Skip to content

Latest commit

 

History

History
52 lines (37 loc) · 4.85 KB

lvm.md

File metadata and controls

52 lines (37 loc) · 4.85 KB
Raw

Logical Volume Management — allows for representing multiple disks as a single one, an online modifications to partitions, etc. Basically, it's just an added easy-of-use and flexibility.

Layers used in order from the most primitive:

  1. physical volumes, prefix pv... — block devices that LVM is making use of. May be a block device as well as a partition, and loopback device, etc. LVM stores a header into it.
  2. volume groups, prefix gv... — storage pools of physical volumes that abstract characteristics of underlying devices.
  3. logical volumes, prefix lv... or lvm... — slices of a volume group, it's like partitions. Creation example: lvcreate -L 100M -T vol_group/logical_volume if thin, lvcreate -L100M -n logical_volume vol_group otherwise.

Displaying can be done with pvdisplay, vgdisplay, lvdisplay and pvs, vgs, lvs. The shorter ones usually print info in a more convenient way. The path /dev/foo/bar that it displays may not exist if LV status is "not available". For it to appear the logical volume needs to be activated with lvchange -ay /dev/foo/bar

Snapshots

Can't do snapshots of snapshots. Cloning a snapshot is also not currently possible. On IRC people recommend using a dd or mounting the partition and copying the content with rsync.

Snapshotting is confusing: in LVM you have usual pools, you have thin pool with usual volumes, and then also thin pools with thin volumes… And the snapshotting commands are slightly different for all of them. From what I gather, a "usual snapshot" (a lvcreate --size 10G -s --name my_disksnap vol_group/my_disk) has to be used for the latter, and a "thin snapshot" (command lvcreate -s vol_group/my_disk --type thin) for the former.

Examples:

  • Creation: 
    pvcreate /dev/sd{g,h,i,j,k,m,o,p,r,v}1           # mark disks
vgcreate vol_group /dev/sd{g,h,i,j,k,m,o,p,r,v}1 # create a volume group
lvcreate -L500G -n disk1 vol_group               # create a disk. It will be at /dev/vol_group/disk1
  • adding a marked disk to a volume group: vgextend my_vol_group /dev/sda1
  • importing a vol_group: either a vgchange -ay vol_group or vgimport vol_group

RAIDs

  • Creation: lvcreate --type raidα -L1G -n my_lv my_vol_group where raidα is a RAID type.
  • RAID status: lvs -o name,lv_health_status. Info for various statuses and what to do on those cases may be found in man lvmraid.
  • Show disks currently used by RAID (in particular): pvdisplay -m.

Compression and dedup (VDO)

Requires kernel ≥ 6.9.

Creation: pvcreate <devs>, vgcreate vg <devs>, lvcreate --type vdo -L 5G -V 3G --compression y --deduplication n vg/lv_vdo. Under the default settings the block device should then appear at /dev/vg/lvol0.

Current usage may be monitored with command vdostats --human-readable. If by any chance the Use% column would get to 100%, writing to the device would result in ENOSPC. It should be noted though that depending on the pool and volume sizes, the Used column may get increased once Use% is ≈85-87%.

Misc

  • exporting/importing: vgexport vg and vgimport vg.
    • activateing has to be done after importing a VG: vgchange -ay vg, and deactivating before exporting: vgchange -an vg.
  • adding disks: done separately for group and volume — e.g. vgextend vol_group /path/to/dev and lvextend -l +100%FREE /dev/vol_group/vol accordingly.
  • pvdisplay -m: check what devices VGs consist of
  • LVM can be made to ignore a device in operations by adding a filter to /etc/lvm/lvm.conf. Syntax of the filter is usually mentioned there, just search for filter word in comments in the file.
  • LVM propagates TRIM/Discard calls on an LV to the underlying physical devices. Worth noting, it is unrelated to issue_discards lvm.conf option, which is rather about LVM TRIMing devices after an LV was removed/reduced. In fact, devs doesn't recommend to enable issue_discards since it makes one unable to restore data from a removed LV. But at least Debian and Ubuntu enable it anyway.
  • if a device with a VG disappeared and re-appeared (e.g. an usb-stick was taken off/on), the VG health status gonna be error. I don't know if there's less intrusive way to fix the status, but one that works for me is deactivating, then activating the VG.
  • printing specific properties and filtering: use pvs and lvs with -o and -S args. Example of printing devices that belong to my_vg volume group: pvs --rows -S "vg_name=my_vg" -o name.
  • Device /dev/foo not found. upon running pvcreate may mean that the device path is filtered out by /etc/lvm/lvm.conf, section devices { … scan = [permitted_device_paths] …}
  • Display raid-types: lvs -ao +layout. The layout field should enlist the raid type, but how exactly it works is quite complicated, as it may belong to one of some internal volumes, hence the need in -a parameter.