From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from gate001.proxmox.com (gate001.proxmox.com [45.144.208.40]) by lore.proxmox.com (Postfix) with ESMTPS id 6ED961FF0DA for ; Thu, 19 Nov 2026 20:25:27 +0100 (CET) Received: from gate001.proxmox.com (localhost.localdomain [127.0.0.1]) by gate001.proxmox.com (Proxmox) with ESMTP id 19DD821BE7; Thu, 19 Nov 2026 20:20:25 +0100 (CET) From: Thomas Lamprecht To: pve-devel@lists.proxmox.com Subject: [PATCH docs v5 38/41] pvecm: document the cluster ID and how to get a new one Date: Thu, 19 Nov 2026 20:19:10 +0100 Message-ID: <20261119191914.2336934-39-t.lamprecht@proxmox.com> X-Mailer: git-send-email 2.47.3 In-Reply-To: <20261119191914.2336934-1-t.lamprecht@proxmox.com> References: <20261119191914.2336934-1-t.lamprecht@proxmox.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Bm-Milter-Handled: 55990f41-d878-4baa-be0a-ee34c49e34d2 X-Bm-Transport-Timestamp: 1795115961508 X-SPAM-LEVEL: Spam detection results: 0 AWL 0.155 Adjusted score from AWL reputation of From: address DMARC_MISSING 0.1 Missing DMARC policy KAM_DMARC_STATUS 0.01 Test Rule for DKIM or SPF Failure with Strict Alignment (newer systems) RCVD_IN_DNSWL_MED -2.3 Sender listed at https://www.dnswl.org/, medium trust SPF_HELO_NONE 0.001 SPF: HELO does not publish an SPF Record SPF_PASS -0.001 SPF: sender matches SPF record Message-ID-Hash: ZZOIZDBZJH7MHAQHMUAYKTQ2WBQQMBIL X-Message-ID-Hash: ZZOIZDBZJH7MHAQHMUAYKTQ2WBQQMBIL X-MailFrom: t.lamprecht@proxmox.com X-Mailman-Rule-Misses: dmarc-mitigation; no-senders; approved; loop; banned-address; emergency; member-moderation; nonmember-moderation; administrivia; implicit-dest; max-recipients; max-size; news-moderation; no-subject; digests; suspicious-header X-Mailman-Version: 3.3.10 Precedence: list List-Id: Proxmox VE development discussion List-Help: List-Owner: List-Post: List-Subscribe: List-Unsubscribe: Since Proxmox VE 9.3, clusters and standalone nodes have a random ID in /etc/pve/cluster-id. Tools need it alongside the guest uid because guests in different clusters can share a uid, for example after a remote migration that keeps the source guest. Explain where to read the ID, why generation can be delayed, and how creating or joining a cluster affects it. pmxcfs protects the file against ordinary changes from user space. Cloned installations and nodes separated without reinstalling keep their old ID. Document removal in forced local mode to get a new one, and include it in the existing node-separation procedure. Warn against doing this on a cluster member: the removal can propagate and cause the entire cluster to generate a new ID. Signed-off-by: Thomas Lamprecht Reviewed-by: Dominik Csapak Tested-by: Dominik Csapak --- changes v4 -> v5: * 'adopts' instead of 'takes over' for a node that gets the ID of its cluster changes v3 -> v4: * the web interface shows the ID in the datacenter summary, not in the 'Guest Identity' window * a node without a cluster that cannot store its ID starts without one * what happens if the generating node gets cut off, and when its ID can replace the one of the others * the reasons for a missing ID as a list, with what to do if a directory takes up the path of the file * the API also omits the field if reading the file fails * the anchor is pvecm_cluster_id changes v2 -> v3: * the file can exist without a valid ID, and forced local mode also allows moving it away pmxcfs.adoc | 1 + pvecm.adoc | 106 ++++++++++++++++++++++++++++++++++++++++++++++++++++ vzdump.adoc | 3 ++ 3 files changed, 110 insertions(+) diff --git a/pmxcfs.adoc b/pmxcfs.adoc index 1128425..be22131 100644 --- a/pmxcfs.adoc +++ b/pmxcfs.adoc @@ -95,6 +95,7 @@ Files |======= |`authkey.pub` | Public key used by the ticket system |`ceph.conf` | Ceph configuration file (note: /etc/ceph/ceph.conf is a symbolic link to this) +|`cluster-id` | Random ID of the cluster, generated by pmxcfs and read-only (see xref:pvecm_cluster_id[Cluster ID]) |`corosync.conf` | Corosync cluster configuration file (prior to {pve} 4.x, this file was called cluster.conf) |`datacenter.cfg` | {pve} datacenter-wide configuration (keyboard layout, proxy, ...) |`domains.cfg` | {pve} authentication domains diff --git a/pvecm.adoc b/pvecm.adoc index cd5a0fb..7a44cbf 100644 --- a/pvecm.adoc +++ b/pvecm.adoc @@ -529,6 +529,14 @@ rm /etc/pve/corosync.conf rm -r /etc/corosync/* ---- +Remove the xref:pvecm_cluster_id[cluster ID], if the cluster has one. +Without this step, the node keeps the ID of its former cluster instead of +generating its own when the file system starts again as a normal service: +[source,bash] +---- +rm /etc/pve/cluster-id +---- + You can now start the file system again as a normal service: [source,bash] ---- @@ -572,6 +580,104 @@ authentication. You should fix this by removing the respective keys from the '/etc/pve/priv/authorized_keys' file. +[[pvecm_cluster_id]] +Cluster ID +---------- + +Since {pve} 9.3, each cluster has a random cluster ID in the same format as the +`uid` of a xref:guest_identity[guest identity]. A node that is not part of a +cluster has an ID of its own. Guests in different clusters can have the same +`uid`, for example after a migration to another cluster that kept the source +guest. The cluster ID tells tools to which cluster the `uid` of a guest belongs. + +The cluster ID is stored in `/etc/pve/cluster-id`: + +---- +# cat /etc/pve/cluster-id +---- + +Through the API, `/cluster/status` returns it in the `cluster-id` field of the +entry of type `cluster`. On a node without a cluster, the field is on the node +entry instead. Reading the cluster ID through the API requires the `Sys.Audit` +privilege on `/`. Users with this privilege also see it in the web interface +(__Datacenter -> Summary__). + +{pve} generates the cluster ID and keeps it unchanged. Creating, changing, +moving or removing `/etc/pve/cluster-id` fails with `Permission denied`. The +only exception is removing or moving it away in local mode (`pmxcfs -l`), as +described below. Nodes running a {pve} version older than 9.3 do not protect the +file. + +A node without a cluster generates its ID when its cluster file system starts. +If it cannot store the ID, for example because the root file system is full, it +starts without one and generates it at the next start of the `pve-cluster` +service. In a cluster, the online node with the lowest node ID generates the ID +once all of the following apply: + +* The cluster is quorate. +* More than half of the configured nodes are online. +* The cluster file systems of the online nodes are in sync. + +If the node that generates the ID loses the connection to the other nodes at +that moment, it reports its ID while the other nodes generate and report another +one. When that node joins again, it normally adopts the ID of the others. +Its own ID replaces theirs only if nothing in `/etc/pve` changed meanwhile and +its clock is a few seconds ahead of theirs. + +Until an ID exists, the API omits the field, and the file is missing or holds +no valid ID. The API also omits the field if reading the file fails. An ID can +be missing for the following reasons: + +* The online node with the lowest node ID runs a {pve} version older than 9.3. + The cluster gets no ID as long as that is the case. +* The online node with the lowest node ID runs {pve} 9.3 or later, but cannot + generate the ID yet. Its `pve-cluster` service then logs the reason in a + message starting with `not generating a cluster ID yet`. +* A directory takes up the path of the file, for example one created on a node + with an older version. {pve} then logs that + `'cluster-id' is not a regular file`. Move the contents of the directory away + and remove it with `rmdir /etc/pve/cluster-id`. A cluster then generates the + ID within a few seconds. A node without a cluster generates it the next time + the `pve-cluster` service starts, for example after + `systemctl restart pve-cluster`. + +Creating a cluster with `pvecm create` keeps the ID of the node, which then +identifies the new cluster. A node that joins a cluster adopts the ID of the +cluster, because joining replaces the configuration of the node in `/etc/pve`. +A copy of an installation, such as a cloned disk or a copy of the cluster file +system database, has the same ID as the original. + +To give a node that is not part of a cluster a new ID, for example after cloning +an installation, follow the steps below. To separate a node from its cluster, +follow xref:pvecm_separate_node_without_reinstall[Separate a Node Without Reinstalling] +instead. That procedure includes the removal of the cluster ID. + +WARNING: Do not remove `/etc/pve/cluster-id` on a node that is still part of a +cluster. When that node rejoins the cluster, the removal is either undone, or it +reaches all nodes and the whole cluster gets a new ID. + +. Stop the cluster file system and start it again in local mode: ++ +---- +# systemctl stop pve-cluster +# pmxcfs -l +---- + +. Remove the cluster ID file: ++ +---- +# rm /etc/pve/cluster-id +---- + +. Stop the cluster file system and start it again as a normal service, which + generates a new ID: ++ +---- +# killall pmxcfs +# systemctl start pve-cluster +---- + + Quorum ------ diff --git a/vzdump.adoc b/vzdump.adoc index cad87dd..dda1b8f 100644 --- a/vzdump.adoc +++ b/vzdump.adoc @@ -684,6 +684,9 @@ The following places show the guest identity. All of them require the The configuration in a backup contains the `meta` property as well. You can show it with `pvesm extractconfig `, for example. +To tell apart guests with the same `uid` in different clusters, combine the +`uid` with the xref:pvecm_cluster_id[cluster ID]. + [[guest_identity_lifecycle]] Lifecycle ~~~~~~~~~ -- 2.47.3