ganeti - cluster-based virtualization management


# gnt-cluster init
# gnt-node add
# gnt-instance add -n \
> -o debootstrap --disk 0:size=30g \
> -t plain


The Ganeti software manages physical nodes and virtual instances of a cluster based on a virtualization software. The current version (2.3) supports Xen 3.x and KVM (72 or above) as hypervisors, and LXC as an experimental hypervisor.

Quick start

First you must install the software on all the cluster nodes, either from sources or (if available) from a package. The next step is to create the initial cluster configuration, using gnt-cluster init.

Then you can add other nodes, or start creating instances.

Cluster architecture

In Ganeti 2.0, the architecture of the cluster is a little more complicated than in 1.2. The cluster is coordinated by a master daemon (ganeti-masterd(8)), running on the master node. Each node runs (as before) a node daemon, and the master has the RAPI daemon running too.

Node roles

Each node can be in one of the following states:


Only one node per cluster can be in this role, and this node is the one holding the authoritative copy of the cluster configuration and the one that can actually execute commands on the cluster and modify the cluster state. See more details under Cluster configuration.


The node receives the full cluster configuration (configuration file and jobs) and can become a master via the gnt-cluster master-failover command. Nodes that are not in this state cannot transition into the master role due to missing state.


This the normal state of a node.


Nodes in this state are functioning normally but cannot receive new instances, because the intention is to set them to offline or remove them from the cluster.


These nodes are still recorded in the Ganeti configuration, but except for the master daemon startup voting procedure, they are not actually contacted by the master. This state was added in order to allow broken machines (that are being repaired) to remain in the cluster but without creating problems.

Node flags

Nodes have two flags which govern which roles they can take:


The node can become a master candidate, and furthermore the master node. When this flag is disabled, the node cannot become a candidate; this can be useful for special networking cases, or less reliable hardware.


The node can host instances. When enabled (the default state), the node will participate in instance allocation, capacity calculation, etc. When disabled, the node will be skipped in many cluster checks and operations.

Node Parameters

The ndparams refer to node parameters. These can be set as defaults on cluster and node group levels, but they take effect for nodes only.

Currently we support the following node parameters:


Path to an executable used as the out-of-band helper as described in the Ganeti Node OOB Management Framework design document.


This should reflect the I/O performance of local attached storage (e.g. for "file", "plain" and "drbd" disk templates). It doesn't have to match the actual spindle count of (any eventual) mechanical hard-drives, its meaning is site-local and just the relative values matter.


When this Boolean flag is enabled, physical disks on the node are assigned to instance disks in an exclusive manner, so as to lower I/O interference between instances. See the Partitioned Ganeti design document for more details. This parameter cannot be set on individual nodes, as its value must be the same within each node group.

Hypervisor State Parameters

Using --hypervisor-state you can set hypervisor specific states as pointed out in Ganeti Resource Model <design-resource-model.rst>.

The format is: hypervisor:option=value.

Currently we support the following hypervisor state values:


Total node memory, as discovered by this hypervisor


Memory used by, or reserved for, the node itself; note that some hypervisors can report this in an authoritative way, other not


Memory used either by the hypervisor itself or lost due to instance allocation rounding; usually this cannot be precisely computed, but only roughly estimated


Total node cpu (core) count; usually this can be discovered automatically


Number of cores reserved for the node itself; this can either be discovered or set manually. Only used for estimating how many VCPUs are left for instances

Note that currently this option is unused by Ganeti; values will be recorded but will not influence the Ganeti operation.

Disk State Parameters

Using --disk-state you can set disk specific states as pointed out in Ganeti Resource Model <design-resource-model.rst>.

The format is: storage_type/identifier:option=value. Where we currently just support lvm as storage type. The identifier in this case is the LVM volume group. By default this is xenvg.

Currently we support the following hypervisor state values:


Total disk size (usually discovered automatically)


Reserved disk size; this is a lower limit on the free space, if such a limit is desired


Disk that is expected to be used by other volumes (set via reserved_lvs); usually should be zero

Note that currently this option is unused by Ganeti; values will be recorded but will not influence the Ganeti operation.

Cluster configuration

The master node keeps and is responsible for the cluster configuration. On the filesystem, this is stored under the /var/ganeti/lib directory, and if the master daemon is stopped it can be backed up normally.

The master daemon will replicate the configuration database called and the job files to all the nodes in the master candidate role. It will also distribute a copy of some configuration values via the ssconf files, which are stored in the same directory and start with a ssconf_ prefix, to all nodes.


All cluster modification are done via jobs. A job consists of one or more opcodes, and the list of opcodes is processed serially. If an opcode fails, the entire job is failed and later opcodes are no longer processed. A job can be in one of the following states:


The job has been submitted but not yet processed by the master daemon.


The job is waiting for for locks before the first of its opcodes.


The job is waiting for locks, but is has been marked for cancellation. It will not transition to running, but to canceled.


The job is currently being executed.


The job has been canceled before starting execution.


The job has finished successfully.


The job has failed during runtime, or the master daemon has been stopped during the job execution.

Common command line features


Many Ganeti commands provide the following options. The availability for a certain command can be checked by calling the command using the --help option.

gnt-... command [--dry-run] [--priority {low | normal | high}]

The --dry-run option can be used to check whether an operation would succeed.

The option --priority sets the priority for opcodes submitted by the command.

The --submit option is used to send the job to the master daemon but not wait for its completion. The job ID will be shown so that it can be examined using gnt-job info.


For certain commands you can use environment variables to provide default command line arguments. Just assign the arguments as a string to the corresponding environment variable. The format of that variable name is binary_command. binary is the name of the gnt-* script all upper case and dashes replaced by underscores, and command is the command invoked on that script.

Currently supported commands are gnt-node list, gnt-group list and gnt-instance list. So you can configure default command line flags by setting GNT_NODE_LIST, GNT_GROUP_LIST and GNT_INSTANCE_LIST.

Field formatting

Multiple ganeti commands use the same framework for tabular listing of resources (e.g. gnt-instance list, gnt-node list, gnt-group list, gnt-debug locks, etc.). For these commands, special states are denoted via a special symbol (in terse mode) or a string (in verbose mode):

*, (offline)

The node in question is marked offline, and thus it cannot be queried for data. This result is persistent until the node is de-offlined.

?, (nodata)

Ganeti expected to receive an answer from this entity, but the cluster RPC call failed and/or we didn't receive a valid answer; usually more information is available in the node daemon log (if the node is alive) or the master daemon log. This result is transient, and re-running command might return a different result.

-, (unavail)

The respective field doesn't make sense for this entity; e.g. querying a down instance for its current memory 'live' usage, or querying a non-vm_capable node for disk/memory data. This result is persistent, and until the entity state is changed via ganeti commands, the result won't change.

??, (unknown)

This field is not known (note that this is different from entity being unknown). Either you have mis-typed the field name, or you are using a field that the running Ganeti master daemon doesn't know. This result is persistent, re-running the command won't change it.

Key-value parameters

Multiple options take parameters that are of the form key=value,key=value,... or category:key=value,.... Examples are the hypervisor parameters, backend parameters, etc. For these, it's possible to use values that contain commas by escaping with via a backslash (which needs two if not single-quoted, due to shell behaviour):

# gnt-instance modify -H kernel_path=an\\,example instance1
# gnt-instance modify -H kernel_path='an\,example' instance1

Query filters

Most commands listing resources (e.g. instances or nodes) support filtering. The filter language is similar to Python expressions with some elements from Perl. The language is not generic. Each condition must consist of a field name and a value (except for boolean checks), a field can not be compared to another field. Keywords are case-sensitive.

Examples (see below for syntax details):

Syntax in pseudo-BNF:

<quoted-string> ::= /* String quoted with single or double quotes,
                       backslash for escaping */

<integer> ::= /* Number in base-10 positional notation */

<re> ::= /* Regular expression */

  Modifier "i": Case-insensitive matching, see

  Modifier "s": Make the "." special character match any character,
  including newline, see
<re-modifiers> ::= /* empty */ | i | s

<value> ::= <quoted-string> | <integer>

<condition> ::=
  { /* Value comparison */
    <field> { == | != | < | <= | >= | > } <value>

    /* Collection membership */
    | <value> [ not ] in <field>

    /* Regular expressions (recognized delimiters
       are "/", "#", "^", and "|"; backslash for escaping)
    | <field> { =~ | !~ } m/<re>/<re-modifiers>

    /* Globbing */
    | <field> { =* | !* } <quoted-string>

    /* Boolean */
    | <field>

<filter> ::=
  { [ not ] <condition> | ( <filter> ) }
  [ { and | or } <filter> ]







Less than


Less than or equal


Greater than


Greater than or equal


Pattern match using regular expression


Logically negated from =~


Globbing, see glob(7), though only * and ? are supported


Logically negated from =*

in, not in

Collection membership and negation

Common daemon functionality

All Ganeti daemons re-open the log file(s) when sent a SIGHUP signal. logrotate(8) can be used to rotate Ganeti's log files.


Report bugs to project website or contact the developers using the Ganeti mailing list.


Ganeti overview and specifications: ganeti(7) (general overview), ganeti-os-interface(7) (guest OS definitions), ganeti-extstorage-interface(7) (external storage providers).

Ganeti commands: gnt-cluster(8) (cluster-wide commands), gnt-job(8) (job-related commands), gnt-node(8) (node-related commands), gnt-instance(8) (instance commands), gnt-os(8) (guest OS commands), gnt-storage(8) (storage commands), gnt-group(8) (node group commands), gnt-backup(8) (instance import/export commands), gnt-debug(8) (debug commands).

Ganeti daemons: ganeti-watcher(8) (automatic instance restarter), ganeti-cleaner(8) (job queue cleaner), ganeti-noded(8) (node daemon), ganeti-masterd(8) (master daemon), ganeti-rapi(8) (remote API daemon).

Ganeti htools: htools(1) (generic binary), hbal(1) (cluster balancer), hspace(1) (capacity calculation), hail(1) (IAllocator plugin), hscan(1) (data gatherer from remote clusters), hinfo(1) (cluster information printer), mon-collector(7) (data collectors interface).


Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 Google Inc. Permission is granted to copy, distribute and/or modify under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

On Debian systems, the complete text of the GNU General Public License can be found in /usr/share/common-licenses/GPL.