---
discourse: lxc:[Online VM live-migration (QEMU to QEMU)](16635)
---

(howto-instances-migrate)=
# How to migrate LXD instances between servers

If you use the LXD client, you can migrate or copy instances from one LXD server (remote or local) to another.

```{note}
{ref}`Remote servers <remotes>` are a concept of the LXD client.
Therefore, there is no direct equivalent for migrating instances between servers in the API or the UI.

However, you can {ref}`export an instance <instances-backup-export-instance>` from one server and {ref}`import it <instances-backup-import-instance>` to another server.
```

## Migrate instances

To migrate an instance (move it from one LXD server to another) using the CLI, use the [`lxc move`](lxc_move.md) command:

    lxc move [<source_remote>:]<source_instance_name> <target_remote>:[<target_instance_name>]

When migrating a container, you must stop it first.
See {ref}`live-migration-containers` for more information.

When migrating a virtual machine, you must either enable {ref}`live-migration-vms` or stop it first.

## Copy instances

Use the [`lxc copy`](lxc_copy.md) command if you want to duplicate the instance instead of migrating it:

    lxc copy [<source_remote>:]<source_instance_name> <target_remote>:[<target_instance_name>]

If the volume already exists in the target location, use the `--refresh` flag to update the copy. To learn about the benefits, see: {ref}`storage-optimized-volume-transfer`.

## Migrate and copy options

For both migrating and copying instances, you don't need to specify the source remote if it is your default remote, and you can leave out the target instance name if you want to use the same instance name on the target remote server.

If you want to migrate the instance to a specific cluster member, specify that member's name with the `--target` flag.
In this case, do not specify the source and target remote.

You can add the `--mode` flag to choose a transfer mode, depending on your network setup:

`pull` (default)
: Instruct the target server to connect to the source server and pull the respective instance.

`push`
: Instruct the source server to connect to the target server and push the instance.

`relay`
: Instruct the client to connect to both the source and the target server and transfer the data through the client.

If you need to adapt the configuration for the instance to run on the target server, you can either specify the new configuration directly (using `--config`, `--device`, `--storage` or `--target-project`) or through profiles (using `--no-profiles` or `--profile`). See [`lxc move --help`](lxc_move.md) for all available flags.

(live-migration)=
## Live migration

Live migration means migrating an instance to another server while it is running. This method is supported for virtual machines. For containers, there is limited support.

(live-migration-vms)=
### Live migration for virtual machines

Virtual machines can be migrated to another server while they are running, thus avoiding any downtime.

For a virtual machine to be eligible for live migration, it must meet the following criteria:

- It must have support for stateful migration enabled. To enable this, set {config:option}`instance-migration:migration.stateful` to `true` on the virtual machine. This setting can only be updated when the machine is stopped. Thus, be sure to configure this setting before you need to live-migrate:

  ```
  lxc config set <instance-name> migration.stateful=true
  ```

  ```{note}
  When {config:option}`instance-migration:migration.stateful` is enabled in LXD, virtiofs shares are disabled, and files are only shared via the 9P protocol. Consequently, guest OSes lacking 9P support, such as CentOS 8, cannot share files with the host unless stateful migration is disabled. Additionally, the `lxd-agent` will not function for these guests under these conditions.
  ```

- When using a local pool, the {config:option}`device-disk-device-conf:size.state` of the virtual machine's root disk device must be set to at least the size of the virtual machine's {config:option}`instance-resource-limits:limits.memory` setting.

  ```{note}
  If you are using a remote storage pool like Ceph RBD to back your instance, you don't need to set {config:option}`device-disk-device-conf:size.state` to perform live migration.
  ```

- The virtual machine must not depend on any resources specific to its current host, such as local storage or a local (non-OVN) bridge network.

(live-migration-containers)=
### Live migration for containers

For containers, there is limited support for live migration using [{abbr}`CRIU (Checkpoint/Restore in Userspace)`](https://criu.org/Main_Page).
However, because of extensive kernel dependencies, only very basic containers (non-`systemd` containers without a network device) can be migrated reliably.
In most real-world scenarios, you should stop the container, migrate it, then start it again.

If you want to use live migration for containers, you must enable CRIU on both the source and the target server.
If you are using the snap, use the following commands to enable CRIU:

    snap set lxd criu.enable=true
    sudo systemctl reload snap.lxd.daemon

Otherwise, make sure you have CRIU installed on both systems.

To optimize the memory transfer for a container, set the {config:option}`instance-migration:migration.incremental.memory` property to `true` to make use of the pre-copy features in CRIU.
With this configuration, LXD instructs CRIU to perform a series of memory dumps for the container.
After each dump, LXD sends the memory dump to the specified remote.
In an ideal scenario, each memory dump will decrease the delta to the previous memory dump, thereby increasing the percentage of memory that is already synced.
When the percentage of synced memory is equal to or greater than the threshold specified via {config:option}`instance-migration:migration.incremental.memory.goal`, or the maximum number of allowed iterations specified via {config:option}`instance-migration:migration.incremental.memory.iterations` is reached, LXD instructs CRIU to perform a final memory dump and transfers it.

## Temporarily migrate all instances from a cluster member

For LXD servers that are members of a cluster, you can use the evacuate and restore operations to temporarily migrate all instances from one cluster member to another. These operations can also live-migrate eligible instances.

For more information, see: {ref}`cluster-evacuate-restore`.

## Related topics

How-to guides:

- {ref}`Migrate instances in a cluster <howto-cluster-manage-instance-migrate>`
- {ref}`howto-projects-work-move-instance`
- {ref}`Import machines to LXD instances <import-machines-to-instances>`
- {ref}`secondary-backup-server`
- {ref}`instances-backup-export-instance`
- {ref}`instances-backup-import-instance`
- {ref}`Move or copy storage volumes <howto-storage-move-volume>`