Build a new server

Revision as of 01:27, 15 October 2011

Details on all the steps necessary to take a new (or old repurposed) piece of hardware and turn it into a happy functional server

Before you begin

• find the machine's MAC address (racadm getsysinfo)
• decide on private / public IP address, and if it will need to exist in some special range
• decide how the disks should be arranged (raid, partitioning, etc.)

Initial hardware setup

• Get the hardware racked and cabled (RobH)
• Get an IP and name (RobH)

• follow DNS How-To section to add the name/ip to DNS
• set up $name.mgmt.$loc.wmnet as well to access the management interface
• set up $assettag.mgmt.$loc.wmnet to the same IP as $name.mgmt.$loc.wmnet.

• Set up DHCP with the MAC address / name info
  • if it's a Dell, get MAC address from the mgmt console, run racadm getsysinfo; we use the first interface
  • log into brewster and edit /etc/dhcp3/* (any new server linux-host-entries.ttyS1-115200), run /etc/init.d/dhcpd3-server restart
• Get the switch set up to pass traffic to the host (Mark)
• set up the hardware raid (if it has it)

PXE boot and the initial OS

• if it's a cluster host, set up netboot to partition the disk
  • log into the install server (presently brewster), edit /srv/autoinstall/netboot.cfg
  • it's a bash case statement. Make sure your hostname is matched by a regex in there.
• if it's a misc host, you'll have to partition by hand (it will prompt you)
  • lvm over raid 1 is a decent config if you don't have anything more specific you need.
• ssh to root@$servername.mgmt.$loc.wmnet, force a restart and pxe boot (if it's a dell, copy/paste this: Dell_PowerEdge_R410_&_R610#PXE_booting dell pxe boot
  • get the password from someone in ops if you don't have it
  • powercycle the host: racadm serveraction powercycle
    • this takes about 15s
  • connect to the cosole: console com2
    • there might be no output for a little bit immediately after the powercycle. wait at least 30s or so
  • during boot, force netboot: F12
    • when connecting from OSX via the Terminal, escape-shift-2 sends F12
  • you can leave this running to watch it complete
    • This is where you'll be prompted for partition info
  • when you're done, ctrl-\ will disconnect you from the console
    • exit will disconnect from the mgmt interface

Get puppet running

Warning: if you are rebuilding a pre-existing server (rather than a brand new name), on sockpuppet, run puppetca --clean $fqdn to clear out the old certificate before beginning this process.

• get a shell on both $server and the puppet master (sockpuppet)
  • Only one key has access to new installs.
  • from sockpuppet, ssh -i ~/.ssh/new_install root@$servername.wikimedia.org to log into $server
• on $server, run puppetd --test It will fail.
• on sockpuppet, run puppetca -s $server-fqdn
• on $server, run puppetd --test It should now succeed.

Set up puppet

• add $server to site.pp, either by hostname or within a regex if it's part of a class (eg srv\d\d*)
• do whatever puppet goodies you want to get the server to do what you want it to.

Related pages

• DNS
• An older version of the preseeding docs: Automated installation
• Another old page: Automated_upgrade(reinstallation)
• More detail around working with the console: Remote_management