SLinstall

Notes

• these instructions are periodically updated to include items needed for older/newer versions of Linux. They are marked like this: (SL4.2+) means Scientific Linux 4.2 and newer; (SL4 is equivalent to FC3). (FC5 only) means Fedora Core 5; etc.
• obsolete items are marked by the "#" sign at the beginning of the line and sometimes have a comment about the reason for removal.
• typically, we do not "upgrade" machines using the Red Hat "upgrade" function. Instead, we save critical files from the old installation and do a "fresh install" from scratch

Preparation

• save /etc, /var, /root, /usr/local, /opt and /tftpboot (tar and scp to another machine or use either of: /triumfcs/trshare/midas/Disks/rsync_all_NIS.csh or /triumfcs/trshare/midas/Disks/rsync_all_noNIS.csh)
• NIS only: ascertain NIS domain name (use authconfig) e.g. "DAQ-NIS","MUSR-NIS" etc.
• check existing partition sizes on machine: df -hl, fdisk -l
• note which are the /home1 and /data partitions
• if /home1 is inside the "/"partition you must save it also
• shutdown

Running SL installer

• Start installation of the new system:
• IMPORTANT: if you have WDC "advanced partitioning disks" (4kB sectors), disks have to be repartitioned before use, see special instructions (TBW) (note: use fdisk -H 224 -S 56 /dev/sdx)
• boot from latest "SL5 kickstart" CD from Kelvin Raywood or PXE boot the latest SL installation image
  
• after the system enters graphical mode, one can remove the CD- the installation is running over the network
• two questions will be asked: how to partition the disks and the root password. The rest of the installation is automatic.
• select "Custom partioniong". You can create either simple normal partitions if you have only one disk or RAID 1(mirror) for / and /home1 if you have two disks.
  

     <strong>For the case of one disk only</strong>: - no "/boot" filesystem,
     - allocate at least 20 GBytes for "/" filesystem (primary partition, hda1)
     - allocate at least 8 GBytes for swap (primary partition, hda2)
     - allocate the rest as /home1 and /data (primary partitions hda3 and hda4)

<strong>For the case of two or more disks:</strong>
- If there are already some partitions on the disks, consider DELETING them all 
- Click New, select software RAID for /dev/sda, 20000MB (20 GB) and Force primary partition
- Same as above for /dev/sdb
- Click RAID, select create a RAID device /dev/md0, mount point /, RAID Level1 (mirror)
- Repeat for the swap partition (/dev/md1, make it at least 8 Gbytes)
- Leave the rest of the disks free. The other RAID partitions will be created AFTER installation.

• enter the root password for the new system
• continue with installation, at the very end, the system will ask you to reboot
  
• boot newly installed system and answer the few questions for SL5
  
• Firewall: Leave it disabled; SELinux: choose Disabled; KDump: Leave it disabled; Date and Time: Leave kickstart defaults (should be NTP using TRIUMF time servers)
  
• Create user: not necessary if you are using NIS; Sound card: ignore possibly.
• The system will reboot again.

Configure SSH

• Login from the console
• restore the SSH keys from backup (/etc/ssh/*key*)
• service sshd restart
• ssh into the new machine as root
• ssh root@localhost, ctrl-C
• scp root@ladd00:/root/authorized_keys ~root/.ssh/
• (not needed for SL5.5 kickstart) check that /etc/ssh/ssh_config contains "ForwardX11 yes" and "ForwardX11Trusted yes".

Configure RAID arrays

If you created only one mirror partitions, it is time to create the others: Create one 60 GB primary partition on each disks for /home1 and another one using the remainder of space.

<strong>fdisk /dev/sda</strong>
...
Command (m for help): <strong>n</strong>
Command action ...   <strong>p</strong>
Partition number ... <strong>2, 3 or 4</strong> according to what has been defined before
First cylinder ... default
Last cylinder ...  <strong>+60000M</strong>  or default
Command action ...   t
Partition number ... : <strong>2, 3 or 4</strong> according to what has been defined before
Hex code ... : fd if you intend to include this into a RAID array
Command action ...   <strong>p to check all is correct</strong>
Command (m for help): <strong>w
.... </strong>The new table will be used at the next reboot.<strong>
</strong><strong><strong>fdisk /dev/sdb  </strong></strong>and repeat as above
<strong><em>-  Reboot the machine  ---
</em></strong>

• Check the newly created partitions: fdisk -l /dev/sda; fdisk -l /dev/sdb
• mdadm --create /dev/md2 -a yes -l 1 -n /dev/sda3 /dev/sdb2
• Check the progress of building the RAID with: more /proc/mdstat
• When finished: mkfs -j -L /home1 /dev/md2; tune2fs -i 0 -c 0 /dev/md2; mkdir /home1
• Add a to /etc/fstab: "/dev/md2 /home1 ext3 defaults 1 2"
• Finally mount this new partition: mount -a
• Repeat from mkfs for each of the data partitions
• At this point you should have these disk partitions (single-disk in parenthesis)
  • /dev/md0 (/dev/hda1) is the system partition, 20 GBytes or more
  • /dev/md1 (/dev/hda2) is the swap partition, 8 GBytes or more
  • /dev/md2 (/dev/hda3) is the /home1 partition, 60 GBytes or more
  • /dev/md3 or /dev/sda4 (/dev/hda4), etc are the data partitions

• (SL5.5 or newer) enable raid1 bitmap files, for each /dev/mdX device, run "mdadm --grow --bitmap=internal /dev/mdX". In /proc/mdstat, verify that there is a "bitmap" entry. This will fail if the md device reconstruction is in progress.

Post installation

(on midm15/midm9b/midm20 only) install correct ethernet driver eepro100 not e100

restore /home (non-NIS) or /home1 (NIS) and other required user directories from backup.
(Can use /triumfcs/trshare/midas/Disks/rsync_back.csh ).

if needed, for non-NIS only, make a softlink for /home1 (NIS automounts /home).
ln -s /home /home1

Restore users accounts (non-NIS and NIS master only):
- edit /etc/passwd and /etc/shadow
Append users' login info to the end of these files from the backup versions.

edit ~root/.forward
(create file if not present, or restore from backup) In file add email address of the person(s) to receive root's email - always include Konstantin - olchansk@triumf.ca)

cleanup network setup (or postfix would not work):

•  check that /etc/sysconfig/network has the entry "HOSTNAME=xxx.triumf.ca" (fully qualified domain name)
• check that /etc/hosts has only the entry for "localhost".
  

Configure NIS and AUTOFS

###run "system-config-authentication", check "select NIS support", click on "configure NIS", set NIS domain, leave NIS server field blank.

run "authconfig --enablenis --nisdomain LADD-NIS --update"

run "sed 's/^hosts:.*/hosts: files dns/' -i /etc/nsswitch.conf" (to undo a mistake from authconfig)

On the master NIS node (ladd00), add this new node to /etc/netgroup, and update NIS maps (cd /var/yp; make)

(if needed) To setup NIS slave server:

•  check if /usr/lib64/yp exists. If not, run "yum install ypserv"

• /usr/lib64/yp/ypinit -s ladd00 (/usr/lib/yp/ypinit on 32-bit machines)

• chkconfig ypserv on

• service ypserv start

• service ypbind restart

• add the new machine to /var/yp/ypservers on the NIS master, run "make -C /var/yp" and also "cd /var/yp; yppush -h newmachine ypservers"

• If using securenets on the NIS master, remember to copy this file to the new slave server in /var/yp

(if NIS master or standalone) check /etc/auto.* against backups, particularly auto.master if NIS master

(if needed) add "+auto.master" at the end of /etc/auto.master

restart autofs to use the newly configured NIS maps: "service autofs restart"

Use "system-config-users" to add local user accounts

NIS: check user accounts: run "ypcat -k passwd"

Configure time

Verify time and date configuration. Run "ntpstat", it should say "synchronised to NTP server (142.90.x.y). If not, run "system-config-time", check "use network time", enter NTP servers: time1, time2, time3

Configure system updates

(If needed) enable automatic system updates: run "/triumfcs/trshare/olchansk/linux/triumf-update/yum-setup.perl -o"

(If needed) enable automatic kernel updates: run "/triumfcs/trshare/olchansk/linux/triumf-update/yum-setup.perl -k -o"

Configure system services

• chkconfig --list | grep on | sort (to see enabled services)
• disable unwanted services:

(only if amanda is not used) -> chkconfig --level 12345 xinetd off
chkconfig --level 12345 canna off
chkconfig --level 12345 FreeWnn off
chkconfig --level 12345 hpoj off
chkconfig --level 12345 ip6tables off
chkconfig --level 12345 iptables off
chkconfig --level 12345 isdn off
chkconfig --level 12345 pcmcia off
chkconfig --level 12345 rhnsd off
chkconfig --level 12345 spamassassin off
chkconfig --level 12345 bluetooth off
chkconfig --level 12345 apmd off
chkconfig --level 12345 iiim off
chkconfig --level 12345 fenced off
chkconfig --level 12345 ccsd off
chkconfig --level 12345 cpuspeed off
chkconfig --level 12345 pcp off
chkconfig --level 12345 pmie off
chkconfig --level 12345 yum-updatesd off
chkconfig --level 12345 clvmd off
chkconfig --level 12345 cman off
chkconfig --level 12345 lvm2-monitor off
chkconfig --level 12345 modclusterd off
chkconfig --level 12345 yum-updateonboot off
chkconfig --level 12345 cmirror off
chkconfig --level 12345 lock_gulmd off
chkconfig --level 12345 firstboot off
chkconfig --level 12345 ricci off
chkconfig --level 12345 gfs off
chkconfig --level 12345 scsi_reserve off
chkconfig --level 12345 openibd off
chkconfig --level 12345 arptables_jf off
chkconfig --level 12345 auditd off
chkconfig --level 12345 avahi-daemon off
chkconfig --level 12345 hplip off
chkconfig --level 12345 iscsi off
chkconfig --level 12345 iscsid off
chkconfig --level 12345 mcstrans off
chkconfig --level 12345 pcscd off
chkconfig --level 12345 restorecond off
chkconfig --level 12345 setroubleshoot off
chkconfig --level 12345 xend off
chkconfig --level 12345 xendomains off
chkconfig --level 12345 kudzu off
#chkconfig --level 12345 yum-cron off

Configure hardware sensors

• yum install lm_sensors
• sensors-detect (accept default answer to all questions - press ENTER)
• service lm_sensors restart (to reload the kernel modules)
• sensors (to see available sensors)
• (not needed) if sensors reports "General parse error", do "cp /triumfcs/trshare/olchansk/linux/lm_sensors/lm_sensors-3.0.1/prog/sensors/sensors /usr/bin/sensors" and try again

Enable NFS

• edit /etc/hosts.allow, add or uncomment "mountd: 142.90.0.0/255.255.0.0"
• create /etc/exports, e.g. "/home1 @daqmachines(rw,no_root_squash,async)"
• chkconfig nfs on
• chkconfig nfslock on
• service nfs restart

Enable AMANDA backups

AMANDA backups are already enabled by TRIUMF kickstart installs. For non-kickstart installation, follow instructions at [http://amanda/~amanda], or look at "/triumfcs/trshare/olchansk/linux/amanda/amanda-enable.perl". As final step, use [https://helpdesk.triumf.ca] to contact TRIUMF CS to add this new machine to the amanda backup list.

Enable DCACHE

• mkdir -p /pnfs/triumf.ca
• edit /etc/rc.local, add to the end of file: "mount -o intr,rw,noac,hard,nfsvers=2 trdata00:/pnfsdoor /pnfs/triumf.ca"
• . /etc/rc.local

section ddd

(not needed for SL5.5 kickstart) erase unwanted packages: yum erase logwatch mailman mrtg inn inn-devel cyrus-imapd cyrus-imapd-devel cyrus-imapd-murder cyrus-imapd-nntp webalizer squirrelmail rhn-applet yumex-applet apt-autoupdate SL_enable_serialconsole tog-pegasus kernel-largesmp kernel-hugemem kernel-largesmp-devel spamassassin slrn-pull openafs kernel-module-openafs openafs-debug openafs-devel openafs-kernel-source kernel-largesmp kernel-hugemem kernel-hugemem-devel xen kernel-xen bash-completion

install xemacs (from EPEL): yum install "xemacs*"

install ganglia:

scp ladd00:/etc/gmond.conf /etc/gmond.conf
yum install ganglia ganglia-gmond
chkconfig gmond on
service gmond restart

install misc stuff

cd /triumfcs/trshare/olchansk/linux/misc-rpms
rpm -vh --install triumf-ko-ganglia-*
rpm -vh --install emailonreboot-*
rpm -vh --install monitor_nfs-*
rpm -vh --install /triumfcs/trshare/olchansk/public_html/diskscrub/download/diskscrub*

install memtest86: "yum install memtest86+", "ls -1 /boot/* | grep memtest" to find out the name of the memtest boot file to use on the "kernel" line of the incantation to be added at the end of /boot/grub/grub.conf:

cat >> /boot/grub/grub.conf

title memtest
      root (hd0,0)
      kernel /boot/memtest86+-1.65

• yum install tkcvs
• install node monitoring

  cd /triumfcs/trshare/olchansk/linux/misc-rpms
  rpm -vh --install triumf_nodeinfo.noarch.rpm
  /usr/sbin/sendnodeinfo.perl --config ladd00:8600
  edit /etc/nodeinfo
  /usr/sbin/sendnodeinfo.perl ladd00:8600

• (before FC5) remove the fancy screen savers

  rm -f /usr/X11R6/lib/xscreensaver/*
  rm -f /usr/bin/*.kss

• chmod a+r /var/log/messages
• (before FC5) edit /etc/updatedb.conf, set "DAILY_UPDATE=yes"
• cd ~; yum update

Install packages needed for ROOT and MIDAS DAQ

yum install giflib.i386 giflib.x86_64 compat-libf2c-34.i386 mysql-devel sysstat "libusb-devel*" unixODBC-devel postgresql-devel libxml2-devel libXpm-devel libgfortran libstdc++-devel.i386 git

Section eee

• yum install xpdf (from EPEL)
• (do not do this) install all missing packages: /triumfcs/trshare/olchansk/linux/triumf-update/yum-everything.perl
• (SL5) If yum complains about unsigned java packages: cd ~; rpm -vh --upgrade /triumfcs/mirror/scientificlinux.org/5x/i386/updates/security/jdk*rpm /triumfcs/mirror/scientificlinux.org/5x/i386/updates/security/java*rpm

Section fff

• (SL5.5 or newer) enable raid1 bitmap files, for each /dev/mdX device, run "mdadm --grow --bitmap=internal /dev/mdX". In /proc/mdstat, verify that there is a "bitmap" entry. This will fail if the md device reconstruction is in progress.
  

Configure GRUB boot loader

• edit /boot/grub/grub.conf, remove the "quiet" and "rhgb" options
• edit /boot/grub/grub.conf, comment out (with "#") the "splashimage=" line
• check that GRUB boot loader is installed on all system disks: dd if=/dev/sda bs=1 count=1024 2>&1 | strings | grep GRUB
• if GRUB is not installed, (i.e. on the 2nd disk of machines with mirrored system disk), (but check that /dev/sdb is the right disk):

# grub
grub> device (hd0) /dev/sdb
grub> root (hd0,0)
grub> setup (hd0)

Configure X11 graphics

Special settings for DAQ

• add the following at the end of /etc/X11/xorg.conf. The enables Ctrl-Alt-KP-/ and Ctrl-Alt-KP-* to unlock the keyboard after Altera Quartus crash:

Section "ServerFlags"
        Option "AllowDeactivateGrabs" "true"
        Option "AllowClosedownGrabs" "true"
EndSection

Install NVIDIA drivers

• yum install nvidia-x11-drv kmod-nvidia (if it fails due to conflict with module-init-tools, run "yum --disablerepo \* --enablerepo elrepo update module-init-tools")
• mv /etc/X11/xorg.conf /etc/X11/xorg.conf-xxx
• nvidia-xconfig
• /dev/MAKEDEV nvidia
• restart the X11 server (Ctrl-Alt-Backspace or killall gdm-binary)
• observe that X11 server restarts using the NVIDIA driver (big NVIDIA logo on startup)
• if needed, login as root and run "nvidia-settings" to setup dual-screen configuration, etc

Manual selection of monitor, video mode and resolution

Automatic selection of monitor and video mode usually works. When it does not, configure it manualls:

• physically go to the computer
• login as root
• run "system-config-display".
• In the "hardware" tab, select monitor type: "generic LCD 1280x1024" or "generic LCD 1600x1200".
• In the "settings" tab, select "1280x1024" or "1600x1200" and "Thousands of colors".
• Press "ok", the display settings application should close.
• Logout, the new login window should use the new settings.

Finish installation

• logout and reboot the computer to have all the changes to take effect