluster administrator can't get to the cluster hardware, and the only person around is the janitor, who can be talked through the process on a mobile phone.
- When the cluster administrator needs to rush off somewhere else but would like to complete the process remotely. Maybe the cluster is in a painfully cold or noisy environment, or maybe the administrator has to go home.
- When the cluster administrator needs help from Bright support due to an obscure problem during installation, and wants Bright to take over the installation.
The procedure
The Bright DVD is put in the head node, and the cluster is made to boot from it.
1. From the installer startup screen seen at the head node, click on the remote-install button:
2. At the head node still, an interface must be chosen through which the Bright installer will be accessible. Its IP address, netmask, and gateway must be configure, or just use DHCP if that is available to the head node. A password must also be setup for ssh root access:
3. After the interface is up, the remote administrator can ssh into the Bright installer to start remote installation. At this point, the third-party datacenter worker/janitor/whatever who is physically at the head node can leave it, and the cluster administrator can take over the installation procedure:
$ ssh root@10.2.184.6
root@10.2.184.6's password:
Welcome to Linux 2.6.32-431.5.1.el6.x86_64.
#
4. The cluster administrator customizes the build-config.xml to suit the hardware configuration of the head node:
# vim /cm/build-config.xml
5. start the installation:
# /cm/cm-master-install
Please specify the build config to use.
The default config is in /cm/build-config.xml
USAGE:
/cm/cm-master-install --config <path to build config> [OPTIONS]
OPTIONS:
--config - The xml build config file
(default: /cm/build-config.xml)
--mountpath - Path of CD/DVD if already mounted
--cdrom - CD/DVD to use (cannot use with --mountpath)
--help - Print this help
EXAMPLES:
/cm/cm-master-install --config /cm/build-config.xml
/cm/cm-master-install --config /cm/build-config.xml --cdrom /dev/hdc
/cm/cm-master-install --config /cm/build-config.xml --mountpath /mnt1/cdrom
/cm/cm-master-install --help
# /cm/cm-master-install --config /cm/build-config.xml
[1/12] Parsing build config ..... [ OK ]
Please set the root password:
Please repeat the password:
[2/12] Mounting CD/DVD-ROM ..... [ OK ]
[3/12] Partitioning harddrives ..... [ OK ]
[4/12] Installing CentOS 6 ..... [ OK ]
[5/12] Installing distribution packages ..... [ OK ]
[6/12] Installing Bright Cluster Manager packages ..... [ OK ]
[7/12] Configuring kernel and setting up bootloader ..... [ OK ]
[8/12] Installing CentOS 6 software image ..... [ OK ]
[9/12] Installing distribution packages to software image ..... [ OK ]
[10/12] Installing Bright Cluster Manager packages to software image ..... [ OK ]
[11/12] Finalizing installation ..... [ OK ]
[12/12] Initializing management daemon ..... [ OK ]
Do you want to reboot[y/n]:y
Rebooting ...