This is a tool for converting a physical computer into a ganeti instance. It consists of two parts, a ganeti instance OS template that allows the instance to be booted to receive the files, and a script that is run on the source machine to make the transfer.
A design document is available in doc/design.rst that describes the functioning of the system. This document will focus on getting the system up and running, and the process for actually performing physical-to-virtual transfers.
The following external programs are used by this package:
The bootstrap OS is a somewhat modified version of the instance-debootstrap OS definition, whose documentation is located in doc/README.debootstrap.rst. The first step will be to install this OS definition onto the nodes of the cluster so that instances can be created.
In order to install this package from source, you need to determine what options ganeti itself has been configured with. If ganeti was built directly from source, then the only place it looks for OS definitions is /srv/ganeti/os, and you need to install the OS under it. Distribute and install the package:
gnt-cluster copyfile /root/ganeti-p2v-transfer-0.1.tar.gz gnt-cluster command "tar xf ganeti-p2v-transfer-0.1.tar.gz && cd ganeti-p2v-transfer-0.1 && ./configure --prefix=/usr --localstatedir=/var \ --sysconfdir=/etc \ --with-os-dir=/srv/ganeti/os && make install-target"
If ganeti was installed from a package, its default OS path should already include /usr/share/ganeti/os, so you can omit --with-os-dir:
gnt-cluster copyfile /root/ganeti-p2v-transfer-0.1.tar.gz gnt-cluster command "tar xf ganeti-p2v-transfer-0.1.tar.gz && cd ganeti-p2v-transfer-0.1 && ./configure --prefix=/usr --localstatedir=/var \ --sysconfdir=/etc && make install-target"
The actual path that ganeti will search for operating system definitions can be determined easily in ganeti 2.4.3 by running gnt-cluster info and looking for the OS search path. In earlier versions, it can be found by looking for a file named _autoconf.py under a ganeti directory in the python modules tree (e.g. /usr/lib/python2.4/site-packages/ganeti/_autoconf.py). In this file, a variable named OS_SEARCH_PATH will list all the directories in which ganeti will look for OS definitions. On of these should be passed to ./configure as the value of --with-os-dir.
Once the package is installed, edit the file /etc/ganeti/instance-p2v-target/p2v-target.conf to uncomment the EXTRA_PKGS value that is appropriate to the hypervisor you are using on this cluster. Then distribute the updated file
Edit /etc/ganeti/instance-p2v-target/p2v-target.conf to uncomment the appropriate value of EXTRA_PKGS. Depending on your setup, you may need to change other values as well. Then, copy the edited file to all nodes:
gnt-cluster copyfile /etc/ganeti/instance-p2v-target/p2v-target.conf
As the bootstrap OS is completely overwritten by the files transferred from the source, it probably doesn't require much configuration. However, all the information in doc/README.debootstrap.rst about configuring the OS is still valid if you need it.
The bootstrap OS must be started with a specialized initrd that moves all files into RAM before finishing the boot. Create this initrd by running:
sudo make_ramboot_initrd.py -V $DEB_KERNEL
on the master node, where $DEB_KERNEL is the name of a kernel that can be used to boot a debootstrap instance. That is, if the filename of the kernel is /boot/vmlinuz-2.6.32-generic, DEB_KERNEL would be "2.6.32-generic". This command should create the file /boot/initrd.img-$DEB_KERNEL-ramboot. Copy this file to the other nodes with:
gnt-cluster copyfile /boot/initrd.img-$DEB_KERNEL-ramboot
The generated initrd depends heavily on the version of initramfs-tools installed on the machine that generates it. As a result, it may not be compatible with the kernel that is to be used for booting the bootstrap OS. In this case, the bootstrap OS may not boot, or may not be able to find the root device. If this happens, a good way to improve compatibility is to use a machine that is already running the instance kernel, perhaps a "normal" (non-p2v) instance on the same cluster. Install and run make_ramboot_initrd.py on this machine to generate the desired initrd.
Users will authenticate to their instances using an SSH keypair generated in advance by the administrator. The public key will be installed into root's .ssh/authorized_keys file on the instance, and the private key will be provided to the user so that they can make the transfer. Generate the keys, with no passphrase, using the commands:
ssh-keygen -t dsa -N "" -f /etc/ganeti/instance-p2v-target/id_dsa gnt-cluster copyfile /etc/ganeti/instance-p2v-target/id_dsa.pub
Keep the private key (/etc/ganeti/instance-p2v-target/id_dsa) somewhere safe, and give it to users who wish to use the P2V system.
Now that the nodes are set up to install instances with the bootstrap OS, we can go ahead and perform a P2V transfer. The first step is to create the instance that will receive the transfer. Create it with the p2v-target+default OS and whatever parameters you need. The default kernel and initrd of the instance should be ones that are both compatible with and installed on the source OS. Also pass the --no-start flag, because we want to use the specially generated initrd for the boot rather than the default one. The command line will look something like the following:
gnt-instance add -t<template> -s<size> -o p2v-target+default \ -n<nodes> --no-start <hostname>
Now boot the instance using the kernel and initrd that work on the initrd:
gnt-instance start -H kernel_path=/boot/vmlinuz-$DEB_KERNEL,\ initrd_path=/boot/initrd.img-$DEB_KERNEL-ramboot <hostname>
Before you begin, you will need the private key corresponding to the public key installed on the instance. Your administrator will provide this to you.
Boot the source machine from a LiveCD or PXE image. Extract the ganeti-p2v-transfer tarball and run:
./configure --prefix=/usr --localstatedir=/var \ --sysconfdir=/etc sudo make install-source
This will install the p2v_transfer.py script. The script requires the following arguments:
Run the script, and your data will be transferred:
sudo p2v_transfer.py $root_dev $target_host $private_key
When the transfer finishes, the script will shut down the instance. When the ganeti watcher restarts it, log in and make sure that everything works.
These instructions suggest building the initrd on a node, for convenience. However, it is possible that there are incompatibilities between the initramfs-tools installed on the node and the kernel that will be used for the bootstrap OS. In this case, the bootstrap OS may not boot, or may not be able to find the root device. If this happens, a good way to improve compatibility is to use a machine that is already running the instance kernel, perhaps a "normal" (non-p2v) instance on the same cluster. Install and run make_ramboot_initrd.py on this machine to generate the desired initrd.
Another possibility is that the bootstrap OS does not have enough RAM to complete its boot. Since the bootstrap OS must be copied entirely into RAM, instances with small memory sizes are not supported. I have had good luck using 768MB of instance memory.