Documents Ganeti version 2.18
Contents
Ganeti is a cluster virtualization management system based on Xen or KVM. This document explains how to bootstrap a Ganeti node (Xen dom0, the host Linux system for KVM), create a running cluster and install virtual instances (Xen domUs, KVM guests). You need to repeat most of the steps in this document for every node you want to install, but of course we recommend creating some semi-automatic procedure if you plan to deploy Ganeti on a medium/large scale.
A basic Ganeti terminology glossary is provided in the introductory section of the Ganeti administrator’s guide. Please refer to that document if you are uncertain about the terms we are using.
Ganeti has been developed for Linux and should be distribution-agnostic. This documentation will use Debian Squeeze as an example system but the examples can be translated to any other distribution. You are expected to be familiar with your distribution, its package management system, and Xen or KVM before trying to use Ganeti.
This document is divided into two main sections:
Each of these is divided into sub-sections. While a full Ganeti system will need all of the steps specified, some are not strictly required for every environment. Which ones they are, and why, is specified in the corresponding sections.
Any system supported by your Linux distribution is fine. 64-bit systems are better as they can support more memory.
Any disk drive recognized by Linux (IDE/SCSI/SATA/etc.) is supported in Ganeti. Note that no shared storage (e.g. SAN) is needed to get high-availability features (but of course, one can be used to store the images). Whilte it is highly recommended to use more than one disk drive in order to improve speed, Ganeti also works with one disk per machine.
Mandatory on all nodes.
It is advised to start with a clean, minimal install of the operating system. The only requirement you need to be aware of at this stage is to partition leaving enough space for a big (minimum 20GiB) LVM volume group which will then host your instance filesystems, if you want to use all Ganeti features. The volume group name Ganeti uses (by default) is xenvg.
You can also use file-based storage only, without LVM, but this setup is not detailed in this document.
If you choose to use RBD-based instances, there’s no need for LVM provisioning. However, this feature is experimental, and is not yet recommended for production clusters.
While you can use an existing system, please note that the Ganeti installation is intrusive in terms of changes to the system configuration, and it’s best to use a newly-installed system without important data on it.
Also, for best results, it’s advised that the nodes have as much as possible the same hardware and software configuration. This will make administration much easier.
Note that Ganeti requires the hostnames of the systems (i.e. what the hostname command outputs to be a fully-qualified name, not a short name. In other words, you should use node1.example.com as a hostname and not just node1.
Debian
Debian usually configures the hostname differently than you need it for Ganeti. For example, this is what it puts in /etc/hosts in certain situations:
127.0.0.1 localhost
127.0.1.1 node1.example.com node1
but for Ganeti you need to have:
127.0.0.1 localhost
192.0.2.1 node1.example.com node1
replacing 192.0.2.1 with your node’s address. Also, the file /etc/hostname which configures the hostname of the system should contain node1.example.com and not just node1 (you need to run the command /etc/init.d/hostname.sh start after changing the file).
Why a fully qualified host name
Although most distributions use only the short name in the /etc/hostname file, we still think Ganeti nodes should use the full name. The reason for this is that calling ‘hostname –fqdn’ requires the resolver library to work and is a ‘guess’ via heuristics at what is your domain name. Since Ganeti can be used among other things to host DNS servers, we don’t want to depend on them as much as possible, and we’d rather have the uname() syscall return the full node name.
We haven’t ever found any breakage in using a full hostname on a Linux system, and anyway we recommend to have only a minimal installation on Ganeti nodes, and to use instances (or other dedicated machines) to run the rest of your network services. By doing this you can change the /etc/hostname file to contain an FQDN without the fear of breaking anything unrelated.
Mandatory on all nodes.
While Ganeti is developed with the ability to modularly run on different virtualization environments in mind the only two currently useable on a live system are Xen and KVM. Supported Xen versions are: 3.0.3 and later 3.x versions, and 4.x (tested up to 4.1). Supported KVM versions are 72 and above.
Please follow your distribution’s recommended way to install and set up Xen, or install Xen from the upstream source, if you wish, following their manual. For KVM, make sure you have a KVM-enabled kernel and the KVM tools.
After installing Xen, you need to reboot into your new system. On some distributions this might involve configuring GRUB appropriately, whereas others will configure it automatically when you install the respective kernels. For KVM no reboot should be necessary.
Xen on Debian
Under Debian you can install the relevant xen-linux-system package, which will pull in both the hypervisor and the relevant kernel. Also, if you are installing a 32-bit system, you should install the libc6-xen package (run apt-get install libc6-xen).
Depending on which toolstack you are using, the hypervisor parameter xen_cmd has to be set to the matching value, either xm or xl.
Some useful best practices for Xen are to restrict the amount of memory dom0 has available, and pin the dom0 to a limited number of CPUs. Instructions for how to achieve this for various toolstacks can be found on the Xen wiki.
It is recommended that you disable Xen’s automatic save of virtual machines at system shutdown and subsequent restore of them at reboot. To obtain this make sure the variable XENDOMAINS_SAVE in the file /etc/default/xendomains is set to an empty value.
You may need to restart the Xen daemon for some of these settings to take effect. The best way to do this depends on your distribution.
After you have installed Xen, you need to tell Ganeti exactly what kernel to use for the instances it will create. This is done by creating a symlink from your actual kernel to /boot/vmlinuz-3-xenU, and one from your initrd to /boot/initrd-3-xenU [1]. Note that if you don’t use an initrd for the domU kernel, you don’t need to create the initrd symlink.
Debian
After installation of the xen-linux-system package, you need to run (replace the exact version number with the one you have):
$ cd /boot
$ ln -s vmlinuz-2.6.26-1-xen-amd64 vmlinuz-3-xenU
$ ln -s initrd.img-2.6.26-1-xen-amd64 initrd-3-xenU
By default, the initrd doesn’t contain the Xen block drivers needed to mount the root device, so it is recommended to update the initrd by following these two steps:
Recommended on all nodes: DRBD is required if you want to use the high availability (HA) features of Ganeti, but optional if you don’t require them or only run Ganeti on single-node clusters. You can upgrade a non-HA cluster to an HA one later, but you might need to convert all your instances to DRBD to take advantage of the new features.
Supported DRBD versions: 8.0-8.4. It’s recommended to have at least version 8.0.12. Note that for version 8.2 and newer it is needed to pass the usermode_helper=/bin/true parameter to the module, either by configuring /etc/modules or when inserting it manually. When using Xen and DRBD 8.3.2 or higher, it is recommended to use the disable_sendpage=1 setting as well.
Now the bad news: unless your distribution already provides it installing DRBD might involve recompiling your kernel or anyway fiddling with it. Hopefully at least the Xen-ified kernel source to start from will be provided (if you intend to use Xen).
The good news is that you don’t need to configure DRBD at all. Ganeti will do it for you for every instance you set up. If you have the DRBD utils installed and the module in your kernel you’re fine. Please check that your system is configured to load the module at every boot, and that it passes the following option to the module: minor_count=NUMBER. We recommend that you use 128 as the value of the minor_count - this will allow you to use up to 64 instances in total per node (both primary and secondary, when using only one disk per instance). You can increase the number up to 255 if you need more instances on a node.
Debian
On Debian, you can just install (build) the DRBD module with the following commands, making sure you are running the target (Xen or KVM) kernel:
$ apt-get install drbd8-source drbd8-utils
$ m-a update
$ m-a a-i drbd8
Or on newer versions, if the kernel already has modules:
$ apt-get install drbd8-utils
Then to configure it for Ganeti:
$ echo "options drbd minor_count=128 usermode_helper=/bin/true" \
> /etc/modprobe.d/drbd.conf
$ echo "drbd" >> /etc/modules
$ depmod -a
$ modprobe drbd
It is also recommended that you comment out the default resources (if any) in the /etc/drbd.conf file, so that the init script doesn’t try to configure any drbd devices. You can do this by prefixing all resource lines in the file with the keyword skip, like this:
skip {
resource r0 {
...
}
}
skip {
resource "r1" {
...
}
}
Recommended on all nodes: RBD is required if you want to create instances with RBD disks residing inside a RADOS cluster (make use of the rbd disk template). RBD-based instances can failover or migrate to any other node in the ganeti cluster, enabling you to exploit of all Ganeti’s high availabilily (HA) features.
Attention
Be careful though: rbd is still experimental! For now it is recommended only for testing purposes. No sensitive data should be stored there.
You will need the rbd and libceph kernel modules, the RBD/Ceph userspace utils (ceph-common Debian package) and an appropriate Ceph/RADOS configuration file on every VM-capable node.
You will also need a working RADOS Cluster accessible by the above nodes.
You will need a working RADOS Cluster accesible by all VM-capable nodes to use the RBD template. For more information on setting up a RADOS Cluster, refer to the official docs.
If you want to use a pool for storing RBD disk images other than the default (rbd), you should first create the pool in the RADOS Cluster, and then set the corresponding rbd disk parameter named pool.
Unless your distribution already provides it, you might need to compile the rbd and libceph modules from source. You will need Linux Kernel 3.2 or above for the kernel modules. Alternatively you will have to build them as external modules (from Linux Kernel source 3.2 or above), if you want to run a less recent kernel, or your kernel doesn’t include them.
The RBD template has been tested with ceph-common v0.38 and above. We recommend using the latest version of ceph-common.
Debian
On Debian, you can just install the RBD/Ceph userspace utils with the following command:
$ apt-get install ceph-common
If your cluster uses a sufficiently new version of KVM (you will need at least QEMU 0.14 with RBD support compiled in), you can take advantage of KVM’s native support for ceph in order to have better performance and avoid potential deadlocks in low memory scenarios.
To initialize a cluster with support for this feature, use a command like the following. Note, that you possibly need to follow the more general installation instructions before invoking this command (see Initializing the cluster ).
- $ gnt-cluster init
- –enabled-disk-templates=rbd –ipolicy-disk-templates=rbd –enabled-hypervisors=kvm -D rbd:access=userspace
(You may want to enable more templates than just rbd.)
You can also change this setting on a live cluster by giving the same switches to gnt-cluster modify, or change those settings at the node group level with gnt-group modify.
You should also provide an appropriate configuration file (ceph.conf) in /etc/ceph. For the rbd userspace utils, you’ll only need to specify the IP addresses of the RADOS Cluster monitors.
ceph.conf
Sample configuration file:
[mon.a]
host = example_monitor_host1
mon addr = 1.2.3.4:6789
[mon.b]
host = example_monitor_host2
mon addr = 1.2.3.5:6789
[mon.c]
host = example_monitor_host3
mon addr = 1.2.3.6:6789
For more information, please see the Ceph Docs
For Gluster integration, Ganeti requires that mount.glusterfs is installed on each and every node. On Debian Wheezy and newer, you can satisfy this requirement with the glusterfs-client package; see this guide for details.
If your cluster uses a sufficiently new version of KVM (you will need at least QEMU 1.3 with Gluster support compiled in), you can take advantage of KVM’s native support for gluster in order to have better performance and avoid potential deadlocks in low memory scenarios.
Please be aware that QEMU 1.3 was released in December 3, 2012, and as such this feature is not available out of the box in any distribution older than Ubuntu 13.04; this excludes Ubuntu 12.04 LTS and Debian Wheezy.
Please install all software requirements mentioned in Ganeti quick installation guide. If you want to build Ganeti from source, don’t forget to follow the steps required for that as well.
Mandatory on all nodes.
You can run Ganeti either in “bridged mode”, “routed mode” or “openvswitch mode”. In bridged mode, the default, the instances network interfaces will be attached to a software bridge running in dom0. Xen by default creates such a bridge at startup, but your distribution might have a different way to do things, and you’ll definitely need to manually set it up under KVM.
Beware that the default name Ganeti uses is xen-br0 (which was used in Xen 2.0) while Xen 3.0 uses xenbr0 by default. See the Initializing the cluster section to learn how to choose a different bridge, or not to use one at all and use “routed mode”.
In order to use “routed mode” under Xen, you’ll need to change the relevant parameters in the Xen config file. Under KVM instead, no config change is necessary, but you still need to set up your network interfaces correctly.
By default, under KVM, the “link” parameter you specify per-nic will represent, if non-empty, a different routing table name or number to use for your instances. This allows isolation between different instance groups, and different routing policies between node traffic and instance traffic.
You will need to configure your routing table basic routes and rules outside of ganeti. The vif scripts will only add /32 routes to your instances, through their interface, in the table you specified (under KVM, and in the main table under Xen).
Also for “openvswitch mode” under Xen a custom network script is needed. Under KVM everything should work, but you’ll need to configure your switches outside of Ganeti (as for bridges).
Bridging issues with certain kernels
Some kernel versions (e.g. 2.6.32) have an issue where the bridge will automatically change its MAC address to the lower-numbered slave on port addition and removal. This means that, depending on the MAC address of the actual NIC on the node and the addresses of the instances, it could be that starting, stopping or migrating instances will lead to timeouts due to the address of the bridge (and thus node itself) changing.
To prevent this, it’s enough to set the bridge manually to a specific MAC address, which will disable this automatic address change. In Debian, this can be done as follows in the bridge configuration snippet:
up ip link set addr $(cat /sys/class/net/$IFACE/address) dev $IFACE
which will “set” the bridge address to the initial one, disallowing changes.
Bridging under Debian
The recommended way to configure the Xen bridge is to edit your /etc/network/interfaces file and substitute your normal Ethernet stanza with the following snippet:
auto xen-br0
iface xen-br0 inet static
address YOUR_IP_ADDRESS
netmask YOUR_NETMASK
network YOUR_NETWORK
broadcast YOUR_BROADCAST_ADDRESS
gateway YOUR_GATEWAY
bridge_ports eth0
bridge_stp off
bridge_fd 0
# example for setting manually the bridge address to the eth0 NIC
up ip link set addr $(cat /sys/class/net/eth0/address) dev $IFACE
The following commands need to be executed on the local console:
$ ifdown eth0
$ ifup xen-br0
To check if the bridge is setup, use the ip and brctl show commands:
$ ip a show xen-br0
9: xen-br0: <BROADCAST,MULTICAST,UP,10000> mtu 1500 qdisc noqueue
link/ether 00:20:fc:1e:d5:5d brd ff:ff:ff:ff:ff:ff
inet 10.1.1.200/24 brd 10.1.1.255 scope global xen-br0
inet6 fe80::220:fcff:fe1e:d55d/64 scope link
valid_lft forever preferred_lft forever
$ brctl show xen-br0
bridge name bridge id STP enabled interfaces
xen-br0 8000.0020fc1ed55d no eth0
In order to have a custom and more advanced networking configuration in Xen which can vary among instances, after having successfully installed Ganeti you have to create a symbolic link to the vif-script provided by Ganeti inside /etc/xen/scripts (assuming you installed Ganeti under /usr/lib):
$ ln -s /usr/lib/ganeti/vif-ganeti /etc/xen/scripts/vif-ganeti
This has to be done on all nodes. Afterwards you can set the vif_script hypervisor parameter to point to that script by:
$ gnt-cluster modify -H xen-pvm:vif_script=/etc/xen/scripts/vif-ganeti
Having this hypervisor parameter you are able to create your own scripts and create instances with different networking configurations.
Mandatory on all nodes.
The volume group is required to be at least 20GiB.
If you haven’t configured your LVM volume group at install time you need to do it before trying to initialize the Ganeti cluster. This is done by formatting the devices/partitions you want to use for it and then adding them to the relevant volume group:
$ pvcreate /dev/sda3
$ vgcreate xenvg /dev/sda3
or:
$ pvcreate /dev/sdb1
$ pvcreate /dev/sdc1
$ vgcreate xenvg /dev/sdb1 /dev/sdc1
If you want to add a device later you can do so with the vgextend command:
$ pvcreate /dev/sdd1
$ vgextend xenvg /dev/sdd1
Optional: it is recommended to configure LVM not to scan the DRBD devices for physical volumes. This can be accomplished by editing /etc/lvm/lvm.conf and adding the /dev/drbd[0-9]+ regular expression to the filter variable, like this:
filter = ["r|/dev/cdrom|", "r|/dev/drbd[0-9]+|" ]
Note that with Ganeti a helper script is provided - lvmstrap which will erase and configure as LVM any not in-use disk on your system. This is dangerous and it’s recommended to read its --help output if you want to use it.
Mandatory on all nodes.
It’s now time to install the Ganeti software itself. Download the source from the project page at http://downloads.ganeti.org/releases/, and install it (replace 2.6.0 with the latest version):
$ tar xvzf ganeti-2.6.0.tar.gz
$ cd ganeti-2.6.0
$ ./configure --localstatedir=/var --sysconfdir=/etc
$ make
$ make install
$ mkdir /srv/ganeti/ /srv/ganeti/os /srv/ganeti/export
You also need to copy the file doc/examples/ganeti.initd from the source archive to /etc/init.d/ganeti and register it with your distribution’s startup scripts, for example in Debian:
$ chmod +x /etc/init.d/ganeti
$ update-rc.d ganeti defaults 20 80
In order to automatically restart failed instances, you need to setup a cron job run the ganeti-watcher command. A sample cron file is provided in the source at doc/examples/ganeti.cron and you can copy that (eventually altering the path) to /etc/cron.d/ganeti. Finally, a sample logrotate snippet is provided in the source at doc/examples/ganeti.logrotate and you can copy it to /etc/logrotate.d/ganeti to have Ganeti’s logs rotated automatically.
The above make install invocation, or installing via your distribution mechanisms, will install on the system:
Mandatory on all nodes.
To be able to install instances you need to have an Operating System installation script. An example OS that works under Debian and can install Debian and Ubuntu instace OSes is provided on the project web site. Download it from the project page and follow the instructions in the README file. Here is the installation procedure (replace 0.14 with the latest version that is compatible with your ganeti version):
$ cd /usr/local/src/
$ wget http://ganeti.googlecode.com/files/ganeti-instance-debootstrap-0.14.tar.gz
$ tar xzf ganeti-instance-debootstrap-0.14.tar.gz
$ cd ganeti-instance-debootstrap-0.14
$ ./configure --with-os-dir=/srv/ganeti/os
$ make
$ make install
In order to use this OS definition, you need to have internet access from your nodes and have the debootstrap, dump and restore commands installed on all nodes. Also, if the OS is configured to partition the instance’s disk in /etc/default/ganeti-instance-debootstrap, you will need kpartx installed.
Debian
Use this command on all nodes to install the required packages:
$ apt-get install debootstrap dump kpartx
Or alternatively install the OS definition from the Debian package:
$ apt-get install ganeti-instance-debootstrap
KVM
In order for debootstrap instances to be able to shutdown cleanly they must install have basic ACPI support inside the instance. Which packages are needed depend on the exact flavor of Debian or Ubuntu which you’re installing, but the example defaults file has a commented out configuration line that works for Debian Lenny and Squeeze:
EXTRA_PKGS="acpi-support-base,console-tools,udev"
kbd can be used instead of console-tools, and more packages can be added, of course, if needed.
Please refer to the README file of ganeti-instance-debootstrap for further documentation.
Alternatively, you can create your own OS definitions. See the manpage ganeti-os-interface(7).
Mandatory once per cluster, on the first node.
The last step is to initialize the cluster. After you have repeated the above process on all of your nodes and choose one as the master. Make sure there is a SSH key pair on the master node (optionally generating one using ssh-keygen). Finally execute:
$ gnt-cluster init CLUSTERNAME
The CLUSTERNAME is a hostname, which must be resolvable (e.g. it must exist in DNS or in /etc/hosts) by all the nodes in the cluster. You must choose a name different from any of the nodes names for a multi-node cluster. In general the best choice is to have a unique name for a cluster, even if it consists of only one machine, as you will be able to expand it later without any problems. Please note that the hostname used for this must resolve to an IP address reserved exclusively for this purpose, and cannot be the name of the first (master) node.
If you want to use a bridge which is not xen-br0, or no bridge at all, change it with the --nic-parameters option. For example to bridge on br0 you can add:
--nic-parameters link=br0
Or to not bridge at all, and use a separate routing table:
--nic-parameters mode=routed,link=100
If you don’t have a xen-br0 interface you also have to specify a different network interface which will get the cluster IP, on the master node, by using the --master-netdev <device> option.
You can use a different name than xenvg for the volume group (but note that the name must be identical on all nodes). In this case you need to specify it by passing the –vg-name <VGNAME> option to gnt-cluster init.
To set up the cluster as an Xen HVM cluster, use the --enabled-hypervisors=xen-hvm option to enable the HVM hypervisor (you can also add ,xen-pvm to enable the PVM one too). You will also need to create the VNC cluster password file /etc/ganeti/vnc-cluster-password which contains one line with the default VNC password for the cluster.
To setup the cluster for KVM-only usage (KVM and Xen cannot be mixed), pass --enabled-hypervisors=kvm to the init command.
You can also invoke the command with the --help option in order to see all the possibilities.
Please note that the default hypervisor/network/cluster parameters may not be the correct one for your environment. Carefully check them, and change them either at cluster init time, or later with gnt-cluster modify.
Your instance types, networking environment, hypervisor type and version may all affect what kind of parameters should be used on your cluster.
KVM
Instances are by default configured to use a host kernel, and to be reached via serial console, which works nice for Linux paravirtualized instances. If you want fully virtualized instances you may want to handle their kernel inside the instance, and to use VNC.
Some versions of KVM have a bug that will make an instance hang when configured to use the serial console (which is the default) unless a connection is made to it within about 2 seconds of the instance’s startup. For such case it’s recommended to disable the serial_console option.
Mandatory for all the other nodes.
After you have initialized your cluster you need to join the other nodes to it. You can do so by executing the following command on the master node:
$ gnt-node add NODENAME
Optional
Ganeti uses DRBD to mirror the disk of the virtual instances between nodes. To use a dedicated network interface for this (in order to improve performance or to enhance security) you need to configure an additional interface for each node. Use the -s option with gnt-cluster init and gnt-node add to specify the IP address of this secondary interface to use for each node. Note that if you specified this option at cluster setup time, you must afterwards use it for every node add operation.
Execute the gnt-node list command to see all nodes in the cluster:
$ gnt-node list
Node DTotal DFree MTotal MNode MFree Pinst Sinst
node1.example.com 197404 197404 2047 1896 125 0 0
The above shows a couple of things:
With Ganeti a tool called burnin is provided that can test most of the Ganeti functionality. The tool is installed under the lib/ganeti/tools directory (either under /usr or /usr/local based on the installation method). See more details under burnin.
You can now proceed either to the Ganeti administrator’s guide, or read the manpages of the various commands (ganeti(7), gnt-cluster(8), gnt-node(8), gnt-instance(8), gnt-job(8)).
Footnotes
| [1] | The kernel and initrd paths can be changed at either cluster level (which changes the default for all instances) or at instance level. |