cluster: network: clarify and ease latency requirements

[pve-docs.git] / pvecm.adoc

[[chapter_pvecm]]
ifdef::manvolnum[]
pvecm(1)
========
:pve-toplevel:

NAME
----

pvecm - Proxmox VE Cluster Manager

SYNOPSIS
--------

include::pvecm.1-synopsis.adoc[]

DESCRIPTION
-----------
endif::manvolnum[]

ifndef::manvolnum[]
Cluster Manager
===============
:pve-toplevel:
endif::manvolnum[]

The {pve} cluster manager `pvecm` is a tool to create a group of
physical servers. Such a group is called a *cluster*. We use the
http://www.corosync.org[Corosync Cluster Engine] for reliable group
communication. There's no explicit limit for the number of nodes in a cluster.
In practice, the actual possible node count may be limited by the host and
network performance. Currently (2021), there are reports of clusters (using
high-end enterprise hardware) with over 50 nodes in production.

`pvecm` can be used to create a new cluster, join nodes to a cluster,
leave the cluster, get status information, and do various other cluster-related
tasks. The **P**rox**m**o**x** **C**luster **F**ile **S**ystem (``pmxcfs'')
is used to transparently distribute the cluster configuration to all cluster
nodes.

Grouping nodes into a cluster has the following advantages:

* Centralized, web-based management

* Multi-master clusters: each node can do all management tasks

* Use of `pmxcfs`, a database-driven file system, for storing configuration
  files, replicated in real-time on all nodes using `corosync`

* Easy migration of virtual machines and containers between physical
  hosts

* Fast deployment

* Cluster-wide services like firewall and HA


Requirements
------------

* All nodes must be able to connect to each other via UDP ports 5404 and 5405
 for corosync to work.

* Date and time must be synchronized.

* An SSH tunnel on TCP port 22 between nodes is required.

* If you are interested in High Availability, you need to have at
  least three nodes for reliable quorum. All nodes should have the
  same version.

* We recommend a dedicated NIC for the cluster traffic, especially if
  you use shared storage.

* The root password of a cluster node is required for adding nodes.

* Online migration of virtual machines is only supported when nodes have CPUs
  from the same vendor. It might work otherwise, but this is never guaranteed.

NOTE: It is not possible to mix {pve} 3.x and earlier with {pve} 4.X cluster
nodes.

NOTE: While it's possible to mix {pve} 4.4 and {pve} 5.0 nodes, doing so is
not supported as a production configuration and should only be done temporarily,
during an upgrade of the whole cluster from one major version to another.

NOTE: Running a cluster of {pve} 6.x with earlier versions is not possible. The
cluster protocol (corosync) between {pve} 6.x and earlier versions changed
fundamentally. The corosync 3 packages for {pve} 5.4 are only intended for the
upgrade procedure to {pve} 6.0.


Preparing Nodes
---------------

First, install {pve} on all nodes. Make sure that each node is
installed with the final hostname and IP configuration. Changing the
hostname and IP is not possible after cluster creation.

While it's common to reference all node names and their IPs in `/etc/hosts` (or
make their names resolvable through other means), this is not necessary for a
cluster to work. It may be useful however, as you can then connect from one node
to another via SSH, using the easier to remember node name (see also
xref:pvecm_corosync_addresses[Link Address Types]). Note that we always
recommend referencing nodes by their IP addresses in the cluster configuration.


[[pvecm_create_cluster]]
Create a Cluster
----------------

You can either create a cluster on the console (login via `ssh`), or through
the API using the {pve} web interface (__Datacenter -> Cluster__).

NOTE: Use a unique name for your cluster. This name cannot be changed later.
The cluster name follows the same rules as node names.

[[pvecm_cluster_create_via_gui]]
Create via Web GUI
~~~~~~~~~~~~~~~~~~

[thumbnail="screenshot/gui-cluster-create.png"]

Under __Datacenter -> Cluster__, click on *Create Cluster*. Enter the cluster
name and select a network connection from the drop-down list to serve as the
main cluster network (Link 0). It defaults to the IP resolved via the node's
hostname.

As of {pve} 6.2, up to 8 fallback links can be added to a cluster. To add a
redundant link, click the 'Add' button and select a link number and IP address
from the respective fields. Prior to {pve} 6.2, to add a second link as
fallback, you can select the 'Advanced' checkbox and choose an additional
network interface (Link 1, see also xref:pvecm_redundancy[Corosync Redundancy]).

NOTE: Ensure that the network selected for cluster communication is not used for
any high traffic purposes, like network storage or live-migration.
While the cluster network itself produces small amounts of data, it is very
sensitive to latency. Check out full
xref:pvecm_cluster_network_requirements[cluster network requirements].

[[pvecm_cluster_create_via_cli]]
Create via the Command Line
~~~~~~~~~~~~~~~~~~~~~~~~~~~

Login via `ssh` to the first {pve} node and run the following command:

----
 hp1# pvecm create CLUSTERNAME
----

To check the state of the new cluster use:

----
 hp1# pvecm status
----

Multiple Clusters in the Same Network
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

It is possible to create multiple clusters in the same physical or logical
network. In this case, each cluster must have a unique name to avoid possible
clashes in the cluster communication stack. Furthermore, this helps avoid human
confusion by making clusters clearly distinguishable.

While the bandwidth requirement of a corosync cluster is relatively low, the
latency of packages and the package per second (PPS) rate is the limiting
factor. Different clusters in the same network can compete with each other for
these resources, so it may still make sense to use separate physical network
infrastructure for bigger clusters.

[[pvecm_join_node_to_cluster]]
Adding Nodes to the Cluster
---------------------------

CAUTION: A node that is about to be added to the cluster cannot hold any guests.
All existing configuration in `/etc/pve` is overwritten when joining a cluster,
since guest IDs could otherwise conflict. As a workaround, you can create a
backup of the guest (`vzdump`) and restore it under a different ID, after the
node has been added to the cluster.

Join Node to Cluster via GUI
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[thumbnail="screenshot/gui-cluster-join-information.png"]

Log in to the web interface on an existing cluster node. Under __Datacenter ->
Cluster__, click the *Join Information* button at the top. Then, click on the
button *Copy Information*. Alternatively, copy the string from the 'Information'
field manually.

[thumbnail="screenshot/gui-cluster-join.png"]

Next, log in to the web interface on the node you want to add.
Under __Datacenter -> Cluster__, click on *Join Cluster*. Fill in the
'Information' field with the 'Join Information' text you copied earlier.
Most settings required for joining the cluster will be filled out
automatically. For security reasons, the cluster password has to be entered
manually.

NOTE: To enter all required data manually, you can disable the 'Assisted Join'
checkbox.

After clicking the *Join* button, the cluster join process will start
immediately. After the node has joined the cluster, its current node certificate
will be replaced by one signed from the cluster certificate authority (CA).
This means that the current session will stop working after a few seconds. You
then might need to force-reload the web interface and log in again with the
cluster credentials.

Now your node should be visible under __Datacenter -> Cluster__.

Join Node to Cluster via Command Line
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Log in to the node you want to join into an existing cluster via `ssh`.

----
 # pvecm add IP-ADDRESS-CLUSTER
----

For `IP-ADDRESS-CLUSTER`, use the IP or hostname of an existing cluster node.
An IP address is recommended (see xref:pvecm_corosync_addresses[Link Address Types]).


To check the state of the cluster use:

----
 # pvecm status
----

.Cluster status after adding 4 nodes
----
 # pvecm status
Cluster information
~~~~~~~~~~~~~~~~~~~
Name:             prod-central
Config Version:   3
Transport:        knet
Secure auth:      on

Quorum information
~~~~~~~~~~~~~~~~~~
Date:             Tue Sep 14 11:06:47 2021
Quorum provider:  corosync_votequorum
Nodes:            4
Node ID:          0x00000001
Ring ID:          1.1a8
Quorate:          Yes

Votequorum information
~~~~~~~~~~~~~~~~~~~~~~
Expected votes:   4
Highest expected: 4
Total votes:      4
Quorum:           3
Flags:            Quorate

Membership information
~~~~~~~~~~~~~~~~~~~~~~
    Nodeid      Votes Name
0x00000001          1 192.168.15.91
0x00000002          1 192.168.15.92 (local)
0x00000003          1 192.168.15.93
0x00000004          1 192.168.15.94
----

If you only want a list of all nodes, use:

----
 # pvecm nodes
----

.List nodes in a cluster
----
 # pvecm nodes

Membership information
~~~~~~~~~~~~~~~~~~~~~~
    Nodeid      Votes Name
         1          1 hp1
         2          1 hp2 (local)
         3          1 hp3
         4          1 hp4
----

[[pvecm_adding_nodes_with_separated_cluster_network]]
Adding Nodes with Separated Cluster Network
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

When adding a node to a cluster with a separated cluster network, you need to
use the 'link0' parameter to set the nodes address on that network:

[source,bash]
----
pvecm add IP-ADDRESS-CLUSTER -link0 LOCAL-IP-ADDRESS-LINK0
----

If you want to use the built-in xref:pvecm_redundancy[redundancy] of the
Kronosnet transport layer, also use the 'link1' parameter.

Using the GUI, you can select the correct interface from the corresponding
'Link X' fields in the *Cluster Join* dialog.

Remove a Cluster Node
---------------------

CAUTION: Read the procedure carefully before proceeding, as it may
not be what you want or need.

Move all virtual machines from the node. Ensure that you have made copies of any
local data or backups that you want to keep. In addition, make sure to remove
any scheduled replication jobs to the node to be removed.

CAUTION: Failure to remove replication jobs to a node before removing said node
will result in the replication job becoming irremovable. Especially note that
replication automatically switches direction if a replicated VM is migrated, so
by migrating a replicated VM from a node to be deleted, replication jobs will be
set up to that node automatically.

In the following example, we will remove the node hp4 from the cluster.

Log in to a *different* cluster node (not hp4), and issue a `pvecm nodes`
command to identify the node ID to remove:

----
 hp1# pvecm nodes

Membership information
~~~~~~~~~~~~~~~~~~~~~~
    Nodeid      Votes Name
         1          1 hp1 (local)
         2          1 hp2
         3          1 hp3
         4          1 hp4
----


At this point, you must power off hp4 and ensure that it will not power on
again (in the network) with its current configuration.

IMPORTANT: As mentioned above, it is critical to power off the node
*before* removal, and make sure that it will *not* power on again
(in the existing cluster network) with its current configuration.
If you power on the node as it is, the cluster could end up broken,
and it could be difficult to restore it to a functioning state.

After powering off the node hp4, we can safely remove it from the cluster.

----
 hp1# pvecm delnode hp4
 Killing node 4
----

NOTE: At this point, it is possible that you will receive an error message
stating `Could not kill node (error = CS_ERR_NOT_EXIST)`. This does not
signify an actual failure in the deletion of the node, but rather a failure in
corosync trying to kill an offline node. Thus, it can be safely ignored.

Use `pvecm nodes` or `pvecm status` to check the node list again. It should
look something like:

----
hp1# pvecm status

...

Votequorum information
~~~~~~~~~~~~~~~~~~~~~~
Expected votes:   3
Highest expected: 3
Total votes:      3
Quorum:           2
Flags:            Quorate

Membership information
~~~~~~~~~~~~~~~~~~~~~~
    Nodeid      Votes Name
0x00000001          1 192.168.15.90 (local)
0x00000002          1 192.168.15.91
0x00000003          1 192.168.15.92
----

If, for whatever reason, you want this server to join the same cluster again,
you have to:

* do a fresh install of {pve} on it,

* then join it, as explained in the previous section.

NOTE: After removal of the node, its SSH fingerprint will still reside in the
'known_hosts' of the other nodes. If you receive an SSH error after rejoining
a node with the same IP or hostname, run `pvecm updatecerts` once on the
re-added node to update its fingerprint cluster wide.

[[pvecm_separate_node_without_reinstall]]
Separate a Node Without Reinstalling
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

CAUTION: This is *not* the recommended method, proceed with caution. Use the
previous method if you're unsure.

You can also separate a node from a cluster without reinstalling it from
scratch. But after removing the node from the cluster, it will still have
access to any shared storage. This must be resolved before you start removing
the node from the cluster. A {pve} cluster cannot share the exact same
storage with another cluster, as storage locking doesn't work over the cluster
boundary. Furthermore, it may also lead to VMID conflicts.

It's suggested that you create a new storage, where only the node which you want
to separate has access. This can be a new export on your NFS or a new Ceph
pool, to name a few examples. It's just important that the exact same storage
does not get accessed by multiple clusters. After setting up this storage, move
all data and VMs from the node to it. Then you are ready to separate the
node from the cluster.

WARNING: Ensure that all shared resources are cleanly separated! Otherwise you
will run into conflicts and problems.

First, stop the corosync and pve-cluster services on the node:
[source,bash]
----
systemctl stop pve-cluster
systemctl stop corosync
----

Start the cluster file system again in local mode:
[source,bash]
----
pmxcfs -l
----

Delete the corosync configuration files:
[source,bash]
----
rm /etc/pve/corosync.conf
rm -r /etc/corosync/*
----

You can now start the file system again as a normal service:
[source,bash]
----
killall pmxcfs
systemctl start pve-cluster
----

The node is now separated from the cluster. You can deleted it from any
remaining node of the cluster with:
[source,bash]
----
pvecm delnode oldnode
----

If the command fails due to a loss of quorum in the remaining node, you can set
the expected votes to 1 as a workaround:
[source,bash]
----
pvecm expected 1
----

And then repeat the 'pvecm delnode' command.

Now switch back to the separated node and delete all the remaining cluster
files on it. This ensures that the node can be added to another cluster again
without problems.

[source,bash]
----
rm /var/lib/corosync/*
----

As the configuration files from the other nodes are still in the cluster
file system, you may want to clean those up too. After making absolutely sure
that you have the correct node name, you can simply remove the entire
directory recursively from '/etc/pve/nodes/NODENAME'.

CAUTION: The node's SSH keys will remain in the 'authorized_key' file. This
means that the nodes can still connect to each other with public key
authentication. You should fix this by removing the respective keys from the
'/etc/pve/priv/authorized_keys' file.


Quorum
------

{pve} use a quorum-based technique to provide a consistent state among
all cluster nodes.

[quote, from Wikipedia, Quorum (distributed computing)]
____
A quorum is the minimum number of votes that a distributed transaction
has to obtain in order to be allowed to perform an operation in a
distributed system.
____

In case of network partitioning, state changes requires that a
majority of nodes are online. The cluster switches to read-only mode
if it loses quorum.

NOTE: {pve} assigns a single vote to each node by default.


Cluster Network
---------------

The cluster network is the core of a cluster. All messages sent over it have to
be delivered reliably to all nodes in their respective order. In {pve} this
part is done by corosync, an implementation of a high performance, low overhead,
high availability development toolkit. It serves our decentralized configuration
file system (`pmxcfs`).

[[pvecm_cluster_network_requirements]]
Network Requirements
~~~~~~~~~~~~~~~~~~~~

The {pve} cluster stack requires a reliable network with latencies under 5
milliseconds (LAN performance) between all nodes to operate stably. While on
setups with a small node count a network with higher latencies _may_ work, this
is not guaranteed and gets rather unlikely with more than three nodes and
latencies above around 10 ms.

The network should not be used heavily by other members, as while corosync does
not uses much bandwidth it is sensitive to latency jitters; ideally corosync
runs on its own physically separated network.  Especially do not use a shared
network for corosync and storage (except as a potential low-priority fallback
in a xref:pvecm_redundancy[redundant] configuration).

Before setting up a cluster, it is good practice to check if the network is fit
for that purpose. To ensure that the nodes can connect to each other on the
cluster network, you can test the connectivity between them with the `ping`
tool.

If the {pve} firewall is enabled, ACCEPT rules for corosync will automatically
be generated - no manual action is required.

NOTE: Corosync used Multicast before version 3.0 (introduced in {pve} 6.0).
Modern versions rely on https://kronosnet.org/[Kronosnet] for cluster
communication, which, for now, only supports regular UDP unicast.

CAUTION: You can still enable Multicast or legacy unicast by setting your
transport to `udp` or `udpu` in your xref:pvecm_edit_corosync_conf[corosync.conf],
but keep in mind that this will disable all cryptography and redundancy support.
This is therefore not recommended.

Separate Cluster Network
~~~~~~~~~~~~~~~~~~~~~~~~

When creating a cluster without any parameters, the corosync cluster network is
generally shared with the web interface and the VMs' network. Depending on
your setup, even storage traffic may get sent over the same network. It's
recommended to change that, as corosync is a time-critical, real-time
application.

Setting Up a New Network
^^^^^^^^^^^^^^^^^^^^^^^^

First, you have to set up a new network interface. It should be on a physically
separate network. Ensure that your network fulfills the
xref:pvecm_cluster_network_requirements[cluster network requirements].

Separate On Cluster Creation
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This is possible via the 'linkX' parameters of the 'pvecm create'
command, used for creating a new cluster.

If you have set up an additional NIC with a static address on 10.10.10.1/25,
and want to send and receive all cluster communication over this interface,
you would execute:

[source,bash]
----
pvecm create test --link0 10.10.10.1
----

To check if everything is working properly, execute:
[source,bash]
----
systemctl status corosync
----

Afterwards, proceed as described above to
xref:pvecm_adding_nodes_with_separated_cluster_network[add nodes with a separated cluster network].

[[pvecm_separate_cluster_net_after_creation]]
Separate After Cluster Creation
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

You can do this if you have already created a cluster and want to switch
its communication to another network, without rebuilding the whole cluster.
This change may lead to short periods of quorum loss in the cluster, as nodes
have to restart corosync and come up one after the other on the new network.

Check how to xref:pvecm_edit_corosync_conf[edit the corosync.conf file] first.
Then, open it and you should see a file similar to:

----
logging {
  debug: off
  to_syslog: yes
}

nodelist {

  node {
    name: due
    nodeid: 2
    quorum_votes: 1
    ring0_addr: due
  }

  node {
    name: tre
    nodeid: 3
    quorum_votes: 1
    ring0_addr: tre
  }

  node {
    name: uno
    nodeid: 1
    quorum_votes: 1
    ring0_addr: uno
  }

}

quorum {
  provider: corosync_votequorum
}

totem {
  cluster_name: testcluster
  config_version: 3
  ip_version: ipv4-6
  secauth: on
  version: 2
  interface {
    linknumber: 0
  }

}
----

NOTE: `ringX_addr` actually specifies a corosync *link address*. The name "ring"
is a remnant of older corosync versions that is kept for backwards
compatibility.

The first thing you want to do is add the 'name' properties in the node entries,
if you do not see them already. Those *must* match the node name.

Then replace all addresses from the 'ring0_addr' properties of all nodes with
the new addresses. You may use plain IP addresses or hostnames here. If you use
hostnames, ensure that they are resolvable from all nodes (see also
xref:pvecm_corosync_addresses[Link Address Types]).

In this example, we want to switch cluster communication to the
10.10.10.1/25 network, so we change the 'ring0_addr' of each node respectively.

NOTE: The exact same procedure can be used to change other 'ringX_addr' values
as well. However, we recommend only changing one link address at a time, so
that it's easier to recover if something goes wrong.

After we increase the 'config_version' property, the new configuration file
should look like:

----
logging {
  debug: off
  to_syslog: yes
}

nodelist {

  node {
    name: due
    nodeid: 2
    quorum_votes: 1
    ring0_addr: 10.10.10.2
  }

  node {
    name: tre
    nodeid: 3
    quorum_votes: 1
    ring0_addr: 10.10.10.3
  }

  node {
    name: uno
    nodeid: 1
    quorum_votes: 1
    ring0_addr: 10.10.10.1
  }

}

quorum {
  provider: corosync_votequorum
}

totem {
  cluster_name: testcluster
  config_version: 4
  ip_version: ipv4-6
  secauth: on
  version: 2
  interface {
    linknumber: 0
  }

}
----

Then, after a final check to see that all changed information is correct, we
save it and once again follow the
xref:pvecm_edit_corosync_conf[edit corosync.conf file] section to bring it into
effect.

The changes will be applied live, so restarting corosync is not strictly
necessary. If you changed other settings as well, or notice corosync
complaining, you can optionally trigger a restart.

On a single node execute:

[source,bash]
----
systemctl restart corosync
----

Now check if everything is okay:

[source,bash]
----
systemctl status corosync
----

If corosync begins to work again, restart it on all other nodes too.
They will then join the cluster membership one by one on the new network.

[[pvecm_corosync_addresses]]
Corosync Addresses
~~~~~~~~~~~~~~~~~~

A corosync link address (for backwards compatibility denoted by 'ringX_addr' in
`corosync.conf`) can be specified in two ways:

* **IPv4/v6 addresses** can be used directly. They are recommended, since they
are static and usually not changed carelessly.

* **Hostnames** will be resolved using `getaddrinfo`, which means that by
default, IPv6 addresses will be used first, if available (see also
`man gai.conf`). Keep this in mind, especially when upgrading an existing
cluster to IPv6.

CAUTION: Hostnames should be used with care, since the addresses they
resolve to can be changed without touching corosync or the node it runs on -
which may lead to a situation where an address is changed without thinking
about implications for corosync.

A separate, static hostname specifically for corosync is recommended, if
hostnames are preferred. Also, make sure that every node in the cluster can
resolve all hostnames correctly.

Since {pve} 5.1, while supported, hostnames will be resolved at the time of
entry. Only the resolved IP is saved to the configuration.

Nodes that joined the cluster on earlier versions likely still use their
unresolved hostname in `corosync.conf`. It might be a good idea to replace
them with IPs or a separate hostname, as mentioned above.


[[pvecm_redundancy]]
Corosync Redundancy
-------------------

Corosync supports redundant networking via its integrated Kronosnet layer by
default (it is not supported on the legacy udp/udpu transports). It can be
enabled by specifying more than one link address, either via the '--linkX'
parameters of `pvecm`, in the GUI as **Link 1** (while creating a cluster or
adding a new node) or by specifying more than one 'ringX_addr' in
`corosync.conf`.

NOTE: To provide useful failover, every link should be on its own
physical network connection.

Links are used according to a priority setting. You can configure this priority
by setting 'knet_link_priority' in the corresponding interface section in
`corosync.conf`, or, preferably, using the 'priority' parameter when creating
your cluster with `pvecm`:

----
 # pvecm create CLUSTERNAME --link0 10.10.10.1,priority=15 --link1 10.20.20.1,priority=20
----

This would cause 'link1' to be used first, since it has the higher priority.

If no priorities are configured manually (or two links have the same priority),
links will be used in order of their number, with the lower number having higher
priority.

Even if all links are working, only the one with the highest priority will see
corosync traffic. Link priorities cannot be mixed, meaning that links with
different priorities will not be able to communicate with each other.

Since lower priority links will not see traffic unless all higher priorities
have failed, it becomes a useful strategy to specify networks used for
other tasks (VMs, storage, etc.) as low-priority links. If worst comes to
worst, a higher latency or more congested connection might be better than no
connection at all.

Adding Redundant Links To An Existing Cluster
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

To add a new link to a running configuration, first check how to
xref:pvecm_edit_corosync_conf[edit the corosync.conf file].

Then, add a new 'ringX_addr' to every node in the `nodelist` section. Make
sure that your 'X' is the same for every node you add it to, and that it is
unique for each node.

Lastly, add a new 'interface', as shown below, to your `totem`
section, replacing 'X' with the link number chosen above.

Assuming you added a link with number 1, the new configuration file could look
like this:

----
logging {
  debug: off
  to_syslog: yes
}

nodelist {

  node {
    name: due
    nodeid: 2
    quorum_votes: 1
    ring0_addr: 10.10.10.2
    ring1_addr: 10.20.20.2
  }

  node {
    name: tre
    nodeid: 3
    quorum_votes: 1
    ring0_addr: 10.10.10.3
    ring1_addr: 10.20.20.3
  }

  node {
    name: uno
    nodeid: 1
    quorum_votes: 1
    ring0_addr: 10.10.10.1
    ring1_addr: 10.20.20.1
  }

}

quorum {
  provider: corosync_votequorum
}

totem {
  cluster_name: testcluster
  config_version: 4
  ip_version: ipv4-6
  secauth: on
  version: 2
  interface {
    linknumber: 0
  }
  interface {
    linknumber: 1
  }
}
----

The new link will be enabled as soon as you follow the last steps to
xref:pvecm_edit_corosync_conf[edit the corosync.conf file]. A restart should not
be necessary. You can check that corosync loaded the new link using:

----
journalctl -b -u corosync
----

It might be a good idea to test the new link by temporarily disconnecting the
old link on one node and making sure that its status remains online while
disconnected:

----
pvecm status
----

If you see a healthy cluster state, it means that your new link is being used.


Role of SSH in {pve} Clusters
-----------------------------

{pve} utilizes SSH tunnels for various features.

* Proxying console/shell sessions (node and guests)
+
When using the shell for node B while being connected to node A, connects to a
terminal proxy on node A, which is in turn connected to the login shell on node
B via a non-interactive SSH tunnel.

* VM and CT memory and local-storage migration in 'secure' mode.
+
During the migration, one or more SSH tunnel(s) are established between the
source and target nodes, in order to exchange migration information and
transfer memory and disk contents.

* Storage replication

.Pitfalls due to automatic execution of `.bashrc` and siblings
[IMPORTANT]
====
In case you have a custom `.bashrc`, or similar files that get executed on
login by the configured shell, `ssh` will automatically run it once the session
is established successfully. This can cause some unexpected behavior, as those
commands may be executed with root permissions on any of the operations
described above. This can cause possible problematic side-effects!

In order to avoid such complications, it's recommended to add a check in
`/root/.bashrc` to make sure the session is interactive, and only then run
`.bashrc` commands.

You can add this snippet at the beginning of your `.bashrc` file:

----
# Early exit if not running interactively to avoid side-effects!
case $- in
    *i*) ;;
      *) return;;
esac
----
====


Corosync External Vote Support
------------------------------

This section describes a way to deploy an external voter in a {pve} cluster.
When configured, the cluster can sustain more node failures without
violating safety properties of the cluster communication.

For this to work, there are two services involved:

* A QDevice daemon which runs on each {pve} node

* An external vote daemon which runs on an independent server

As a result, you can achieve higher availability, even in smaller setups (for
example 2+1 nodes).

QDevice Technical Overview
~~~~~~~~~~~~~~~~~~~~~~~~~~

The Corosync Quorum Device (QDevice) is a daemon which runs on each cluster
node. It provides a configured number of votes to the cluster's quorum
subsystem, based on an externally running third-party arbitrator's decision.
Its primary use is to allow a cluster to sustain more node failures than
standard quorum rules allow. This can be done safely as the external device
can see all nodes and thus choose only one set of nodes to give its vote.
This will only be done if said set of nodes can have quorum (again) after
receiving the third-party vote.

Currently, only 'QDevice Net' is supported as a third-party arbitrator. This is
a daemon which provides a vote to a cluster partition, if it can reach the
partition members over the network. It will only give votes to one partition
of a cluster at any time.
It's designed to support multiple clusters and is almost configuration and
state free. New clusters are handled dynamically and no configuration file
is needed on the host running a QDevice.

The only requirements for the external host are that it needs network access to
the cluster and to have a corosync-qnetd package available. We provide a package
for Debian based hosts, and other Linux distributions should also have a package
available through their respective package manager.

NOTE: Unlike corosync itself, a QDevice connects to the cluster over TCP/IP.
The daemon can also run outside the LAN of the cluster and isn't limited to the
low latencies requirements of corosync.

Supported Setups
~~~~~~~~~~~~~~~~

We support QDevices for clusters with an even number of nodes and recommend
it for 2 node clusters, if they should provide higher availability.
For clusters with an odd node count, we currently discourage the use of
QDevices. The reason for this is the difference in the votes which the QDevice
provides for each cluster type. Even numbered clusters get a single additional
vote, which only increases availability, because if the QDevice
itself fails, you are in the same position as with no QDevice at all.

On the other hand, with an odd numbered cluster size, the QDevice provides
'(N-1)' votes -- where 'N' corresponds to the cluster node count. This
alternative behavior makes sense; if it had only one additional vote, the
cluster could get into a split-brain situation. This algorithm allows for all
nodes but one (and naturally the QDevice itself) to fail. However, there are two
drawbacks to this:

* If the QNet daemon itself fails, no other node may fail or the cluster
  immediately loses quorum. For example, in a cluster with 15 nodes, 7
  could fail before the cluster becomes inquorate. But, if a QDevice is
  configured here and it itself fails, **no single node** of the 15 may fail.
  The QDevice acts almost as a single point of failure in this case.

* The fact that all but one node plus QDevice may fail sounds promising at
  first, but this may result in a mass recovery of HA services, which could
  overload the single remaining node. Furthermore, a Ceph server will stop
  providing services if only '((N-1)/2)' nodes or less remain online.

If you understand the drawbacks and implications, you can decide yourself if
you want to use this technology in an odd numbered cluster setup.

QDevice-Net Setup
~~~~~~~~~~~~~~~~~

We recommend running any daemon which provides votes to corosync-qdevice as an
unprivileged user. {pve} and Debian provide a package which is already
configured to do so.
The traffic between the daemon and the cluster must be encrypted to ensure a
safe and secure integration of the QDevice in {pve}.

First, install the 'corosync-qnetd' package on your external server

----
external# apt install corosync-qnetd
----

and the 'corosync-qdevice' package on all cluster nodes

----
pve# apt install corosync-qdevice
----

After doing this, ensure that all the nodes in the cluster are online.

You can now set up your QDevice by running the following command on one
of the {pve} nodes:

----
pve# pvecm qdevice setup <QDEVICE-IP>
----

The SSH key from the cluster will be automatically copied to the QDevice.

NOTE: Make sure that the SSH configuration on your external server allows root
login via password, if you are asked for a password during this step.

After you enter the password and all the steps have successfully completed, you
will see "Done". You can verify that the QDevice has been set up with:

----
pve# pvecm status

...

Votequorum information
~~~~~~~~~~~~~~~~~~~~~
Expected votes:   3
Highest expected: 3
Total votes:      3
Quorum:           2
Flags:            Quorate Qdevice

Membership information
~~~~~~~~~~~~~~~~~~~~~~
    Nodeid      Votes    Qdevice Name
    0x00000001          1    A,V,NMW 192.168.22.180 (local)
    0x00000002          1    A,V,NMW 192.168.22.181
    0x00000000          1            Qdevice

----


Frequently Asked Questions
~~~~~~~~~~~~~~~~~~~~~~~~~~

Tie Breaking
^^^^^^^^^^^^

In case of a tie, where two same-sized cluster partitions cannot see each other
but can see the QDevice, the QDevice chooses one of those partitions randomly
and provides a vote to it.

Possible Negative Implications
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

For clusters with an even node count, there are no negative implications when
using a QDevice. If it fails to work, it is the same as not having a QDevice
at all.

Adding/Deleting Nodes After QDevice Setup
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

If you want to add a new node or remove an existing one from a cluster with a
QDevice setup, you need to remove the QDevice first. After that, you can add or
remove nodes normally. Once you have a cluster with an even node count again,
you can set up the QDevice again as described previously.

Removing the QDevice
^^^^^^^^^^^^^^^^^^^^

If you used the official `pvecm` tool to add the QDevice, you can remove it
by running:

----
pve# pvecm qdevice remove
----

//Still TODO
//^^^^^^^^^^
//There is still stuff to add here


Corosync Configuration
----------------------

The `/etc/pve/corosync.conf` file plays a central role in a {pve} cluster. It
controls the cluster membership and its network.
For further information about it, check the corosync.conf man page:
[source,bash]
----
man corosync.conf
----

For node membership, you should always use the `pvecm` tool provided by {pve}.
You may have to edit the configuration file manually for other changes.
Here are a few best practice tips for doing this.

[[pvecm_edit_corosync_conf]]
Edit corosync.conf
~~~~~~~~~~~~~~~~~~

Editing the corosync.conf file is not always very straightforward. There are
two on each cluster node, one in `/etc/pve/corosync.conf` and the other in
`/etc/corosync/corosync.conf`. Editing the one in our cluster file system will
propagate the changes to the local one, but not vice versa.

The configuration will get updated automatically, as soon as the file changes.
This means that changes which can be integrated in a running corosync will take
effect immediately. Thus, you should always make a copy and edit that instead,
to avoid triggering unintended changes when saving the file while editing.

[source,bash]
----
cp /etc/pve/corosync.conf /etc/pve/corosync.conf.new
----

Then, open the config file with your favorite editor, such as `nano` or
`vim.tiny`, which come pre-installed on every {pve} node.

NOTE: Always increment the 'config_version' number after configuration changes;
omitting this can lead to problems.

After making the necessary changes, create another copy of the current working
configuration file. This serves as a backup if the new configuration fails to
apply or causes other issues.

[source,bash]
----
cp /etc/pve/corosync.conf /etc/pve/corosync.conf.bak
----

Then replace the old configuration file with the new one:
[source,bash]
----
mv /etc/pve/corosync.conf.new /etc/pve/corosync.conf
----

You can check if the changes could be applied automatically, using the following
commands:
[source,bash]
----
systemctl status corosync
journalctl -b -u corosync
----

If the changes could not be applied automatically, you may have to restart the
corosync service via:
[source,bash]
----
systemctl restart corosync
----

On errors, check the troubleshooting section below.

Troubleshooting
~~~~~~~~~~~~~~~

Issue: 'quorum.expected_votes must be configured'
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

When corosync starts to fail and you get the following message in the system log:

----
[...]
corosync[1647]:  [QUORUM] Quorum provider: corosync_votequorum failed to initialize.
corosync[1647]:  [SERV  ] Service engine 'corosync_quorum' failed to load for reason
    'configuration error: nodelist or quorum.expected_votes must be configured!'
[...]
----

It means that the hostname you set for a corosync 'ringX_addr' in the
configuration could not be resolved.

Write Configuration When Not Quorate
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

If you need to change '/etc/pve/corosync.conf' on a node with no quorum, and you
understand what you are doing, use:
[source,bash]
----
pvecm expected 1
----

This sets the expected vote count to 1 and makes the cluster quorate. You can
then fix your configuration, or revert it back to the last working backup.

This is not enough if corosync cannot start anymore. In that case, it is best to
edit the local copy of the corosync configuration in
'/etc/corosync/corosync.conf', so that corosync can start again. Ensure that on
all nodes, this configuration has the same content to avoid split-brain
situations.


[[pvecm_corosync_conf_glossary]]
Corosync Configuration Glossary
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

ringX_addr::
This names the different link addresses for the Kronosnet connections between
nodes.


Cluster Cold Start
------------------

It is obvious that a cluster is not quorate when all nodes are
offline. This is a common case after a power failure.

NOTE: It is always a good idea to use an uninterruptible power supply
(``UPS'', also called ``battery backup'') to avoid this state, especially if
you want HA.

On node startup, the `pve-guests` service is started and waits for
quorum. Once quorate, it starts all guests which have the `onboot`
flag set.

When you turn on nodes, or when power comes back after power failure,
it is likely that some nodes will boot faster than others. Please keep in
mind that guest startup is delayed until you reach quorum.


Guest Migration
---------------

Migrating virtual guests to other nodes is a useful feature in a
cluster. There are settings to control the behavior of such
migrations. This can be done via the configuration file
`datacenter.cfg` or for a specific migration via API or command line
parameters.

It makes a difference if a guest is online or offline, or if it has
local resources (like a local disk).

For details about virtual machine migration, see the
xref:qm_migration[QEMU/KVM Migration Chapter].

For details about container migration, see the
xref:pct_migration[Container Migration Chapter].

Migration Type
~~~~~~~~~~~~~~

The migration type defines if the migration data should be sent over an
encrypted (`secure`) channel or an unencrypted (`insecure`) one.
Setting the migration type to `insecure` means that the RAM content of a
virtual guest is also transferred unencrypted, which can lead to
information disclosure of critical data from inside the guest (for
example, passwords or encryption keys).

Therefore, we strongly recommend using the secure channel if you do
not have full control over the network and can not guarantee that no
one is eavesdropping on it.

NOTE: Storage migration does not follow this setting. Currently, it
always sends the storage content over a secure channel.

Encryption requires a lot of computing power, so this setting is often
changed to `insecure` to achieve better performance. The impact on
modern systems is lower because they implement AES encryption in
hardware. The performance impact is particularly evident in fast
networks, where you can transfer 10 Gbps or more.

Migration Network
~~~~~~~~~~~~~~~~~

By default, {pve} uses the network in which cluster communication
takes place to send the migration traffic. This is not optimal both because
sensitive cluster traffic can be disrupted and this network may not
have the best bandwidth available on the node.

Setting the migration network parameter allows the use of a dedicated
network for all migration traffic. In addition to the memory,
this also affects the storage traffic for offline migrations.

The migration network is set as a network using CIDR notation. This
has the advantage that you don't have to set individual IP addresses
for each node. {pve} can determine the real address on the
destination node from the network specified in the CIDR form. To
enable this, the network must be specified so that each node has exactly one
IP in the respective network.

Example
^^^^^^^

We assume that we have a three-node setup, with three separate
networks. One for public communication with the Internet, one for
cluster communication, and a very fast one, which we want to use as a
dedicated network for migration.

A network configuration for such a setup might look as follows:

----
iface eno1 inet manual

# public network
auto vmbr0
iface vmbr0 inet static
    address 192.X.Y.57/24
    gateway 192.X.Y.1
    bridge-ports eno1
    bridge-stp off
    bridge-fd 0

# cluster network
auto eno2
iface eno2 inet static
    address  10.1.1.1/24

# fast network
auto eno3
iface eno3 inet static
    address  10.1.2.1/24
----

Here, we will use the network 10.1.2.0/24 as a migration network. For
a single migration, you can do this using the `migration_network`
parameter of the command line tool:

----
# qm migrate 106 tre --online --migration_network 10.1.2.0/24
----

To configure this as the default network for all migrations in the
cluster, set the `migration` property of the `/etc/pve/datacenter.cfg`
file:

----
# use dedicated migration network
migration: secure,network=10.1.2.0/24
----

NOTE: The migration type must always be set when the migration network
is set in `/etc/pve/datacenter.cfg`.


ifdef::manvolnum[]
include::pve-copyright.adoc[]
endif::manvolnum[]