proxmove 1.2

Migrate virtual machines between different Proxmox VM clusters

The Proxmox VM migrator: migrates VMs between different Proxmox VE clusters.

Migrating a virtual machine (VM) on a PVE-cluster from one node to another is implemented in the Proxmox Virtual Environment (PVE). But migrating a VM from one PVE-cluster to another is not.

proxmove helps you move VMs between PVE-clusters with minimal hassle. And if you use ZFS, with minimal downtime too.

Example invocation:

$ proxmove SOURCE_CLUSTER DEST_CLUSTER DEST_NODE DEST_STORAGE VM_NAME1...

But, to get it to work, you’ll need to configure ~/.proxmoverc first. See Configuration.

Additional tips:

• If source and destination filesystems use ZFS, the move is done in two stages, by copying an initial snapshot while the source VM is still up. Combine with --wait-before-stop for additional control.

• Use --debug; it doesn’t flood your screen, but provides useful clues about what it’s doing.

• If your network bridge is different on the DEST_CLUSTER, use --skip-start; that way proxmove “completes” successfully when done with the move. (You’ll still need to change the bridge before starting the VM obviously.)

• If proxmove detects that a move was in progress, it will interactively attempt a resume. For ZFS to ZFS syncs, it will do another initial resync before shutting down the source VM.

Full invocation specification (--help):

usage: proxmove [-c FILENAME] [-n] [--bwlimit MBPS] [--no-verify-ssl]
                [--skip-disks] [--skip-start] [--wait-before-stop]
                [--ssh-ciphers CIPHERS] [--debug] [--ignore-exists]
                [-h] [--version]
                source destination nodeid storage vm [vm ...]

Migrate VMs from one Proxmox cluster to another.

positional arguments:
  source                alias of source cluster
  destination           alias of destination cluster
  nodeid                node on destination cluster
  storage               storage on destination node
  vm                    one or more VMs (guests) to move

optional arguments:
  -c FILENAME, --config FILENAME
                        use alternate configuration inifile
  -n, --dry-run         stop before doing any writes
  --bwlimit MBPS        limit bandwidth in Mbit/s
  --no-verify-ssl       skip ssl verification on the api hosts
  --skip-disks          do the move, but skip copying of the disks;
                        implies --skip-start
  --skip-start          do the move, but do not start the new instance
  --wait-before-stop    prepare the move, but ask for user
                        confirmation before shutting down the old
                        instance (useful if you have to move
                        networks/IPs)
  --ssh-ciphers CIPHERS
                        comma separated list of ssh -c ciphers to
                        prefer, ([e-mail] is supposed to
                        be fast if you have aes on your cpu); set to
                        "-" to use ssh defaults

debug arguments:
  --debug               enables extra debug logging
  --ignore-exists       continue when target VM already exists; allows
                        moving to same cluster

other actions:
  -h, --help            show this help message and exit
  --version             show program's version number and exit

Cluster aliases and storage locations should be defined in
~/.proxmoverc (or see -c option). See the example proxmoverc.sample.
It requires [pve:CLUSTER_ALIAS] sections for the proxmox "api" URL and
[storage:CLUSTER_ALIAS:STORAGE_NAME] sections with "ssh", "path" and
"temp" settings.

Example run

First you need to configure ~/.proxmoverc; see below.

When configured, you can do something like this:

$ proxmove apple-cluster banana-cluster node2 node2-ssd the-vm-to-move
12:12:27: Attempt moving apple-cluster<e1400248> => banana-cluster<6669ad2c> (node 'node2'): the-vm-to-move
12:12:27: - source VM the-vm-to-move@node1<qemu/565/running>
12:12:27: - storage 'ide2': None,media=cdrom (host=<unknown>, guest=<unknown>)
12:12:27: - storage 'virtio0': sharedsan:565/vm-565-disk-1.qcow2,format=qcow2,iops_rd=4000,iops_wr=500,size=50G (host=37.7GiB, guest=50.0GiB)
12:12:27: Creating new VM 'the-vm-to-move' on 'banana-cluster', node 'node2'
12:12:27: - created new VM 'the-vm-to-move--CREATING' as UPID:node2:00005977:1F4D78F4:57C55C0B:qmcreate:126:user@pve:; waiting for it to show up
12:12:34: - created new VM 'the-vm-to-move--CREATING': the-vm-to-move--CREATING@node2<qemu/126/stopped>
12:12:34: Stopping VM the-vm-to-move@node1<qemu/565/running>
12:12:42: - stopped VM the-vm-to-move@node1<qemu/565/stopped>
12:12:42: Ejected (cdrom?) volume 'ide2' (none) added to the-vm-to-move--CREATING@node2<qemu/126/stopped>
12:12:42: Begin copy of 'virtio0' (sharedsan:565/vm-565-disk-1.qcow2,format=qcow2,iops_rd=4000,iops_wr=500,size=50G) to local-ssd
12:12:42: scp(1) copy from '/pool0/san/images/565/vm-565-disk-1.qcow2' (on sharedsan) to '[e-mail]:/node2-ssd/temp/temp-proxmove/vm-126-virtio0'
Warning: Permanently added 'node2.banana-cluster.com' (ECDSA) to the list of known hosts.
vm-565-disk-1.qcow2   100%   50GB   90.5MB/s   09:26
Connection to san.apple-cluster.com closed.
12:22:08: Temp data '/node2-ssd/temp/temp-proxmove/vm-126-virtio0' on local-ssd
12:22:08: Writing data from temp '/node2-ssd/temp/temp-proxmove/vm-126-virtio0' to '/dev/zvol/node2-ssd/vm-126-virtio0' (on local-ssd)
    (100.00/100%)
Connection to node2.banana-cluster.com closed.
12:24:25: Removing temp '/node2-ssd/temp/temp-proxmove/vm-126-virtio0' (on local-ssd)
12:24:26: Starting VM the-vm-to-move@node2<qemu/126/stopped>
12:24:27: - started VM the-vm-to-move@node2<qemu/126/running>
12:24:27: Completed moving apple-cluster<e1400248> => banana-cluster<6669ad2c> (node 'node2'): the-vm-to-move

Before, the-vm-to-move was running on apple-cluster on node1.

Afterwards, the-vm-to-move is running on banana-cluster on node2. The the-vm-to-move on the apple-cluster has been stopped and renamed to the-vm-to-move--MIGRATED.

Configuration

Set up the ~/.proxmoverc config file. First you need to define which clusters you have. For example apple-cluster and banana-cluster.

; Example cluster named "apple-cluster" with 3 storage devices, one
; shared, and two which exist on a single node only.
;
; The user requires various permissions found in the PVEVMAdmin role (VM
; allocate + audit) and PVEAuditor role (Datastore audit) and PVEPoolAdmin
; (to inspect and create pools).
;
[pve:apple-cluster]
api=https://user@pve:[e-mail]:443

; Example cluster named "banana-cluster" with 2 storage devices; both
; storage devices exist on the respective nodes only.
[pve:banana-cluster]
api=https://user@pve:[e-mail]:443

Next, it needs configuration for the storage devices. They are expected to be reachable over SSH; both from the caller and from each other (using SSH-agent forwarding).

The following defines two storage devices for the apple-cluster, one shared and one local to node1 only.

If on sharedsan, the images are probably called something like /pool0/san/images/VMID/vm-VMID-disk1.qcow2, while in Proxmox, they are referred to as sharedsan:VMID/vm-VMID-disk1.qcow2.

[storage:apple-cluster:sharedsan] ; "sharedsan" is available on all nodes
ssh=[e-mail]
path=/pool0/san/images
temp=/pool0/san/private

[storage:apple-cluster:local@node1] ; local disk on node1 only
ssh=[e-mail]
path=/srv/images
temp=/srv/temp

If you use ZFS storage on banana-cluster, the storage config could look like this. Disk volumes exist on the ZFS filesystem node1-ssd/images and node2-ssd/images on the nodes node1 and node2 respectively.

Note that the temp= path is always a regular path.

[storage:banana-cluster:node1-ssd@node1]
ssh=[e-mail]
path=zfs:node1-ssd/images
temp=/node1-ssd/temp

[storage:banana-cluster:node2-ssd@node2]
ssh=[e-mail]
path=zfs:node2-ssd/images
temp=/node2-ssd/temp

The config file looks better with indentation. The author suggests this layout:

[pve:apple-cluster]
...

  [storage:apple-cluster:sharedsan]
  ...
  [storage:apple-cluster:local@node1]
  ...

[pve:banana-cluster]
...

  [storage:banana-cluster:node1-ssd@node1]
  ...

Debugging

If you run into a ResourceException, you may want to patch proxmoxer 1.0.3 to show the HTTP error reason as well.

--- proxmoxer/core.py       2019-04-04 09:13:16.832961589 +0200
+++ proxmoxer/core.py       2019-04-04 09:15:45.434175030 +0200
@@ -75,8 +75,10 @@ class ProxmoxResource(ProxmoxResourceBas
         logger.debug('Status code: %s, output: %s', resp.status_code, resp.content)

         if resp.status_code >= 400:
-            raise ResourceException("{0} {1}: {2}".format(resp.status_code, httplib.responses[resp.status_code],
-                                                          resp.content))
+            raise ResourceException('{0} {1} ("{2}"): {3}'.format(
+                resp.status_code, httplib.responses[resp.status_code],
+                resp.reason,  # reason = textual status_code
+                resp.content))
         elif 200 <= resp.status_code <= 299:
             return self._store["serializer"].loads(resp)

It might reveal a bug (or new feature), like:

proxmoxer.core.ResourceException:
  500 Internal Server Error ("only root can set 'vmgenid' config"):
  b'{"data":null}'

License

proxmove is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 3 or any later version.

Changes

• 1.2 - 2022-09-21

  Changes:

  • Add ZFS snapshot migration for minimal downtime: full sync, poweroff, final sync, poweron.

  • Add –wait-before-stop to allow manual actions after initial sync.

  • Add discard=on on ZFS volumes by default.

  • Prefer rsync over scp.

  • Add ssh_cipher setting (for AES speedup) for zfs-to-zfs transfers.

  • Add workaround to fix boot order on 6.1+6.3 mixed clusters

  • Show –average-rate in pv(1) viewer.

  Tweaks/fixes:

  • Create temp dir automatically, instead of complaining.

  • Set onboot=0 on old-VM after migration so it does not auto-start.

  • Improve various pre-copy checks.

  • Fix confusing double exception, fix various other minor annoyances.

  • Better ssh handling (accept fingerprints, forwarding only when necessary).

  • Better handling of unsupported LXC.

  • Also copy the source pool membership to the destination.

• 1.1 - 2021-06-11

  Features/fixes:

  • Add basic resume support.

  • Allow :port in ssh=user@host:port config.

  • Unmount cdrom media before moving.

  • Fix destination volume naming (use vm-NNN-disk-N instead of vm-NNN-virtioN).

  • Some documentation improvements.

• 1.0 - 2020-01-17

  Features/fixes:

  • Fix disk I/O resource hog overuse on newer PVE clusters.

  • Fix API connection to newer PVE clusters.

  • Add faster ssh cipher by default.

  • Work around Proxmox API timeout.

  • Improved usability through better logging and prepare checks.

• 0.1.0 - 2018-11-22

  Bugs fixed:

  • Show that not seeing a VM is probably a PVEAdmin-permissions issue.

  • Don’t die if image_size cannot be determined.

  • Place the sample files and docs in a /usr/share/…proxmove subdir.

• 0.0.9 - 2017-03-28

  New features:

  • Added –no-verify-ssl option.

  Bugs fixed:

  • Fix str-cast bug with ZFS destination creation.

  • Fix ignoring of non-volume properties like “scsihw”.

• 0.0.8 - 2017-01-26

  New features:

  • Partial LXC container move implemented. Not complete.

  Bugs fixed:

  • Allow ZFS paths to be more than just the a pool name. Instead of e.g. path=zfs:mc9-8-ssd we now also allow path=zfs:rpool/data/images. Closes #7.

• v0.0.7 - 2016-10-07

  Bugs fixed:

  • Instead of trusting on the “size=XXG” which may or may not be present in the storage volume config, it reads the QCOW header or ZFS volume size directly. Also checks that the values are available before attempting a move.

• v0.0.6 - 2016-09-21

  New features:

  • Add –bwlimit in Mbit/s to limit bandwidth during transfer. Will use the scp(1) -l option or for ZFS use the mbuffer(1) auxiliary. As an added bonus mbuffer may improve ZFS send/recv speed through buffering. Closes #4.

  • Add –skip-disks option to skip copying of the disks. Use this if you want to copy the disks manually. Closes #3.

  • Add –skip-start option to skip autostarting of the VM.

  • Adds optional pv(1) pipe viewer progress bar to get ETA in ZFS transfers.

  • Add hidden –debug option for more verbosity.

  • Add hidden –ignore-exists option that allows you to test moves between the same cluster by creating an alias (second config).

  Bugs fixed:

  • Format is not always specified in the properties. If it isn’t, use the image filename suffix when available.

  • Sometimes old values aren’t available in the “pending” list. Don’t croak. Closes #2.

  • Begun refactoring. Testing bettercodehub.com.

  • Also check whether temporary (renamed) VMs exist before starting.

• v0.0.5 - 2016-08-30

  • Added support for ZFS to ZFS disk copy. QCOW2 to ZFS and ZFS to ZFS is now tested.

• v0.0.4 - 2016-08-30

  • Major overhaul of configuration format and other changes.

TODO

• On missing disk (bad config), the zfs send stalls and mbuffer waits for infinity.

• Rename “VM” to “Instance” so “LXC” containers don’t feel misrepresented.

• Communicate with the storage API to check/configure storage (more easily).

• Create a --config command to set up a basic configuration. Combine with storage API reading.

• Fix so LXC-copying works (this is a bit tricky because of the necessity for a temporary image/tarball to install).

• Create a proxmoxer_api example that returns test data, so we can run tests.

• Let it work with pve-zsync – a daemon that syncs data between nodes: https://pve.proxmox.com/wiki/PVE-zsync