10 pvecm - Proxmox VE Cluster Manager
15 include::pvecm.1-synopsis.adoc[]
27 The {pve} cluster manager `pvecm` is a tool to create a group of
28 physical servers. Such a group is called a *cluster*. We use the
29 http://www.corosync.org[Corosync Cluster Engine] for reliable group
30 communication. There's no explicit limit for the number of nodes in a cluster.
31 In practice, the actual possible node count may be limited by the host and
32 network performance. Currently (2021), there are reports of clusters (using
33 high-end enterprise hardware) with over 50 nodes in production.
35 `pvecm` can be used to create a new cluster, join nodes to a cluster,
36 leave the cluster, get status information, and do various other cluster-related
37 tasks. The **P**rox**m**o**x** **C**luster **F**ile **S**ystem (``pmxcfs'')
38 is used to transparently distribute the cluster configuration to all cluster
41 Grouping nodes into a cluster has the following advantages:
43 * Centralized, web-based management
45 * Multi-master clusters: each node can do all management tasks
47 * Use of `pmxcfs`, a database-driven file system, for storing configuration
48 files, replicated in real-time on all nodes using `corosync`
50 * Easy migration of virtual machines and containers between physical
55 * Cluster-wide services like firewall and HA
61 * All nodes must be able to connect to each other via UDP ports 5405-5412
64 * Date and time must be synchronized.
66 * An SSH tunnel on TCP port 22 between nodes is required.
68 * If you are interested in High Availability, you need to have at
69 least three nodes for reliable quorum. All nodes should have the
72 * We recommend a dedicated NIC for the cluster traffic, especially if
73 you use shared storage.
75 * The root password of a cluster node is required for adding nodes.
77 * Online migration of virtual machines is only supported when nodes have CPUs
78 from the same vendor. It might work otherwise, but this is never guaranteed.
80 NOTE: It is not possible to mix {pve} 3.x and earlier with {pve} 4.X cluster
83 NOTE: While it's possible to mix {pve} 4.4 and {pve} 5.0 nodes, doing so is
84 not supported as a production configuration and should only be done temporarily,
85 during an upgrade of the whole cluster from one major version to another.
87 NOTE: Running a cluster of {pve} 6.x with earlier versions is not possible. The
88 cluster protocol (corosync) between {pve} 6.x and earlier versions changed
89 fundamentally. The corosync 3 packages for {pve} 5.4 are only intended for the
90 upgrade procedure to {pve} 6.0.
96 First, install {pve} on all nodes. Make sure that each node is
97 installed with the final hostname and IP configuration. Changing the
98 hostname and IP is not possible after cluster creation.
100 While it's common to reference all node names and their IPs in `/etc/hosts` (or
101 make their names resolvable through other means), this is not necessary for a
102 cluster to work. It may be useful however, as you can then connect from one node
103 to another via SSH, using the easier to remember node name (see also
104 xref:pvecm_corosync_addresses[Link Address Types]). Note that we always
105 recommend referencing nodes by their IP addresses in the cluster configuration.
108 [[pvecm_create_cluster]]
112 You can either create a cluster on the console (login via `ssh`), or through
113 the API using the {pve} web interface (__Datacenter -> Cluster__).
115 NOTE: Use a unique name for your cluster. This name cannot be changed later.
116 The cluster name follows the same rules as node names.
118 [[pvecm_cluster_create_via_gui]]
122 [thumbnail="screenshot/gui-cluster-create.png"]
124 Under __Datacenter -> Cluster__, click on *Create Cluster*. Enter the cluster
125 name and select a network connection from the drop-down list to serve as the
126 main cluster network (Link 0). It defaults to the IP resolved via the node's
129 As of {pve} 6.2, up to 8 fallback links can be added to a cluster. To add a
130 redundant link, click the 'Add' button and select a link number and IP address
131 from the respective fields. Prior to {pve} 6.2, to add a second link as
132 fallback, you can select the 'Advanced' checkbox and choose an additional
133 network interface (Link 1, see also xref:pvecm_redundancy[Corosync Redundancy]).
135 NOTE: Ensure that the network selected for cluster communication is not used for
136 any high traffic purposes, like network storage or live-migration.
137 While the cluster network itself produces small amounts of data, it is very
138 sensitive to latency. Check out full
139 xref:pvecm_cluster_network_requirements[cluster network requirements].
141 [[pvecm_cluster_create_via_cli]]
142 Create via the Command Line
143 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
145 Login via `ssh` to the first {pve} node and run the following command:
148 hp1# pvecm create CLUSTERNAME
151 To check the state of the new cluster use:
157 Multiple Clusters in the Same Network
158 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
160 It is possible to create multiple clusters in the same physical or logical
161 network. In this case, each cluster must have a unique name to avoid possible
162 clashes in the cluster communication stack. Furthermore, this helps avoid human
163 confusion by making clusters clearly distinguishable.
165 While the bandwidth requirement of a corosync cluster is relatively low, the
166 latency of packages and the package per second (PPS) rate is the limiting
167 factor. Different clusters in the same network can compete with each other for
168 these resources, so it may still make sense to use separate physical network
169 infrastructure for bigger clusters.
171 [[pvecm_join_node_to_cluster]]
172 Adding Nodes to the Cluster
173 ---------------------------
175 CAUTION: All existing configuration in `/etc/pve` is overwritten when joining a
176 cluster. In particular, a joining node cannot hold any guests, since guest IDs
177 could otherwise conflict, and the node will inherit the cluster's storage
178 configuration. To join a node with existing guest, as a workaround, you can
179 create a backup of each guest (using `vzdump`) and restore it under a different
180 ID after joining. If the node's storage layout differs, you will need to re-add
181 the node's storages, and adapt each storage's node restriction to reflect on
182 which nodes the storage is actually available.
184 Join Node to Cluster via GUI
185 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
187 [thumbnail="screenshot/gui-cluster-join-information.png"]
189 Log in to the web interface on an existing cluster node. Under __Datacenter ->
190 Cluster__, click the *Join Information* button at the top. Then, click on the
191 button *Copy Information*. Alternatively, copy the string from the 'Information'
194 [thumbnail="screenshot/gui-cluster-join.png"]
196 Next, log in to the web interface on the node you want to add.
197 Under __Datacenter -> Cluster__, click on *Join Cluster*. Fill in the
198 'Information' field with the 'Join Information' text you copied earlier.
199 Most settings required for joining the cluster will be filled out
200 automatically. For security reasons, the cluster password has to be entered
203 NOTE: To enter all required data manually, you can disable the 'Assisted Join'
206 After clicking the *Join* button, the cluster join process will start
207 immediately. After the node has joined the cluster, its current node certificate
208 will be replaced by one signed from the cluster certificate authority (CA).
209 This means that the current session will stop working after a few seconds. You
210 then might need to force-reload the web interface and log in again with the
213 Now your node should be visible under __Datacenter -> Cluster__.
215 Join Node to Cluster via Command Line
216 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
218 Log in to the node you want to join into an existing cluster via `ssh`.
221 # pvecm add IP-ADDRESS-CLUSTER
224 For `IP-ADDRESS-CLUSTER`, use the IP or hostname of an existing cluster node.
225 An IP address is recommended (see xref:pvecm_corosync_addresses[Link Address Types]).
228 To check the state of the cluster use:
234 .Cluster status after adding 4 nodes
246 Date: Tue Sep 14 11:06:47 2021
247 Quorum provider: corosync_votequorum
253 Votequorum information
254 ~~~~~~~~~~~~~~~~~~~~~~
261 Membership information
262 ~~~~~~~~~~~~~~~~~~~~~~
264 0x00000001 1 192.168.15.91
265 0x00000002 1 192.168.15.92 (local)
266 0x00000003 1 192.168.15.93
267 0x00000004 1 192.168.15.94
270 If you only want a list of all nodes, use:
276 .List nodes in a cluster
280 Membership information
281 ~~~~~~~~~~~~~~~~~~~~~~
289 [[pvecm_adding_nodes_with_separated_cluster_network]]
290 Adding Nodes with Separated Cluster Network
291 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
293 When adding a node to a cluster with a separated cluster network, you need to
294 use the 'link0' parameter to set the nodes address on that network:
298 # pvecm add IP-ADDRESS-CLUSTER --link0 LOCAL-IP-ADDRESS-LINK0
301 If you want to use the built-in xref:pvecm_redundancy[redundancy] of the
302 Kronosnet transport layer, also use the 'link1' parameter.
304 Using the GUI, you can select the correct interface from the corresponding
305 'Link X' fields in the *Cluster Join* dialog.
307 Remove a Cluster Node
308 ---------------------
310 CAUTION: Read the procedure carefully before proceeding, as it may
311 not be what you want or need.
313 Move all virtual machines from the node. Ensure that you have made copies of any
314 local data or backups that you want to keep. In addition, make sure to remove
315 any scheduled replication jobs to the node to be removed.
317 CAUTION: Failure to remove replication jobs to a node before removing said node
318 will result in the replication job becoming irremovable. Especially note that
319 replication automatically switches direction if a replicated VM is migrated, so
320 by migrating a replicated VM from a node to be deleted, replication jobs will be
321 set up to that node automatically.
323 In the following example, we will remove the node hp4 from the cluster.
325 Log in to a *different* cluster node (not hp4), and issue a `pvecm nodes`
326 command to identify the node ID to remove:
331 Membership information
332 ~~~~~~~~~~~~~~~~~~~~~~
341 At this point, you must power off hp4 and ensure that it will not power on
342 again (in the network) with its current configuration.
344 IMPORTANT: As mentioned above, it is critical to power off the node
345 *before* removal, and make sure that it will *not* power on again
346 (in the existing cluster network) with its current configuration.
347 If you power on the node as it is, the cluster could end up broken,
348 and it could be difficult to restore it to a functioning state.
350 After powering off the node hp4, we can safely remove it from the cluster.
353 hp1# pvecm delnode hp4
357 NOTE: At this point, it is possible that you will receive an error message
358 stating `Could not kill node (error = CS_ERR_NOT_EXIST)`. This does not
359 signify an actual failure in the deletion of the node, but rather a failure in
360 corosync trying to kill an offline node. Thus, it can be safely ignored.
362 Use `pvecm nodes` or `pvecm status` to check the node list again. It should
370 Votequorum information
371 ~~~~~~~~~~~~~~~~~~~~~~
378 Membership information
379 ~~~~~~~~~~~~~~~~~~~~~~
381 0x00000001 1 192.168.15.90 (local)
382 0x00000002 1 192.168.15.91
383 0x00000003 1 192.168.15.92
386 If, for whatever reason, you want this server to join the same cluster again,
389 * do a fresh install of {pve} on it,
391 * then join it, as explained in the previous section.
393 The configuration files for the removed node will still reside in
394 '/etc/pve/nodes/hp4'. Recover any configuration you still need and remove the
395 directory afterwards.
397 NOTE: After removal of the node, its SSH fingerprint will still reside in the
398 'known_hosts' of the other nodes. If you receive an SSH error after rejoining
399 a node with the same IP or hostname, run `pvecm updatecerts` once on the
400 re-added node to update its fingerprint cluster wide.
402 [[pvecm_separate_node_without_reinstall]]
403 Separate a Node Without Reinstalling
404 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
406 CAUTION: This is *not* the recommended method, proceed with caution. Use the
407 previous method if you're unsure.
409 You can also separate a node from a cluster without reinstalling it from
410 scratch. But after removing the node from the cluster, it will still have
411 access to any shared storage. This must be resolved before you start removing
412 the node from the cluster. A {pve} cluster cannot share the exact same
413 storage with another cluster, as storage locking doesn't work over the cluster
414 boundary. Furthermore, it may also lead to VMID conflicts.
416 It's suggested that you create a new storage, where only the node which you want
417 to separate has access. This can be a new export on your NFS or a new Ceph
418 pool, to name a few examples. It's just important that the exact same storage
419 does not get accessed by multiple clusters. After setting up this storage, move
420 all data and VMs from the node to it. Then you are ready to separate the
421 node from the cluster.
423 WARNING: Ensure that all shared resources are cleanly separated! Otherwise you
424 will run into conflicts and problems.
426 First, stop the corosync and pve-cluster services on the node:
429 systemctl stop pve-cluster
430 systemctl stop corosync
433 Start the cluster file system again in local mode:
439 Delete the corosync configuration files:
442 rm /etc/pve/corosync.conf
443 rm -r /etc/corosync/*
446 You can now start the file system again as a normal service:
450 systemctl start pve-cluster
453 The node is now separated from the cluster. You can deleted it from any
454 remaining node of the cluster with:
457 pvecm delnode oldnode
460 If the command fails due to a loss of quorum in the remaining node, you can set
461 the expected votes to 1 as a workaround:
467 And then repeat the 'pvecm delnode' command.
469 Now switch back to the separated node and delete all the remaining cluster
470 files on it. This ensures that the node can be added to another cluster again
475 rm /var/lib/corosync/*
478 As the configuration files from the other nodes are still in the cluster
479 file system, you may want to clean those up too. After making absolutely sure
480 that you have the correct node name, you can simply remove the entire
481 directory recursively from '/etc/pve/nodes/NODENAME'.
483 CAUTION: The node's SSH keys will remain in the 'authorized_key' file. This
484 means that the nodes can still connect to each other with public key
485 authentication. You should fix this by removing the respective keys from the
486 '/etc/pve/priv/authorized_keys' file.
492 {pve} use a quorum-based technique to provide a consistent state among
495 [quote, from Wikipedia, Quorum (distributed computing)]
497 A quorum is the minimum number of votes that a distributed transaction
498 has to obtain in order to be allowed to perform an operation in a
502 In case of network partitioning, state changes requires that a
503 majority of nodes are online. The cluster switches to read-only mode
506 NOTE: {pve} assigns a single vote to each node by default.
512 The cluster network is the core of a cluster. All messages sent over it have to
513 be delivered reliably to all nodes in their respective order. In {pve} this
514 part is done by corosync, an implementation of a high performance, low overhead,
515 high availability development toolkit. It serves our decentralized configuration
516 file system (`pmxcfs`).
518 [[pvecm_cluster_network_requirements]]
522 The {pve} cluster stack requires a reliable network with latencies under 5
523 milliseconds (LAN performance) between all nodes to operate stably. While on
524 setups with a small node count a network with higher latencies _may_ work, this
525 is not guaranteed and gets rather unlikely with more than three nodes and
526 latencies above around 10 ms.
528 The network should not be used heavily by other members, as while corosync does
529 not uses much bandwidth it is sensitive to latency jitters; ideally corosync
530 runs on its own physically separated network. Especially do not use a shared
531 network for corosync and storage (except as a potential low-priority fallback
532 in a xref:pvecm_redundancy[redundant] configuration).
534 Before setting up a cluster, it is good practice to check if the network is fit
535 for that purpose. To ensure that the nodes can connect to each other on the
536 cluster network, you can test the connectivity between them with the `ping`
539 If the {pve} firewall is enabled, ACCEPT rules for corosync will automatically
540 be generated - no manual action is required.
542 NOTE: Corosync used Multicast before version 3.0 (introduced in {pve} 6.0).
543 Modern versions rely on https://kronosnet.org/[Kronosnet] for cluster
544 communication, which, for now, only supports regular UDP unicast.
546 CAUTION: You can still enable Multicast or legacy unicast by setting your
547 transport to `udp` or `udpu` in your xref:pvecm_edit_corosync_conf[corosync.conf],
548 but keep in mind that this will disable all cryptography and redundancy support.
549 This is therefore not recommended.
551 Separate Cluster Network
552 ~~~~~~~~~~~~~~~~~~~~~~~~
554 When creating a cluster without any parameters, the corosync cluster network is
555 generally shared with the web interface and the VMs' network. Depending on
556 your setup, even storage traffic may get sent over the same network. It's
557 recommended to change that, as corosync is a time-critical, real-time
560 Setting Up a New Network
561 ^^^^^^^^^^^^^^^^^^^^^^^^
563 First, you have to set up a new network interface. It should be on a physically
564 separate network. Ensure that your network fulfills the
565 xref:pvecm_cluster_network_requirements[cluster network requirements].
567 Separate On Cluster Creation
568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
570 This is possible via the 'linkX' parameters of the 'pvecm create'
571 command, used for creating a new cluster.
573 If you have set up an additional NIC with a static address on 10.10.10.1/25,
574 and want to send and receive all cluster communication over this interface,
579 pvecm create test --link0 10.10.10.1
582 To check if everything is working properly, execute:
585 systemctl status corosync
588 Afterwards, proceed as described above to
589 xref:pvecm_adding_nodes_with_separated_cluster_network[add nodes with a separated cluster network].
591 [[pvecm_separate_cluster_net_after_creation]]
592 Separate After Cluster Creation
593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
595 You can do this if you have already created a cluster and want to switch
596 its communication to another network, without rebuilding the whole cluster.
597 This change may lead to short periods of quorum loss in the cluster, as nodes
598 have to restart corosync and come up one after the other on the new network.
600 Check how to xref:pvecm_edit_corosync_conf[edit the corosync.conf file] first.
601 Then, open it and you should see a file similar to:
635 provider: corosync_votequorum
639 cluster_name: testcluster
651 NOTE: `ringX_addr` actually specifies a corosync *link address*. The name "ring"
652 is a remnant of older corosync versions that is kept for backwards
655 The first thing you want to do is add the 'name' properties in the node entries,
656 if you do not see them already. Those *must* match the node name.
658 Then replace all addresses from the 'ring0_addr' properties of all nodes with
659 the new addresses. You may use plain IP addresses or hostnames here. If you use
660 hostnames, ensure that they are resolvable from all nodes (see also
661 xref:pvecm_corosync_addresses[Link Address Types]).
663 In this example, we want to switch cluster communication to the
664 10.10.10.0/25 network, so we change the 'ring0_addr' of each node respectively.
666 NOTE: The exact same procedure can be used to change other 'ringX_addr' values
667 as well. However, we recommend only changing one link address at a time, so
668 that it's easier to recover if something goes wrong.
670 After we increase the 'config_version' property, the new configuration file
685 ring0_addr: 10.10.10.2
692 ring0_addr: 10.10.10.3
699 ring0_addr: 10.10.10.1
705 provider: corosync_votequorum
709 cluster_name: testcluster
721 Then, after a final check to see that all changed information is correct, we
722 save it and once again follow the
723 xref:pvecm_edit_corosync_conf[edit corosync.conf file] section to bring it into
726 The changes will be applied live, so restarting corosync is not strictly
727 necessary. If you changed other settings as well, or notice corosync
728 complaining, you can optionally trigger a restart.
730 On a single node execute:
734 systemctl restart corosync
737 Now check if everything is okay:
741 systemctl status corosync
744 If corosync begins to work again, restart it on all other nodes too.
745 They will then join the cluster membership one by one on the new network.
747 [[pvecm_corosync_addresses]]
751 A corosync link address (for backwards compatibility denoted by 'ringX_addr' in
752 `corosync.conf`) can be specified in two ways:
754 * **IPv4/v6 addresses** can be used directly. They are recommended, since they
755 are static and usually not changed carelessly.
757 * **Hostnames** will be resolved using `getaddrinfo`, which means that by
758 default, IPv6 addresses will be used first, if available (see also
759 `man gai.conf`). Keep this in mind, especially when upgrading an existing
762 CAUTION: Hostnames should be used with care, since the addresses they
763 resolve to can be changed without touching corosync or the node it runs on -
764 which may lead to a situation where an address is changed without thinking
765 about implications for corosync.
767 A separate, static hostname specifically for corosync is recommended, if
768 hostnames are preferred. Also, make sure that every node in the cluster can
769 resolve all hostnames correctly.
771 Since {pve} 5.1, while supported, hostnames will be resolved at the time of
772 entry. Only the resolved IP is saved to the configuration.
774 Nodes that joined the cluster on earlier versions likely still use their
775 unresolved hostname in `corosync.conf`. It might be a good idea to replace
776 them with IPs or a separate hostname, as mentioned above.
783 Corosync supports redundant networking via its integrated Kronosnet layer by
784 default (it is not supported on the legacy udp/udpu transports). It can be
785 enabled by specifying more than one link address, either via the '--linkX'
786 parameters of `pvecm`, in the GUI as **Link 1** (while creating a cluster or
787 adding a new node) or by specifying more than one 'ringX_addr' in
790 NOTE: To provide useful failover, every link should be on its own
791 physical network connection.
793 Links are used according to a priority setting. You can configure this priority
794 by setting 'knet_link_priority' in the corresponding interface section in
795 `corosync.conf`, or, preferably, using the 'priority' parameter when creating
796 your cluster with `pvecm`:
799 # pvecm create CLUSTERNAME --link0 10.10.10.1,priority=15 --link1 10.20.20.1,priority=20
802 This would cause 'link1' to be used first, since it has the higher priority.
804 If no priorities are configured manually (or two links have the same priority),
805 links will be used in order of their number, with the lower number having higher
808 Even if all links are working, only the one with the highest priority will see
809 corosync traffic. Link priorities cannot be mixed, meaning that links with
810 different priorities will not be able to communicate with each other.
812 Since lower priority links will not see traffic unless all higher priorities
813 have failed, it becomes a useful strategy to specify networks used for
814 other tasks (VMs, storage, etc.) as low-priority links. If worst comes to
815 worst, a higher latency or more congested connection might be better than no
818 Adding Redundant Links To An Existing Cluster
819 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
821 To add a new link to a running configuration, first check how to
822 xref:pvecm_edit_corosync_conf[edit the corosync.conf file].
824 Then, add a new 'ringX_addr' to every node in the `nodelist` section. Make
825 sure that your 'X' is the same for every node you add it to, and that it is
826 unique for each node.
828 Lastly, add a new 'interface', as shown below, to your `totem`
829 section, replacing 'X' with the link number chosen above.
831 Assuming you added a link with number 1, the new configuration file could look
846 ring0_addr: 10.10.10.2
847 ring1_addr: 10.20.20.2
854 ring0_addr: 10.10.10.3
855 ring1_addr: 10.20.20.3
862 ring0_addr: 10.10.10.1
863 ring1_addr: 10.20.20.1
869 provider: corosync_votequorum
873 cluster_name: testcluster
887 The new link will be enabled as soon as you follow the last steps to
888 xref:pvecm_edit_corosync_conf[edit the corosync.conf file]. A restart should not
889 be necessary. You can check that corosync loaded the new link using:
892 journalctl -b -u corosync
895 It might be a good idea to test the new link by temporarily disconnecting the
896 old link on one node and making sure that its status remains online while
903 If you see a healthy cluster state, it means that your new link is being used.
906 Role of SSH in {pve} Clusters
907 -----------------------------
909 {pve} utilizes SSH tunnels for various features.
911 * Proxying console/shell sessions (node and guests)
913 When using the shell for node B while being connected to node A, connects to a
914 terminal proxy on node A, which is in turn connected to the login shell on node
915 B via a non-interactive SSH tunnel.
917 * VM and CT memory and local-storage migration in 'secure' mode.
919 During the migration, one or more SSH tunnel(s) are established between the
920 source and target nodes, in order to exchange migration information and
921 transfer memory and disk contents.
923 * Storage replication
925 .Pitfalls due to automatic execution of `.bashrc` and siblings
928 In case you have a custom `.bashrc`, or similar files that get executed on
929 login by the configured shell, `ssh` will automatically run it once the session
930 is established successfully. This can cause some unexpected behavior, as those
931 commands may be executed with root permissions on any of the operations
932 described above. This can cause possible problematic side-effects!
934 In order to avoid such complications, it's recommended to add a check in
935 `/root/.bashrc` to make sure the session is interactive, and only then run
938 You can add this snippet at the beginning of your `.bashrc` file:
941 # Early exit if not running interactively to avoid side-effects!
950 Corosync External Vote Support
951 ------------------------------
953 This section describes a way to deploy an external voter in a {pve} cluster.
954 When configured, the cluster can sustain more node failures without
955 violating safety properties of the cluster communication.
957 For this to work, there are two services involved:
959 * A QDevice daemon which runs on each {pve} node
961 * An external vote daemon which runs on an independent server
963 As a result, you can achieve higher availability, even in smaller setups (for
966 QDevice Technical Overview
967 ~~~~~~~~~~~~~~~~~~~~~~~~~~
969 The Corosync Quorum Device (QDevice) is a daemon which runs on each cluster
970 node. It provides a configured number of votes to the cluster's quorum
971 subsystem, based on an externally running third-party arbitrator's decision.
972 Its primary use is to allow a cluster to sustain more node failures than
973 standard quorum rules allow. This can be done safely as the external device
974 can see all nodes and thus choose only one set of nodes to give its vote.
975 This will only be done if said set of nodes can have quorum (again) after
976 receiving the third-party vote.
978 Currently, only 'QDevice Net' is supported as a third-party arbitrator. This is
979 a daemon which provides a vote to a cluster partition, if it can reach the
980 partition members over the network. It will only give votes to one partition
981 of a cluster at any time.
982 It's designed to support multiple clusters and is almost configuration and
983 state free. New clusters are handled dynamically and no configuration file
984 is needed on the host running a QDevice.
986 The only requirements for the external host are that it needs network access to
987 the cluster and to have a corosync-qnetd package available. We provide a package
988 for Debian based hosts, and other Linux distributions should also have a package
989 available through their respective package manager.
991 NOTE: Unlike corosync itself, a QDevice connects to the cluster over TCP/IP.
992 The daemon can also run outside the LAN of the cluster and isn't limited to the
993 low latencies requirements of corosync.
998 We support QDevices for clusters with an even number of nodes and recommend
999 it for 2 node clusters, if they should provide higher availability.
1000 For clusters with an odd node count, we currently discourage the use of
1001 QDevices. The reason for this is the difference in the votes which the QDevice
1002 provides for each cluster type. Even numbered clusters get a single additional
1003 vote, which only increases availability, because if the QDevice
1004 itself fails, you are in the same position as with no QDevice at all.
1006 On the other hand, with an odd numbered cluster size, the QDevice provides
1007 '(N-1)' votes -- where 'N' corresponds to the cluster node count. This
1008 alternative behavior makes sense; if it had only one additional vote, the
1009 cluster could get into a split-brain situation. This algorithm allows for all
1010 nodes but one (and naturally the QDevice itself) to fail. However, there are two
1013 * If the QNet daemon itself fails, no other node may fail or the cluster
1014 immediately loses quorum. For example, in a cluster with 15 nodes, 7
1015 could fail before the cluster becomes inquorate. But, if a QDevice is
1016 configured here and it itself fails, **no single node** of the 15 may fail.
1017 The QDevice acts almost as a single point of failure in this case.
1019 * The fact that all but one node plus QDevice may fail sounds promising at
1020 first, but this may result in a mass recovery of HA services, which could
1021 overload the single remaining node. Furthermore, a Ceph server will stop
1022 providing services if only '((N-1)/2)' nodes or less remain online.
1024 If you understand the drawbacks and implications, you can decide yourself if
1025 you want to use this technology in an odd numbered cluster setup.
1030 We recommend running any daemon which provides votes to corosync-qdevice as an
1031 unprivileged user. {pve} and Debian provide a package which is already
1032 configured to do so.
1033 The traffic between the daemon and the cluster must be encrypted to ensure a
1034 safe and secure integration of the QDevice in {pve}.
1036 First, install the 'corosync-qnetd' package on your external server
1039 external# apt install corosync-qnetd
1042 and the 'corosync-qdevice' package on all cluster nodes
1045 pve# apt install corosync-qdevice
1048 After doing this, ensure that all the nodes in the cluster are online.
1050 You can now set up your QDevice by running the following command on one
1054 pve# pvecm qdevice setup <QDEVICE-IP>
1057 The SSH key from the cluster will be automatically copied to the QDevice.
1059 NOTE: Make sure to setup key-based access for the root user on your external
1060 server, or temporarily allow root login with password during the setup phase.
1061 If you receive an error such as 'Host key verification failed.' at this
1062 stage, running `pvecm updatecerts` could fix the issue.
1064 After all the steps have successfully completed, you will see "Done". You can
1065 verify that the QDevice has been set up with:
1072 Votequorum information
1073 ~~~~~~~~~~~~~~~~~~~~~
1078 Flags: Quorate Qdevice
1080 Membership information
1081 ~~~~~~~~~~~~~~~~~~~~~~
1082 Nodeid Votes Qdevice Name
1083 0x00000001 1 A,V,NMW 192.168.22.180 (local)
1084 0x00000002 1 A,V,NMW 192.168.22.181
1085 0x00000000 1 Qdevice
1089 [[pvecm_qdevice_status_flags]]
1090 QDevice Status Flags
1091 ^^^^^^^^^^^^^^^^^^^^
1093 The status output of the QDevice, as seen above, will usually contain three
1096 * `A` / `NA`: Alive or Not Alive. Indicates if the communication to the external
1097 `corosync-qnetd` daemon works.
1098 * `V` / `NV`: If the QDevice will cast a vote for the node. In a split-brain
1099 situation, where the corosync connection between the nodes is down, but they
1100 both can still communicate with the external `corosync-qnetd` daemon,
1101 only one node will get the vote.
1102 * `MW` / `NMW`: Master wins (`MV`) or not (`NMW`). Default is `NMW`, see
1103 footnote:[`votequorum_qdevice_master_wins` manual page
1104 https://manpages.debian.org/bookworm/libvotequorum-dev/votequorum_qdevice_master_wins.3.en.html].
1105 * `NR`: QDevice is not registered.
1107 NOTE: If your QDevice is listed as `Not Alive` (`NA` in the output above),
1108 ensure that port `5403` (the default port of the qnetd server) of your external
1109 server is reachable via TCP/IP!
1112 Frequently Asked Questions
1113 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1118 In case of a tie, where two same-sized cluster partitions cannot see each other
1119 but can see the QDevice, the QDevice chooses one of those partitions randomly
1120 and provides a vote to it.
1122 Possible Negative Implications
1123 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1125 For clusters with an even node count, there are no negative implications when
1126 using a QDevice. If it fails to work, it is the same as not having a QDevice
1129 Adding/Deleting Nodes After QDevice Setup
1130 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1132 If you want to add a new node or remove an existing one from a cluster with a
1133 QDevice setup, you need to remove the QDevice first. After that, you can add or
1134 remove nodes normally. Once you have a cluster with an even node count again,
1135 you can set up the QDevice again as described previously.
1137 Removing the QDevice
1138 ^^^^^^^^^^^^^^^^^^^^
1140 If you used the official `pvecm` tool to add the QDevice, you can remove it
1144 pve# pvecm qdevice remove
1149 //There is still stuff to add here
1152 Corosync Configuration
1153 ----------------------
1155 The `/etc/pve/corosync.conf` file plays a central role in a {pve} cluster. It
1156 controls the cluster membership and its network.
1157 For further information about it, check the corosync.conf man page:
1163 For node membership, you should always use the `pvecm` tool provided by {pve}.
1164 You may have to edit the configuration file manually for other changes.
1165 Here are a few best practice tips for doing this.
1167 [[pvecm_edit_corosync_conf]]
1171 Editing the corosync.conf file is not always very straightforward. There are
1172 two on each cluster node, one in `/etc/pve/corosync.conf` and the other in
1173 `/etc/corosync/corosync.conf`. Editing the one in our cluster file system will
1174 propagate the changes to the local one, but not vice versa.
1176 The configuration will get updated automatically, as soon as the file changes.
1177 This means that changes which can be integrated in a running corosync will take
1178 effect immediately. Thus, you should always make a copy and edit that instead,
1179 to avoid triggering unintended changes when saving the file while editing.
1183 cp /etc/pve/corosync.conf /etc/pve/corosync.conf.new
1186 Then, open the config file with your favorite editor, such as `nano` or
1187 `vim.tiny`, which come pre-installed on every {pve} node.
1189 NOTE: Always increment the 'config_version' number after configuration changes;
1190 omitting this can lead to problems.
1192 After making the necessary changes, create another copy of the current working
1193 configuration file. This serves as a backup if the new configuration fails to
1194 apply or causes other issues.
1198 cp /etc/pve/corosync.conf /etc/pve/corosync.conf.bak
1201 Then replace the old configuration file with the new one:
1204 mv /etc/pve/corosync.conf.new /etc/pve/corosync.conf
1207 You can check if the changes could be applied automatically, using the following
1211 systemctl status corosync
1212 journalctl -b -u corosync
1215 If the changes could not be applied automatically, you may have to restart the
1216 corosync service via:
1219 systemctl restart corosync
1222 On errors, check the troubleshooting section below.
1227 Issue: 'quorum.expected_votes must be configured'
1228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1230 When corosync starts to fail and you get the following message in the system log:
1234 corosync[1647]: [QUORUM] Quorum provider: corosync_votequorum failed to initialize.
1235 corosync[1647]: [SERV ] Service engine 'corosync_quorum' failed to load for reason
1236 'configuration error: nodelist or quorum.expected_votes must be configured!'
1240 It means that the hostname you set for a corosync 'ringX_addr' in the
1241 configuration could not be resolved.
1243 Write Configuration When Not Quorate
1244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1246 If you need to change '/etc/pve/corosync.conf' on a node with no quorum, and you
1247 understand what you are doing, use:
1253 This sets the expected vote count to 1 and makes the cluster quorate. You can
1254 then fix your configuration, or revert it back to the last working backup.
1256 This is not enough if corosync cannot start anymore. In that case, it is best to
1257 edit the local copy of the corosync configuration in
1258 '/etc/corosync/corosync.conf', so that corosync can start again. Ensure that on
1259 all nodes, this configuration has the same content to avoid split-brain
1263 [[pvecm_corosync_conf_glossary]]
1264 Corosync Configuration Glossary
1265 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1268 This names the different link addresses for the Kronosnet connections between
1275 It is obvious that a cluster is not quorate when all nodes are
1276 offline. This is a common case after a power failure.
1278 NOTE: It is always a good idea to use an uninterruptible power supply
1279 (``UPS'', also called ``battery backup'') to avoid this state, especially if
1282 On node startup, the `pve-guests` service is started and waits for
1283 quorum. Once quorate, it starts all guests which have the `onboot`
1286 When you turn on nodes, or when power comes back after power failure,
1287 it is likely that some nodes will boot faster than others. Please keep in
1288 mind that guest startup is delayed until you reach quorum.
1291 [[pvecm_next_id_range]]
1292 Guest VMID Auto-Selection
1293 ------------------------
1295 When creating new guests the web interface will ask the backend for a free VMID
1296 automatically. The default range for searching is `100` to `1000000` (lower
1297 than the maximal allowed VMID enforced by the schema).
1299 Sometimes admins either want to allocate new VMIDs in a separate range, for
1300 example to easily separate temporary VMs with ones that choose a VMID manually.
1301 Other times its just desired to provided a stable length VMID, for which
1302 setting the lower boundary to, for example, `100000` gives much more room for.
1304 To accommodate this use case one can set either lower, upper or both boundaries
1305 via the `datacenter.cfg` configuration file, which can be edited in the web
1306 interface under 'Datacenter' -> 'Options'.
1308 NOTE: The range is only used for the next-id API call, so it isn't a hard
1314 Migrating virtual guests to other nodes is a useful feature in a
1315 cluster. There are settings to control the behavior of such
1316 migrations. This can be done via the configuration file
1317 `datacenter.cfg` or for a specific migration via API or command-line
1320 It makes a difference if a guest is online or offline, or if it has
1321 local resources (like a local disk).
1323 For details about virtual machine migration, see the
1324 xref:qm_migration[QEMU/KVM Migration Chapter].
1326 For details about container migration, see the
1327 xref:pct_migration[Container Migration Chapter].
1332 The migration type defines if the migration data should be sent over an
1333 encrypted (`secure`) channel or an unencrypted (`insecure`) one.
1334 Setting the migration type to `insecure` means that the RAM content of a
1335 virtual guest is also transferred unencrypted, which can lead to
1336 information disclosure of critical data from inside the guest (for
1337 example, passwords or encryption keys).
1339 Therefore, we strongly recommend using the secure channel if you do
1340 not have full control over the network and can not guarantee that no
1341 one is eavesdropping on it.
1343 NOTE: Storage migration does not follow this setting. Currently, it
1344 always sends the storage content over a secure channel.
1346 Encryption requires a lot of computing power, so this setting is often
1347 changed to `insecure` to achieve better performance. The impact on
1348 modern systems is lower because they implement AES encryption in
1349 hardware. The performance impact is particularly evident in fast
1350 networks, where you can transfer 10 Gbps or more.
1355 By default, {pve} uses the network in which cluster communication
1356 takes place to send the migration traffic. This is not optimal both because
1357 sensitive cluster traffic can be disrupted and this network may not
1358 have the best bandwidth available on the node.
1360 Setting the migration network parameter allows the use of a dedicated
1361 network for all migration traffic. In addition to the memory,
1362 this also affects the storage traffic for offline migrations.
1364 The migration network is set as a network using CIDR notation. This
1365 has the advantage that you don't have to set individual IP addresses
1366 for each node. {pve} can determine the real address on the
1367 destination node from the network specified in the CIDR form. To
1368 enable this, the network must be specified so that each node has exactly one
1369 IP in the respective network.
1374 We assume that we have a three-node setup, with three separate
1375 networks. One for public communication with the Internet, one for
1376 cluster communication, and a very fast one, which we want to use as a
1377 dedicated network for migration.
1379 A network configuration for such a setup might look as follows:
1382 iface eno1 inet manual
1386 iface vmbr0 inet static
1387 address 192.X.Y.57/24
1395 iface eno2 inet static
1400 iface eno3 inet static
1404 Here, we will use the network 10.1.2.0/24 as a migration network. For
1405 a single migration, you can do this using the `migration_network`
1406 parameter of the command-line tool:
1409 # qm migrate 106 tre --online --migration_network 10.1.2.0/24
1412 To configure this as the default network for all migrations in the
1413 cluster, set the `migration` property of the `/etc/pve/datacenter.cfg`
1417 # use dedicated migration network
1418 migration: secure,network=10.1.2.0/24
1421 NOTE: The migration type must always be set when the migration network
1422 is set in `/etc/pve/datacenter.cfg`.
1426 include::pve-copyright.adoc[]