Ensure both nodes are up.

If you planning to use the secondary network for SAN and DRBD synchronization, you
should configure secondary IP interfaces manually on both nodes at this time.

Log in to the first node via ssh. Due to lack of DNS there may be
a minute timeout before password prompt.


Edit /etc/sci/sci.conf

  • You should specify node1 and node2 data as you have installed them.
    NOTE: You can setup the cluster even with one node. In this case just leave NODE2_
    lines as is. In fact this is a dangerous setup, so you will be warned about this duging
    the procedures.
  • You should specify the cluster's name and IP.
  • NODE#_SAN_IP should be specified on both nodes or none.
  • If you haven't Internet uplink or have a local package mirrors, you should correct
    APT_ - settings.
  • If you need to uplink to the DNS hierarchy other than root hint zones, specify DNS_FORWARDERS
    (note trailing ';').



# sci-setup cluster

Check and confirm settings printed.

The process will go on.

Next you will be prompted to accept ssh key from node2 and for the root's password to node2.

On finish you will look something like this:

Wed Jan 12 15:36:10 2011 * Verifying global settings
Wed Jan 12 15:36:10 2011 * Gathering data (1 nodes)
Wed Jan 12 15:36:11 2011 * Verifying node status
Wed Jan 12 15:36:11 2011 * Verifying instance status
Wed Jan 12 15:36:11 2011 * Verifying orphan volumes
Wed Jan 12 15:36:11 2011 * Verifying orphan instances
Wed Jan 12 15:36:11 2011 * Verifying N+1 Memory redundancy
Wed Jan 12 15:36:11 2011 * Other Notes
Wed Jan 12 15:36:11 2011 * Hooks Results
Node                    DTotal  DFree MTotal MNode MFree Pinst Sinst 100.0G 100.0G  1020M  379M  625M     0     0 100.0G 100.0G  1020M  379M  625M     0     0
If all is ok, proceed with /usr/local/sbin/sci-setup service


The service instance is named 'sci' and have a few aliases.
On setup, it's IP address is determined from /etc/resolv.conf of your first node.
This instance will be hardcoded in /etc/hosts file of all cluster nodes and instances.


# sci-setup service

You'll see the progress of DRBD syncing disks, then the message

* running the instance OS create scripts...

appears. The further may take a while. The process finishes with
* starting instance...


Now you can log on to the sci instance using:

# gnt-instance console sci

Log in as root, the password is empty.
NOTE: Due to empty password all remote connections to new instance is prohibited.
You should change password and install openssh-server package manually after
successful bootstrap procedure.


The system will setup itself via puppet. This is the iterative process. You can monitor
it by looking into /var/log/daemon.log. At start there is no less command yet, so
you can use more, cat, tail or tail -f until less will be auto-installed.

By default the iterations are repeated in 20 minutes. To shorten the wait time you can

# /etc/init.d/puppet restart

and then look into daemon.log how it finishes.

Repeat this a few times until puppet will do nothing in turn.


New instances are created just by regular Ganeti commands such as:

gnt-instance add -t drbd -o debootstrap+default -s 10g -B memory=256m -n NODE1_NAME:NODE2_NAME INSTANCE_NAME
Altought, some tuning hooks are provided by SCI-CD project:
  1. Each instance has installed puppet for autoconfiguration and openssh-client for file transfers etc.
  2. The instance uses pygrub to boot kernel from /vmlinuz & Co on the innstance's own disk.
  3. The instance's network interfaces may be set up automatically as described below.


If your instances may sit on several networks and you need static addressing in them, you should fulfill
the file /etc/ganeti/networks with all known networks you want to attach your instances.
Each line in the file has format


Ganeti instance debootstrap hook looks in this file for the network, mathing the address of bootstraped
instance and fulfill it's /etc/network/interfaces accordingly.

NOTE: If you have only one default network, you shouldn't care because it's data are preinstalled.
NOTE: networks file must be copied to all cluster nodes (not automated yet).