improve error recovery section

[pve-docs.git] / pvecm.adoc

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, and such clusters can consist of up to 32 physical nodes
(probably more, dependent on network latency).

`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 task

* `pmxcfs`: 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 in the same network as `corosync` uses IP Multicast
 to communicate between nodes (also see
 http://www.corosync.org[Corosync Cluster Engine]). Corosync uses UDP
 ports 5404 and 5405 for cluster communication.
+
NOTE: Some switches do not support IP multicast by default and must be
manually enabled first.

* Date and time have to be synchronized.

* SSH tunnel on TCP port 22 between nodes is used. 

* 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.

NOTE: It is not possible to mix Proxmox VE 3.x and earlier with
Proxmox VE 4.0 cluster nodes.


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.

Currently the cluster creation has to be done on the console, so you
need to login via `ssh`.

Create the Cluster
------------------

Login via `ssh` to the first {pve} node. Use a unique name for your cluster.
This name cannot be changed later.

 hp1# pvecm create YOUR-CLUSTER-NAME

CAUTION: The cluster name is used to compute the default multicast
address. Please use unique cluster names if you run more than one
cluster inside your network.

To check the state of your cluster use:

 hp1# pvecm status


Adding Nodes to the Cluster
---------------------------

Login via `ssh` to the node you want to add.

 hp2# pvecm add IP-ADDRESS-CLUSTER

For `IP-ADDRESS-CLUSTER` use the IP from an existing cluster node.

CAUTION: A new node cannot hold any VMs, because you would get
conflicts about identical VM IDs. Also, all existing configuration in
`/etc/pve` is overwritten when you join a new node to the cluster. To
workaround, use `vzdump` to backup and restore to a different VMID after
adding the node to the cluster.

To check the state of cluster:

 # pvecm status

.Cluster status after adding 4 nodes
----
hp2# pvecm status
Quorum information
~~~~~~~~~~~~~~~~~~
Date:             Mon Apr 20 12:30:13 2015
Quorum provider:  corosync_votequorum
Nodes:            4
Node ID:          0x00000001
Ring ID:          1928
Quorate:          Yes

Votequorum information
~~~~~~~~~~~~~~~~~~~~~~
Expected votes:   4
Highest expected: 4
Total votes:      4
Quorum:           2
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 the list of all nodes use:

 # pvecm nodes

.List nodes in a cluster
----
hp2# pvecm nodes

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

Adding Nodes With Separated Cluster Network
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

When adding a node to a cluster with a separated cluster network you need to
use the 'ringX_addr' parameters to set the nodes address on those networks:

[source,bash]
----
pvecm add IP-ADDRESS-CLUSTER -ring0_addr IP-ADDRESS-RING0
----

If you want to use the Redundant Ring Protocol you will also want to pass the
'ring1_addr' parameter.


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

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

Move all virtual machines from the node. Make sure you have no local
data or backups you want to keep, or save them accordingly.

Log in to one remaining node via ssh. Issue a `pvecm nodes` command to
identify the node ID:

----
hp1# pvecm status

Quorum information
~~~~~~~~~~~~~~~~~~
Date:             Mon Apr 20 12:30:13 2015
Quorum provider:  corosync_votequorum
Nodes:            4
Node ID:          0x00000001
Ring ID:          1928
Quorate:          Yes

Votequorum information
~~~~~~~~~~~~~~~~~~~~~~
Expected votes:   4
Highest expected: 4
Total votes:      4
Quorum:           2
Flags:            Quorate

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

IMPORTANT: at this point you must power off the node to be removed and
make sure that it will not power on again (in the network) as it
is.

----
hp1# pvecm nodes

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

Log in to one remaining node via ssh. Issue the delete command (here
deleting node `hp4`):

 hp1# pvecm delnode hp4

If the operation succeeds no output is returned, just check the node
list again with `pvecm nodes` or `pvecm status`. You should see
something like:

----
hp1# pvecm status

Quorum information
~~~~~~~~~~~~~~~~~~
Date:             Mon Apr 20 12:44:28 2015
Quorum provider:  corosync_votequorum
Nodes:            3
Node ID:          0x00000001
Ring ID:          1992
Quorate:          Yes

Votequorum information
~~~~~~~~~~~~~~~~~~~~~~
Expected votes:   3
Highest expected: 3
Total votes:      3
Quorum:           3
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
----

IMPORTANT: as said above, it is very important to power off the node
*before* removal, and make sure that it will *never* power on again
(in the existing cluster network) as it is.

If you power on the node as it is, your cluster will be screwed up and
it could be difficult to restore a clean cluster state.

If, for whatever reason, you want that this server joins the same
cluster again, you have to

* reinstall {pve} on it from scratch

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

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

CAUTION: This is *not* the recommended method, proceed with caution. Use the
above mentioned 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 the shared storages! 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 it leads to VMID conflicts.

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

WARNING: Ensure all shared resources are cleanly separated! You will run into
conflicts and problems else.

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

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

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

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

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

If the command failed, because the remaining node in the cluster lost quorum
when the now separate node exited, you may set the expected votes to 1 as a workaround:
[source,bash]
----
pvecm expected 1
----

And the repeat the 'pvecm delnode' command.

Now switch back to the separated node, here delete all remaining files left
from the old cluster. 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
filesystem you may want to clean those up too.  Remove simply the whole
directory recursive from '/etc/pve/nodes/NODENAME', but check three times that
you used the correct one before deleting it.

CAUTION: The nodes SSH keys are still in the 'authorized_key' file, this means
the nodes can still connect to each other with public key authentication. This
should be fixed 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 reliable 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`).

[[cluster-network-requirements]]
Network Requirements
~~~~~~~~~~~~~~~~~~~~
This needs a reliable network with latencies under 2 milliseconds (LAN
performance) to work properly. While corosync can also use unicast for
communication between nodes its **highly recommended** to have a multicast
capable network. The network should not be used heavily by other members,
ideally corosync runs on its own network.
*never* share it with network where storage communicates too.

Before setting up a cluster it is good practice to check if the network is fit
for that purpose.

* Ensure that all nodes are in the same subnet. This must only be true for the
  network interfaces used for cluster communication (corosync).

* Ensure all nodes can reach each other over those interfaces, using `ping` is
  enough for a basic test.

* Ensure that multicast works in general and a high package rates. This can be
  done with the `omping` tool. The final "%loss" number should be < 1%.
[source,bash]
----
omping -c 10000 -i 0.001 -F -q NODE1-IP NODE2-IP ...
----

* Ensure that multicast communication works over an extended period of time.
  This covers up problems where IGMP snooping is activated on the network but
  no multicast querier is active. This test has a duration of around 10
  minutes.
[source,bash]
----
omping -c 600 -i 1 -q NODE1-IP NODE2-IP ...
----

Your network is not ready for clustering if any of these test fails. Recheck
your network configuration. Especially switches are notorious for having
multicast disabled by default or IGMP snooping enabled with no IGMP querier
active.

In smaller cluster its also an option to use unicast if you really cannot get
multicast to work.

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

When creating a cluster without any parameters the cluster network is generally
shared with the Web UI and the VMs and its traffic. Depending on your setup
even storage traffic may get sent over the same network. Its recommended to
change that, as corosync is a time critical real time application.

Setting Up A New Network
^^^^^^^^^^^^^^^^^^^^^^^^

First you have to setup a new network interface. It should be on a physical
separate network. Ensure that your network fulfills the
<<cluster-network-requirements,cluster network requirements>>.

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

This is possible through the 'ring0_addr' and 'bindnet0_addr' parameter of
the 'pvecm create' command used for creating a new cluster.

If you have setup a 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 --ring0_addr 10.10.10.1 --bindnet0_addr 10.10.10.0
----

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

[[separate-cluster-net-after-creation]]
Separate After Cluster Creation
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

You can do this also 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 durations 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 <<edit-corosync-conf,edit the corosync.conf file>> first.
The 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: thomas-testcluster
  config_version: 3
  ip_version: ipv4
  secauth: on
  version: 2
  interface {
    bindnetaddr: 192.168.30.50
    ringnumber: 0
  }

}
----

The first 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 the address from the 'ring0_addr' properties with the new
addresses.  You may use plain IP addresses or also hostnames here. If you use
hostnames ensure that they are resolvable from all nodes.

In my example I want to switch my cluster communication to the 10.10.10.1/25
network. So I replace all 'ring0_addr' respectively. I also set the bindetaddr
in the totem section of the config to an address of the new network. It can be
any address from the subnet configured on the new network interface.

After you increased 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: thomas-testcluster
  config_version: 4
  ip_version: ipv4
  secauth: on
  version: 2
  interface {
    bindnetaddr: 10.10.10.1
    ringnumber: 0
  }

}
----

Now after a final check whether all changed information is correct we save it
and see again the <<edit-corosync-conf,edit corosync.conf file>> section to
learn how to bring it in effect.

As our change cannot be enforced live from corosync we have to do an restart.

On a single node execute:
[source,bash]
----
systemctl restart corosync
----

Now check if everything is fine:

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

If corosync runs again correct restart corosync also on all other nodes.
They will then join the cluster membership one by one on the new network.

Redundant Ring Protocol
~~~~~~~~~~~~~~~~~~~~~~~
To avoid a single point of failure you should implement counter measurements.
This can be on the hardware and operating system level through network bonding.

Corosync itself offers also a possibility to add redundancy through the so
called 'Redundant Ring Protocol'. This protocol allows running a second totem
ring on another network, this network should be physically separated from the
other rings network to actually increase availability.

RRP On Cluster Creation
~~~~~~~~~~~~~~~~~~~~~~~

The 'pvecm create' command provides the additional parameters 'bindnetX_addr',
'ringX_addr' and 'rrp_mode', can be used for RRP configuration.

NOTE: See the <<corosync-conf-glossary,glossary>> if you do not know what each parameter means.

So if you have two networks, one on the 10.10.10.1/24 and the other on the
10.10.20.1/24 subnet you would execute:

[source,bash]
----
pvecm create CLUSTERNAME -bindnet0_addr 10.10.10.1 -ring0_addr 10.10.10.1 \
-bindnet1_addr 10.10.20.1 -ring1_addr 10.10.20.1
----

RRP On A Created Cluster
~~~~~~~~~~~~~~~~~~~~~~~~

When enabling an already running cluster to use RRP you will take similar steps
as describe in
<<separate-cluster-net-after-creation,separating the cluster network>>. You
just do it on another ring.

First add a new `interface` subsection in the `totem` section, set its
`ringnumber` property to `1`. Set the interfaces `bindnetaddr` property to an
address of the subnet you have configured for your new ring.
Further set the `rrp_mode` to `passive`, this is the only stable mode.

Then add to each node entry in the `nodelist` section its new `ring1_addr`
property with the nodes additional ring address.

So if you have two networks, one on the 10.10.10.1/24 and the other on the
10.10.20.1/24 subnet, the final configuration file should look like:

----
totem {
  cluster_name: tweak
  config_version: 9
  ip_version: ipv4
  rrp_mode: passive
  secauth: on
  version: 2
  interface {
    bindnetaddr: 10.10.10.1
    ringnumber: 0
  }
  interface {
    bindnetaddr: 10.10.20.1
    ringnumber: 1
  }
}

nodelist {
  node {
    name: pvecm1
    nodeid: 1
    quorum_votes: 1
    ring0_addr: 10.10.10.1
    ring1_addr: 10.10.20.1
  }

 node {
    name: pvecm2
    nodeid: 2
    quorum_votes: 1
    ring0_addr: 10.10.10.2
    ring1_addr: 10.10.20.2
  }

  [...] # other cluster nodes here
}

[...] # other remaining config sections here

----

Bring it in effect like described in the
<<edit-corosync-conf,edit the corosync.conf file>> section.

This is a change which cannot take live in effect and needs at least a restart
of corosync. Recommended is a restart of the whole cluster.

If you cannot reboot the whole cluster ensure no High Availability services are
configured and the stop the corosync service on all nodes. After corosync is
stopped on all nodes start it one after the other again.

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

The `/ect/pve/corosync.conf` file plays a central role in {pve} cluster. It
controls the cluster member ship and its network.
For reading more 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.

[[edit-corosync-conf]]
Edit corosync.conf
~~~~~~~~~~~~~~~~~~

Editing the corosync.conf file can be not always straight forward. There are
two on each cluster, 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 changes which can be integrated in a running corosync will take
instantly effect. So you should always make a copy and edit that instead, to
avoid triggering some unwanted changes by an in between safe.

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

Then open the Config file with your favorite editor, `nano` and `vim.tiny` are
preinstalled on {pve} for example.

NOTE: Always increment the 'config_version' number on 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 makes problems in other ways.

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

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

You may check with the commands
[source,bash]
----
systemctl status corosync
journalctl -b -u corosync
----

If the change could applied automatically. If not 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 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 an node with no quorum, and you
know what you do, use:
[source,bash]
----
pvecm expected 1
----

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

This is not enough if corosync cannot start anymore. Here its 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 brains. If you are not sure what went wrong
it's best to ask the Proxmox Community to help you.


[[corosync-conf-glossary]]
Corosync Configuration Glossary
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

ringX_addr::
This names the different ring addresses for the corosync totem rings used for
the cluster communication.

bindnetaddr::
Defines to which interface the ring should bind to. It may be any address of
the subnet configured on the interface we want to use. In general its the
recommended to just use an address a node uses on this interface.

rrp_mode::
Specifies the mode of the redundant ring protocol and may be passive, active or
none. Note that use of active is highly experimental and not official
supported. Passive is the preferred mode, it may double the cluster
communication throughput and increases availability.


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, service `pve-manager` 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 boots faster than others. Please keep in
mind that guest startup is delayed until you reach quorum.


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