LXD supports creating and managing storage pools and storage volumes. General keys are top-level. Driver specific keys are namespaced by driver name. Volume keys apply to any volume created in the pool unless the value is overridden on a per-volume basis.
|size||string||appropriate driver and source||0||storage||Size of the storage pool in bytes (suffixes supported). (Currently valid for loop based pools and zfs.)|
|source||string||-||-||storage||Path to block device or loop file or filesystem entry|
|btrfs.mount_options||string||btrfs driver||user_subvol_rm_allowed||storage_btrfs_mount_options||Mount options for block devices|
|ceph.cluster_name||string||ceph driver||ceph||storage_driver_ceph||Name of the ceph cluster in which to create new storage pools.|
|ceph.osd.force_reuse||bool||ceph driver||false||storage_ceph_force_osd_reuse||Force using an osd storage pool that is already in use by another LXD instance.|
|ceph.osd.pg_num||string||ceph driver||32||storage_driver_ceph||Number of placement groups for the osd storage pool.|
|ceph.osd.pool_name||string||ceph driver||name of the pool||storage_driver_ceph||Name of the osd storage pool.|
|ceph.osd.data_pool_name||string||ceph driver||-||storage_driver_ceph||Name of the osd data pool.|
|ceph.rbd.clone_copy||string||ceph driver||true||storage_driver_ceph||Whether to use RBD lightweight clones rather than full dataset copies.|
|ceph.user.name||string||ceph driver||admin||storage_ceph_user_name||The ceph user to use when creating storage pools and volumes.|
|cephfs.cluster_name||string||cephfs driver||ceph||storage_driver_cephfs||Name of the ceph cluster in which to create new storage pools.|
|cephfs.path||string||cephfs driver||/||storage_driver_cephfs||The base path for the CEPHFS mount|
|cephfs.user.name||string||cephfs driver||admin||storage_driver_cephfs||The ceph user to use when creating storage pools and volumes.|
|lvm.thinpool_name||string||lvm driver||LXDThinPool||storage||Thin pool where volumes are created.|
|lvm.use_thinpool||bool||lvm driver||true||storage_lvm_use_thinpool||Whether the storage pool uses a thinpool for logical volumes.|
|lvm.vg_name||string||lvm driver||name of the pool||storage||Name of the volume group to create.|
|lvm.vg.force_reuse||bool||lvm driver||false||storage_lvm_vg_force_reuse||Force using an existing non-empty volume group.|
|volume.lvm.stripes||string||lvm driver||-||storage_lvm_stripes||Number of stripes to use for new volumes (or thin pool volume).|
|volume.lvm.stripes.size||string||lvm driver||-||storage_lvm_stripes||Size of stripes to use (at least 4096 bytes and multiple of 512bytes).|
|rsync.bwlimit||string||-||0 (no limit)||storage_rsync_bwlimit||Specifies the upper limit to be placed on the socket I/O whenever rsync has to be used to transfer storage entities.|
|volatile.initial_source||string||-||-||storage_volatile_initial_source||Records the actual source passed during creating (e.g. /dev/sdb).|
|volatile.pool.pristine||string||-||true||storage_driver_ceph||Whether the pool has been empty on creation time.|
|volume.block.filesystem||string||block based driver (lvm)||ext4||storage||Filesystem to use for new volumes|
|volume.block.mount_options||string||block based driver (lvm)||discard||storage||Mount options for block devices|
|volume.size||string||appropriate driver||unlimited (10GB for block)||storage||Default volume size|
|volume.zfs.remove_snapshots||bool||zfs driver||false||storage||Remove snapshots as needed|
|volume.zfs.use_refquota||bool||zfs driver||false||storage||Use refquota instead of quota for space.|
|zfs.clone_copy||bool||zfs driver||true||storage_zfs_clone_copy||Whether to use ZFS lightweight clones rather than full dataset copies.|
|zfs.pool_name||string||zfs driver||name of the pool||storage||Name of the zpool|
Storage pool configuration keys can be set using the lxc tool with:
lxc storage set [<remote>:]<pool> <key> <value>
|size||string||appropriate driver||same as volume.size||storage||Size of the storage volume|
|block.filesystem||string||block based driver||same as volume.block.filesystem||storage||Filesystem of the storage volume|
|block.mount_options||string||block based driver||same as volume.block.mount_options||storage||Mount options for block devices|
|security.shifted||bool||custom volume||false||storage_shifted||Enable id shifting overlay (allows attach by multiple isolated instances)|
|security.unmapped||bool||custom volume||false||storage_unmapped||Disable id mapping for the volume|
|lvm.stripes||string||lvm driver||-||storage_lvm_stripes||Number of stripes to use for new volumes (or thin pool volume).|
|lvm.stripes.size||string||lvm driver||-||storage_lvm_stripes||Size of stripes to use (at least 4096 bytes and multiple of 512bytes).|
|snapshots.expiry||string||custom volume||-||custom_volume_snapshot_expiry||Controls when snapshots are to be deleted (expects expression like
|snapshots.schedule||string||custom volume||-||volume_snapshot_scheduling||Cron expression (
|snapshots.pattern||string||custom volume||snap%d||volume_snapshot_scheduling||Pongo2 template string which represents the snapshot name (used for scheduled snapshots and unnamed snapshots)|
|zfs.remove_snapshots||string||zfs driver||same as volume.zfs.remove_snapshots||storage||Remove snapshots as needed|
|zfs.use_refquota||string||zfs driver||same as volume.zfs.zfs_requota||storage||Use refquota instead of quota for space|
Storage volume configuration keys can be set using the lxc tool with:
lxc storage volume set [<remote>:]<pool> <volume> <key> <value>
LXD supports using ZFS, btrfs, LVM or just plain directories for storage of images, instances and custom volumes.
Where possible, LXD tries to use the advanced features of each system to optimize operations.
|Optimized image storage||no||yes||yes||yes||yes|
|Optimized instance creation||no||yes||yes||yes||yes|
|Optimized snapshot creation||no||yes||yes||yes||yes|
|Optimized image transfer||no||yes||no||yes||yes|
|Optimized instance transfer||no||yes||no||yes||yes|
|Copy on write||no||yes||yes||yes||yes|
|Storage driver usable inside a container||yes||yes||no||no||no|
|Restore from older snapshots (not latest)||yes||yes||yes||no||yes|
The two best options for use with LXD are ZFS and btrfs.
They have about similar functionalities but ZFS is more reliable if available on your particular platform.
Whenever possible, you should dedicate a full disk or partition to your LXD storage pool.
While LXD will let you create loop based storage, this isn't recommended for production use.
Similarly, the directory backend is to be considered as a last resort option.
It does support all main LXD features, but is terribly slow and inefficient as it can't perform
instant copies or snapshots and so needs to copy the entirety of the instance's storage every time.
Currently, the Linux Kernel may not apply mount options and silently ignore
them when a block-based filesystem (e.g.
ext4) is already mounted with
different options. This means when dedicated disk devices are shared between
different storage pools with different mount options set, the second mount may
not have the expected mount options. This becomes security relevant, when e.g.
one storage pool is supposed to provide
acl support and the second one is
supposed to not provide
acl support. For this reason it is currently
recommended to either have dedicated disk devices per storage pool or ensure
that all storage pools that share the same dedicated disk device use the same
All backends but the directory backend have some kind of optimized image storage format.
This is used by LXD to make instance creation near instantaneous by simply cloning a pre-made
image volume rather than unpack the image tarball from scratch.
As it would be wasteful to prepare such a volume on a storage pool that may never be used with that image,
the volume is generated on demand, causing the first instance to take longer to create than subsequent ones.
ZFS, btrfs and CEPH RBD have an internal send/receive mechanisms which allow for optimized volume transfer.
LXD uses those features to transfer instances and snapshots between servers.
When such capabilities aren't available, either because the storage driver doesn't support it
or because the storage backend of the source and target servers differ,
LXD will fallback to using rsync to transfer the individual files instead.
When rsync has to be used LXD allows to specify an upper limit on the amount of
socket I/O by setting the
rsync.bwlimit storage pool property to a non-zero
There is no concept of a default storage pool in LXD.
Instead, the pool to use for the instance's root is treated as just another "disk" device in LXD.
The device entry looks like:
root: type: disk path: / pool: default
And it can be directly set on an instance ("-s" option to "lxc launch" and "lxc init")
or it can be set through LXD profiles.
That latter option is what the default LXD setup (through "lxd init") will do for you.
The same can be done manually against any profile using (for the "default" profile):
lxc profile device add default root disk path=/ pool=default
I/O limits in IOp/s or MB/s can be set on storage devices when attached to an instance (see Instances).
Those are applied through the Linux
blkio cgroup controller which makes it possible
to restrict I/O at the disk level (but nothing finer grained than that).
Because those apply to a whole physical disk rather than a partition or path, the following restrictions apply:
- Limits will not apply to filesystems that are backed by virtual devices (e.g. device mapper).
- If a filesystem is backed by multiple block devices, each device will get the same limit.
- If the instance is passed two disk devices that are each backed by the same disk,
the limits of the two devices will be averaged.
It's also worth noting that all I/O limits only apply to actual block device access,
so you will need to consider the filesystem's own overhead when setting limits.
This also means that access to cached data will not be affected by the limit.
- While this backend is fully functional, it's also much slower than all the others due to it having to unpack images or do instant copies of instances, snapshots and images.
- Quotas are supported with the directory backend when running on either ext4 or XFS with project quotas enabled at the filesystem level.
- Create a new directory pool called "pool1".
lxc storage create pool1 dir
- Use an existing directory for "pool2".
lxc storage create pool2 dir source=/data/lxd
- Uses RBD images for images, then snapshots and clones to create instances and snapshots.
- Due to the way copy-on-write works in RBD, parent filesystems can't be removed until all children are gone. As a result, LXD will automatically prefix any removed but still referenced object with "zombie_" and keep it until such time the references are gone and it can safely be removed.
- Note that LXD will assume it has full control over the osd storage pool. It is recommended to not maintain any non-LXD owned filesystem entities in a LXD OSD storage pool since LXD might delete them.
- Note that sharing the same osd storage pool between multiple LXD instances is
not supported. LXD only allows sharing of an OSD storage pool between
multiple LXD instances only for backup purposes of existing instances via
lxd import. In line with this, LXD requires the "ceph.osd.force_reuse" property to be set to true. If not set, LXD will refuse to reuse an osd storage pool it detected as being in use by another LXD instance.
- When setting up a ceph cluster that LXD is going to use we recommend using
xfsas the underlying filesystem for the storage entities that are used to hold OSD storage pools. Using
ext4as the underlying filesystem for the storage entities is not recommended by Ceph upstream. You may see unexpected and erratic failures which are unrelated to LXD itself.
- Create a osd storage pool named "pool1" in the CEPH cluster "ceph".
lxc storage create pool1 ceph
- Create a osd storage pool named "pool1" in the CEPH cluster "my-cluster".
lxc storage create pool1 ceph ceph.cluster\_name=my-cluster
- Create a osd storage pool named "pool1" with the on-disk name "my-osd".
lxc storage create pool1 ceph ceph.osd.pool\_name=my-osd
- Use the existing osd storage pool "my-already-existing-osd".
lxc storage create pool1 ceph source=my-already-existing-osd
- Can only be used for custom storage volumes
- Supports snapshots if enabled on the server side
- Uses a subvolume per instance, image and snapshot, creating btrfs snapshots when creating a new object.
- btrfs can be used as a storage backend inside a container (nesting), so long as the parent container is itself on btrfs. (But see notes about btrfs quota via qgroups.)
- btrfs supports storage quotas via qgroups. While btrfs qgroups are hierarchical, new subvolumes will not automatically be added to the qgroups of their parent subvolumes. This means that users can trivially escape any quotas that are set. If adherence to strict quotas is a necessity users should be mindful of this and maybe consider using a zfs storage pool with refquotas.
- Create loop-backed pool named "pool1".
lxc storage create pool1 btrfs
- Create a new pool called "pool1" using an existing btrfs filesystem at
lxc storage create pool1 btrfs source=/some/path
- Create a new pool called "pool1" on
lxc storage create pool1 btrfs source=/dev/sdX
LXD doesn't let you directly grow a loop backed btrfs pool, but you can do so with:
sudo truncate -s +5G /var/lib/lxd/disks/<POOL>.img sudo losetup -c <LOOPDEV> sudo btrfs filesystem resize max /var/lib/lxd/storage-pools/<POOL>/
(NOTE: For users of the snap, use
/var/snap/lxd/common/lxd/ instead of /var/lib/lxd/)
- Uses LVs for images, then LV snapshots for instances and instance snapshots.
- The filesystem used for the LVs is ext4 (can be configured to use xfs instead).
- By default, all LVM storage pools use an LVM thinpool in which logical volumes for all LXD storage entities (images, instances, etc.) are created. This behavior can be changed by setting "lvm.use_thinpool" to "false". In this case, LXD will use normal logical volumes for all non-instance snapshot storage entities (images, instances, etc.). This means most storage operations will need to fallback to rsyncing since non-thinpool logical volumes do not support snapshots of snapshots. Note that this entails serious performance impacts for the LVM driver causing it to be close to the fallback DIR driver both in speed and storage usage. This option should only be chosen if the use-case renders it necessary.
- For environments with high instance turn over (e.g continuous integration)
it may be important to tweak the archival
/etc/lvm/lvm.confto avoid slowdowns when interacting with LXD.
- Create a loop-backed pool named "pool1". The LVM Volume Group will also be called "pool1".
lxc storage create pool1 lvm
- Use the existing LVM Volume Group called "my-pool"
lxc storage create pool1 lvm source=my-pool
- Use the existing LVM Thinpool called "my-pool" in Volume Group "my-vg".
lxc storage create pool1 lvm source=my-vg lvm.thinpool_name=my-pool
- Create a new pool named "pool1" on
/dev/sdX. The LVM Volume Group will also be called "pool1".
lxc storage create pool1 lvm source=/dev/sdX
- Create a new pool called "pool1" using
/dev/sdXwith the LVM Volume Group called "my-pool".
lxc storage create pool1 lvm source=/dev/sdX lvm.vg_name=my-pool
- When LXD creates a ZFS pool, compression is enabled by default.
- Uses ZFS filesystems for images, then snapshots and clones to create instances and snapshots.
- Due to the way copy-on-write works in ZFS, parent filesystems can't be removed until all children are gone. As a result, LXD will automatically rename any removed but still referenced object to a random deleted/ path and keep it until such time the references are gone and it can safely be removed.
- ZFS as it is today doesn't support delegating part of a pool to a container user. Upstream is actively working on this.
- ZFS doesn't support restoring from snapshots other than the latest one. You can however create new instances from older snapshots which makes it possible to confirm the snapshots is indeed what you want to restore before you remove the newer snapshots.
Also note that instance copies use ZFS snapshots, so you also cannot restore an instance to a snapshot taken before the last copy without having to also delete instance copies.
Copying the wanted snapshot into a new instance and then deleting the old instance does however work, at the cost of losing any other snapshot the instance may have had.
- Note that LXD will assume it has full control over the ZFS pool or dataset. It is recommended to not maintain any non-LXD owned filesystem entities in a LXD zfs pool or dataset since LXD might delete them.
- When quotas are used on a ZFS dataset LXD will set the ZFS "quota" property. In order to have LXD set the ZFS "refquota" property, either set "zfs.use_refquota" to "true" for the given dataset or set "volume.zfs.use_refquota" to true on the storage pool. The former option will make LXD use refquota only for the given storage volume the latter will make LXD use refquota for all storage volumes in the storage pool.
- I/O quotas (IOps/MBs) are unlikely to affect ZFS filesystems very much. That's because of ZFS being a port of a Solaris module (using SPL) and not a native Linux filesystem using the Linux VFS API which is where I/O limits are applied.
- Create a loop-backed pool named "pool1". The ZFS Zpool will also be called "pool1".
lxc storage create pool1 zfs
- Create a loop-backed pool named "pool1" with the ZFS Zpool called "my-tank".
lxc storage create pool1 zfs zfs.pool\_name=my-tank
- Use the existing ZFS Zpool "my-tank".
lxc storage create pool1 zfs source=my-tank
- Use the existing ZFS dataset "my-tank/slice".
lxc storage create pool1 zfs source=my-tank/slice
- Create a new pool called "pool1" on
/dev/sdX. The ZFS Zpool will also be called "pool1".
lxc storage create pool1 zfs source=/dev/sdX
- Create a new pool on
/dev/sdXwith the ZFS Zpool called "my-tank".
lxc storage create pool1 zfs source=/dev/sdX zfs.pool_name=my-tank
LXD doesn't let you directly grow a loop backed ZFS pool, but you can do so with:
sudo truncate -s +5G /var/lib/lxd/disks/<POOL>.img sudo zpool set autoexpand=on lxd sudo zpool online -e lxd /var/lib/lxd/disks/<POOL>.img sudo zpool set autoexpand=off lxd
(NOTE: For users of the snap, use
/var/snap/lxd/common/lxd/ instead of /var/lib/lxd/)