Replacing disks and everyday maintenance

Disks age, fill up and get replaced. With plain partitions that means downtime and copying. With LVM you can move the data from an old disk to a new one while the volume is mounted and in use, then remove the old disk from the pool. This lesson covers that move, plus the everyday commands: renaming and removing volumes, using a volume for swap, and making volumes permanent in /etc/fstab.

Replace a physical volume without downtime with vgextend, pvmove, vgreduce and pvremove.
Rename and remove logical volumes, and remove a volume group safely.
Make LVM mounts and swap permanent in /etc/fstab.

Moving data off a disk

Replace a disk without downtime: Step through moving data off an old disk while it is in use.

[root@servera ~]# lvs -o lv_name,devices vgdata
  LV     Devices
  lvdata /dev/sdb1(0)
[root@servera ~]# pvcreate /dev/sdb2; vgextend vgdata /dev/sdb2
  Volume group "vgdata" successfully extended
[root@servera ~]# pvmove /dev/sdb1 /dev/sdb2
  /dev/sdb1: Moved: 30.50%
  /dev/sdb1: Moved: 100.00%
[root@servera ~]# lvs -o lv_name,devices vgdata
  LV     Devices
  lvdata /dev/sdb2(0)
[root@servera ~]# vgreduce vgdata /dev/sdb1
  Removed "/dev/sdb1" from volume group "vgdata"
[root@servera ~]# pvremove /dev/sdb1
  Labels on physical volume "/dev/sdb1" successfully wiped.

The Devices column shows where each extent lives: /dev/sdb1(0) means "starting at extent 0 of sdb1". pvmove needs enough free extents on the target; if there are not, it fails with Insufficient free space and nothing is changed. The volume stays usable during the move. The order matters: move first, then reduce, then remove. If you take a PV out of a group while it still holds data, vgreduce refuses (or, with --removemissing after the disk has really failed, the affected volumes are lost).

Everyday maintenance

Task	Command
Rename a volume	`lvrename vgdata lvdata lvstore`
Remove a volume (destroys its data)	`umount DIR; lvremove vgdata/lvstore` (asks to confirm; `-y` skips)
Remove an empty group	`vgremove vgdata`
Show where each LV lives	`lvs -o +devices`
Show the device path	`lvs -o +lv_path`

After renaming, fix every place that uses the old name, above all /etc/fstab. A line that uses UUID= is not affected, which is one more reason for UUIDs.

Making volumes permanent

The device path /dev/VG/LV is stable, so an fstab line can use it directly (or the UUID that blkid shows):

[root@servera ~]# cp /etc/fstab /root/fstab.bak
[root@servera ~]# echo "/dev/vgdata/lvstore  /srv/data  xfs  defaults  0 0" >> /etc/fstab
[root@servera ~]# systemctl daemon-reload
[root@servera ~]# findmnt --verify | tail -1
0 parse errors, 0 errors, 1 warning
[root@servera ~]# mount -a && df -h /srv/data | tail -1
/dev/mapper/vgdata-lvstore  736M   38M  699M   6% /srv/data

Never use the /dev/dm-N names: they can change between boots. Volumes activate automatically at boot; if one does not, see the next lesson.

Swap on a volume

Swap space works the same as on a partition, and a volume is easy to prepare:

[root@servera ~]# lvcreate -n lvswap -L 256M vgdata
  Logical volume "lvswap" created.
[root@servera ~]# mkswap /dev/vgdata/lvswap
Setting up swapspace version 1, size = 256 MiB (268431360 bytes)
[root@servera ~]# echo "/dev/vgdata/lvswap  none  swap  defaults  0 0" >> /etc/fstab
[root@servera ~]# systemctl daemon-reload; swapon -a; swapon --show
NAME      TYPE      SIZE USED PRIO
/dev/dm-1 partition 256M   0B   -2

The list shows the kernel name (/dev/dm-1), but the fstab line uses the stable path. A swap volume cannot be resized in place: turn it off, extend it, run mkswap again, and turn it on.

Try it

Check your understanding

Which command moves the data of one PV to another while mounted?
What must you do before pulling a disk out of a volume group?
Which fstab device name stays valid when you rename an LV?
lvremove asks to confirm. Why?
How do you use an LV as swap?
Which command removes a PV from a volume group?