How Ganeti works - The Basics

This document is to help give you a basic understanding of how Ganeti works. If you’d like to see more detailed information, please visit their upstream docs site.

Ganeti is written primarily in python. Each node runs several daemons which take care of various tasks for Ganeti such as config syncing, instance creation, lvm, etc. Every Ganeti cluster must have one node act as a master. This master is the only node allowed to run cluster-wide commands such as adding new instances. Each node can be a master (if setup as a master candidate) and can become a master.

Communication between the nodes happens via an encrypted connection using a restful API. It currently only supports Xen, KVM for HVM backends and some limited support for LXC. The most common setup using using LVM with DRBD to have a cluster failover solution. It also has support for plain LVM, and file-based although is less tested. Ganeti handles all of the storage backend setup for LVM and DRBD so you don’t need to worry about that.

All of the remote commands are executed via ssh and require the same ssh key on every node.

Managing the Cluster

Managing all aspects of the cluster is done via gnt- prefixed commands on the master node. The ssh host key is copied between all the nodes so that there should be no warning when you ssh to the host. Here’s a list of commands you can use to manage the cluster:




managing cluster-wide actions


manage node-level actions


manage instance (VM) level actions


manage import/export actions for instances


list all operating system types and variants


show and manage ganeti jobs

All commands are well documented in both man pages and command line based help. You can get help on every command by doing:

gnt-$command $action --help

When you call a command, Ganeti will communicate to the other nodes it needs to via the daemons, gather up information about the node, verify several factors, and then create a job to complete the task. Logs can be found in /var/log/ganeti/. You should look on the master node for more logs however you may need to look at node specific logs during instance creation, etc.

Some useful common commands are documented here in more detail.

Instance (VMs) Management

Ganeti uses an OS instance definition to create new VMs. The most common instance definition is instance-debootstrap which uses debootstrap to create a debian-based install. We have written an image based instance definition called ganeti-instance-image which lets you use either a tarball, filesystem dump, or qemu-img file for OS deployments. This allows us to deploy a diverse environment using upstream installation methods. You need to install the same instance definition on ALL nodes otherwise Ganeti will complain.

The instance definitions almost all use bash, but they could be written in any language. They have four actions:

  • create – create a new instance

  • rename – rename an instance

  • export – export an instance

  • import – import an instance

You can add simple post-install scripts so that you can do things like setup cfengine.

Exporting/Importing the instances is primarily used when you want to move between clusters. Most are just tarball filesystem dumps with meta information about the VM.