Skip to content

Latest commit

 

History

History
127 lines (89 loc) · 3.67 KB

lvm.md

File metadata and controls

127 lines (89 loc) · 3.67 KB
Raw

LVM

Concepts

  • PVs (Physical Volumes) are full disk devices, or just partitions, or loopback files.
  • VGs (Volume Groups) are virtual groups of PVs.
  • LVs (Logical Volumes) reside in VGs. They are block devices, and are FS formatted.

PEs (Physical Extents) are the smallest contiguous extents in PVs assigned to LVs.

Introduction 101 Demo

Unless you have spare physical disks, let's create 2 loopback devices, to illustrate:

# truncate --size=1G /var/loop0.img
# truncate --size=1G /var/loop1.img
# file /var/loop0.img /var/loop1.img

# modprobe loop
# mknod -m640 /dev/loop0 b 7 0
# mknod -m640 /dev/loop0 b 7 1
# losetup /dev/loop0 --nooverlap --show /var/loop0.img
# losetup /dev/loop1 --nooverlap --show /var/loop1.img
# losetup --all --list

Now let's create x2 PVs:

# lsblk
# lvmdiskscan
# pvcreate -v /dev/loop0 /dev/loop1
# file /var/loop0.img /var/loop1.img
# pvs

Now let's create a VG, and initially use only 1 of our PVs:

# vgcreate -v vg0 /dev/loop0
# vgdisplay -v vg0
# vgs

Now let's create a LV:

# lvcreate --size 100M --name lv1 vg0
# lvs

This LV is now available as the /dev/vg0/lv1 block device, and we can use it like any other, to format it, mount it, and then writing 2 files which fit on it:

# mkfs.ext4 -v -L lv1 -F /dev/vg0/lv1
# mkdir /mnt/lv1
# mount /dev/vg0/lv1 /mnt/lv1
# echo "hello, world" >/mnt/lv1/hello.txt
# dd if=/dev/urandom of=/mnt/lv1/big.data bs=1M count=80
# df -h /mnt/lv1

If we now write another big file to it, that will obviously fail with No space left on device. But what we can do with LVM is resize the LV, given that we still have space on our (first) PV, and then write a much bigger file to it:

# lvresize --size 1000M --resizefs vg0/lv1
# dd if=/dev/urandom of=/mnt/lv1/bigger.data bs=1M count=800
# df -h /mnt/lv1

If we need even more space, we can extend the LV using our 2nd (virtual, here) drive:

# vgextend vg0 /dev/loop1
# lvresize --size 1500M --resizefs vg0/lv1
# df -h /mnt/lv1
# dd if=/dev/urandom of=/mnt/lv1/more.data bs=1M count=400

What's fun is that if we only need space that fits on the 2nd new PV, we could get rid of the 1st original one - without loosing any data:

# rm /mnt/lv1/more.data /mnt/lv1/bigger.data
# lvresize --size 500M --resizefs vg0/lv1
# pvmove /dev/loop0
# vgreduce vg0 /dev/loop0

To clean-up completely, let's entirely remove PVs and loopback mounts and image files:

# umount /mnt/lv1
# lvremove vg0/lv1
# vgchange -a n vg0
# vgremove vg0
# pvremove /dev/loop0
# pvremove /dev/loop1

# losetup --detach /dev/loop0
# losetup --detach /dev/loop1
# rm /dev/loop0
# rm /dev/loop1
# rm /var/loop0.img
# rm /var/loop1.img

Configuration

/etc/lvm/lvm.conf

systemctl status lvm2-monitor.service

Caching

TODO https://en.wikipedia.org/wiki/Dm-cache and https://wiki.archlinux.org/title/LVM#Cache and https://man7.org/linux/man-pages/man7/lvmcache.7.html

TODO How would one monitor this to see how effective is?

Striping

TODO https://www.youtube.com/watch?v=_dsHWq_MuUo

Alternatives

Both BTRFS and ZFS offer similar logical volume management, and Wikipedia lists other options.

Background

TODO

  1. Test filesystem size overheads
  2. lvmthin Thin Pools, lvcreate --thin --thinpool --virtualsize. Try --errorwhenfull on lvcreate
  3. lvmvdo for compression
  4. Caching
  5. Striping
  6. pvscan as on https://www.linuxsysadmins.com/moving-a-volume-group-from-server-to-server/
  7. vgchange
  8. lvchange