This chapter describes how to recover a cluster node. There are several methods available to recover a node:
A simple, kick-start procedure
Documented in this chapter.
SystemImager
Located on the Linux Graphics Cluster Resource CD in /unsupported/systemimager/*.
Mindi and Mondo
Accessible at http://www.microwerks.net/~hugo/.
This chapter describes the simple, kick-start recovery procedure in the following sections:
To set up your system for the installation and to install Red Hat Linux, perform the following steps.
Mount the Linux Graphics Cluster Resource CD and create a kick-start floppy disk by using the following command:
# dd of=/dev/fd0 if=/mnt/cdrom/unsupported/ks_gc_1.0.img |
You may wish to edit or replace the /dev/fd0/ks.cfg file. SGI has pre-installed several applications that may not be useful to your activities. The installation from the Red Hat 7.1 distribution consumes 1.765 GB of disk space and installs 701 packages by using the the included ks.cfg file.
The following CDs will be used with the kick-start floppy disk:
SGI XFS 1.0 for Red Hat 7.1
Red Hat Linux 7.1 Operating System, Binary CD 1
Red Hat Linux 7.1 Operating System, Binary CD 2
Before starting the installation, back up any files from the node you wish to keep (assuming the node is somewhat functional).
| Note: During a kick-start installation, the Red Hat installer will remove any partitions and reformat the hard drive. If multiple hard drives are found, it will span Linux partitions across all of the hard drives it finds. |
Power down the node and disconnect or remove power from all hard drives, except the system drive.
Ensure you have a monitor, mouse, and keyboard attached to the node being recovered.
Power up the node, insert the kick-start floppy disk into the floppy disk drive, and insert the SGI XFS 1.0 for Red Hat 7.1 CD into the CD-ROM drive.
The installer creates the following partitions:
| /boot | 32 MB, XFS format | |
| / | 8 GB, XFS format | |
| swap | 2047 MB | |
| /home | Remaining free disk space, XFS format |
The installer copies the installation image onto the system drive and then prompts for the Red Hat Linux 7.1 CD 1.
Load the Red Hat Linux 7.1 CD 1.
It installs several files up to the point of installing the kernel. It ejects the Red Hat Linux 7.1 CD 1 and asks for the SGI XFS 1.0 for Red Hat 7.1 CD.
Load the SGI XFS 1.0 for Red Hat 7.1 CD and await the next prompt to re-insert the Red Hat Linux 7.1 CD 1.
When prompted, load the Red Hat Linux 7.1 CD 1.
Subsequently, the installer requests the Red Hat Linux 7.1 CD 2.
Load the Red Hat Linux 7.1 CD 2 and continue the installation.
Using the kick-start floppy disk, the installation of the base Red Hat Linux distribution will consume approximately 30 minutes.
When prompted, press the Enter key to end the installation.
As the node reboots, remove the kick-start floppy disk and Red Hat Linux CD 2 from the CD-ROM drive .
The node comes up without X windows. The following steps describe how to install the X Window System.
Log in as root.
Place the SGI Linux Graphics Cluster Resource CD into the CD-ROM drive and mount it, as shown in the following:
[root@master-channel /root]# mount /dev/cdrom -t iso9660 /mnt/cdrom |
Change the current directory to that of the CD and start the installation script, as shown in the following:
[root@master-channel /root]# cd /mnt/cdrom [root@master-channel /root]# ./INSTALL |
When prompted, select the appropriate optional software and follow the instructions as documented in the installation section of the README.GCE file located on the Graphics Cluster Resource CD.
When the installation has completed, switch to the home directory, umount the CD, remove it from the CD-ROM drive, and reboot the system, as shown in the following:
[root@master-channel /root]# cd $HOME [root@master-channel /root]# umount /mnt/cdrom [root@master-channel /root]# eject [root@master-channel /root]# sync [root@master-channel /root]# reboot |
When the system has booted to multi-user level, log in as root, using the password sgisgi. Change the password to one suitable for your installation.
Switch the desktop to K Desktop Environment (KDE) using the GUI menu as follows:
Gnome -> Programs -> System -> Desktop Switching Tool |
Select the KDE desktop.
You now need to appropriately configure the network. The following steps describe how to do this.
Log out and log in as root.
Bring up the network configuration GUI as follows:
KDE -> System -> Network Configuration |
Set up the host name and domain definitions.
Select Hosts and add in the host names and IP address for the network interfaces.
| Note: Master nodes have up to three host name/IP pairs. Channel nodes have up to two host name/IP pairs. |
Select Interfaces and associate the main IP address of the node with eth0.
Associate any additional IP addresses with appropriate additional interfaces.
Ensure that the button Activate on Boot is depressed. Click OK for each interface definition and save the definition.
Select Routing.
Ensure that the default gateway is the master-channel's IP address for channel nodes.
Enter the appropriate Ethernet IP address for the master channel.
Enter the gateway device's IP address.
Use eth0 for the default gateway device.
Save the configuration and quit from the network configuration GUI.
Reboot the system.
You may do so from a terminal at the command line. If you log out, you will experience a gray screen with the X mouse cursor. This is due to the network definitions having changed and the window manager KDE, in this case, is trying to use the host name.
To effect a reboot at this stage, you need to do the following two things:
Restart X Windows.
While the screen is blank, reboot the system
These two steps are accomplished by the following two sets of keyboard actions:
Ctrl-Alt-Backspace Ctrl-Alt-Delete |
Note that Backspace is sometimes the <- key.
When the system has rebooted, log in as root and check the /etc/modules.conf file to ensure that the network interfaces have been properly assigned to the proper driver, as shown in the following example:
[root@master-channel /root]# cat /etc/modules.conf alias eth0 eepro100 alias eth1 eepro100 alias scsi_hostadapter aic7xxx alias eth2 acenic alias parport_lowlevel parport_pc post-install sound-slot-0 /bin/aumix-minimal -f /etc/.aumixrc -L >/dev/null 2>&1 || : pre-remove sound-slot-0 /bin/aumix-minimal -f /etc/.aumixrc -S >/dev/null 2>&1 || : alias usb-controller usb-uhci alias char-major-195 NVdriver options NVdriver NVreg_EnableVia4x=1 alias char-major-253 imagesync alias char-major-254 windrvr include /etc/modules.alsa |
Note that the Gigabit Ethernet driver acenic has been aliased to eth2. Linux will sometimes assign the wrong driver to the wrong adapter and while the network will appear to function, no packets will be sent or received.
If necessary, manually edit the file /etc/modules.conf to correct driver misassignments and enter the following:
[root@master-channel /root]# depmod |
To further verify that the network is functioning, enter the following:
[root@master-channel /root]# netstat -ina Kernel Interface table Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg eth0 1500 0 91918 0 0 0 122900 0 0 0 BRU eth1 1500 0 1 0 0 0 55 0 0 0 BRU eth2 1500 0 110353 0 0 0 2 0 0 0 BRU lo 16436 0 21 0 0 0 21 0 0 0 LRU |
Note that in the preceding example, all interfaces have activity.
Further, detailed verification of network setup can be determined from the following files:
/etc/sysconfig/network
/etc/sysconfig/network-scripts/ifcfg-eth{0,1,2}
/etc/resolv.conf
/etc/hosts
If the system is a master channel, ensure pconsole functions properly by using it to attach to one or more channels, as shown in the following example:
[root@master-channel /root]# pconsole.sh channel0 channel1 |
Bring up the User Manager GUI or use the adduser command line interface. Add the gcadmin account with the following parameters:
| login | gcadmin | |
| Name | Graphics Cluster Admin. | |
| User id | 501 | |
| Use Private Group | uncheck | |
| Password | sgisgi (Do change.) | |
| Primary Group | adm | |
| Groups Belong to | sys |
If you are using the User Manager GUI, save the new user information and exit out of the GUI.
Add the appropriate entries into the file /etc/fstab for any additional hard drives you may have. At a minimum, you must add the following line to /etc/fstab:
/dev/cdrom /mnt/cdrom auto noauto,owner 0 0 |
Now shut down the node and re-attach and re-power any additional hard drives. Power the system back up and ensure they all mount properly during the boot up procedure.
If the node being recovered is a channel node, use your editor of choice to create the file /etc/sysconfig/autologin with the following contents:
AUTOLOGIN=yes EXEC=/usr/X11R6/bin/startx USER=gcadmin |
Save the file, reboot the node, and select KDE as the window manager (see “Changing the Desktop to KDE”).
You may now shut down the node and remove the mouse and keyboard. If appropriate, re-attach the proper monitor to the node. Re-attach and re-power any additional hard drives and power up the node. Verify that any additional hard drives mount properly during bootup.
If you want to reinstate utilities such as rsync, cron jobs, or other customizations, now is an appropriate time to add them back to the channel node. When doing any desktop customizations, shut off screen savers and the like.