Backing up a LXD server

What to backup

When planning to backup a LXD server, consider all the different objects that are stored/managed by LXD:

  • Instances (database records and filesystems)
  • Images (database records, image files and filesystems)
  • Networks (database records and state files)
  • Profiles (database records)
  • Storage volumes (database records and filesystems)

Only backing up the database or only backing up the instances will not get you a fully functional backup.

In some disaster recovery scenarios, that may be reasonable but if your goal is to get back online quickly, consider all the different pieces of LXD you're using.

Full backup

A full backup would include the entirety of /var/lib/lxd or /var/snap/lxd/common/lxd for snap users.

You will also need to appropriately backup any external storage that you made LXD use, this can be LVM volume groups, ZFS zpools or any other resource which isn't directly self-contained to LXD.

Restoring involves stopping LXD on the target server, wiping the lxd directory, restoring the backup and any external dependency it requires.

If not using the snap package and your source system has a /etc/subuid and /etc/subgid file, restoring those or at least the entries inside them for both the lxd and root user is also a good idea (avoids needless shifting of container filesystems).

Then start LXD again and check that everything works fine.

Secondary backup LXD server

LXD supports copying and moving instances and storage volumes between two hosts.

So with a spare server, you can copy your instances and storage volumes to that secondary server every so often, allowing it to act as either an offline spare or just as a storage server that you can copy your instances back from if needed.

Instance backups

The lxc export command can be used to export instances to a backup tarball. Those tarballs will include all snapshots by default and an "optimized" tarball can be obtained if you know that you'll be restoring on a LXD server using the same storage pool backend.

You can use any compressor installed on the server using the --compression flag. There is no validation on the LXD side, any command that is available to LXD and supports -c for stdout should work.

Those tarballs can be saved any way you want on any filesystem you want and can be imported back into LXD using the lxc import command.

Disaster recovery

LXD provides the lxd recover command (note the the lxd command rather than the normal lxc command). This is an interactive CLI tool that will attempt to scan all storage pools that exist in the database looking for missing volumes that can be recovered. It also provides the ability for the user to specify the details of any unknown storage pools (those that exist on disk but do not exist in the database) and it will attempt to scan those too.

Because LXD maintains a backup.yaml file in each instance's storage volume which contains all necessary information to recover a given instance (including instance configuration, attached devices, storage volume and pool configuration) it can be used to rebuild the instance, storage volume and storage pool database records.

The lxd recover tool will attempt to mount the storage pool (if not already mounted) and scan it for unknown volumes that look like they are associated with LXD. For each instance volume LXD will attempt to mount it and access the backup.yaml file. From there it will perform some consistency checks to compare what is in the backup.yaml file with what is actually on disk (such as matching snapshots) and if all checks out then the database records are recreated.

If the storage pool database record also needs to be created then it will prefer to use an instance backup.yaml file as the basis of its config, rather than what the user provided during the discovery phase, however if not available then it will fallback to restoring the pool's database record with what was provided by the user.