[aleksandar@thinkpad]

Adding Persistent Swap on an LVM Logical Volume

Extra swap carved out of the vg_data volume group from the LVM lab: a new logical volume, mkswap, swapon, a priority higher than the existing swap, an /etc/fstab entry tested with swapon -a (mount -a ignores swap), and growing it later - including why lvextend -r refuses swap and a mkswap that has to be rerun.

Goal: add 2 GiB of swap to a RHEL system as a new logical volume in vg_data - the volume group built in the LVM from scratch lab - keep the swap the installer created, make the new one survive a reboot, and later grow it without breaking anything.

Why on LVM and not a swap file or partition: a partition can’t be resized without repartitioning, and a swap file depends on the filesystem underneath it. A logical volume can be grown later the same way as any other LV, and its /dev/vg_data/lv_swap path never changes. It’s also how RHEL itself does it - a default install already puts its swap on LVM, as rhel/swap.

Tested on: a Linux VM with lvm2 and util-linux, the same tools RHEL 10 uses, on throwaway loop-device disks. Every command and output below comes from that run, except the final reboot - loop devices don’t survive one - so the Verification section shows what to expect rather than captured output. Device numbers like dm-2, and the sizes, will differ on another system.

1. Look at the starting point

free -h
swapon --show
vgs

NAME      TYPE       SIZE USED PRIO
/dev/dm-0 partition 1024M   0B   -1
  VG      #PV #LV #SN Attr    VSize  VFree
  rhel      1   1   0 wz--n-- <3.00g <2.00g
  vg_data   1   1   0 wz--n-- <3.00g     0

One swap device is already active - the installer’s rhel/swap, shown by its device-mapper name dm-0. vg_data has VFree of 0: the LVM lab ended with lv_apps taking every free extent. The new swap needs space from somewhere first.

2. Make room in the volume group

Add another disk to vg_data, exactly as in the LVM lab:

lsblk                        # find the new, empty disk - here /dev/sdd
sudo blkid /dev/sdd          # empty output = no existing signature
sudo pvcreate /dev/sdd
sudo vgextend vg_data /dev/sdd
vgs vg_data

  VG      #PV #LV #SN Attr    VSize VFree
  vg_data   2   1   0 wz--n-- 5.99g <3.00g

If VFree is already big enough on your system, skip this step.

3. Create the logical volume and format it as swap

sudo lvcreate -L 2G -n lv_swap vg_data
sudo mkswap /dev/vg_data/lv_swap

  Logical volume "lv_swap" created.
Setting up swapspace version 1, size = 2 GiB (2147479552 bytes)
no label, UUID=43308a47-5f98-4142-8eac-1b485b70ea27

mkswap writes a swap signature - the swap equivalent of mkfs. Without it, swapon refuses the device.

mkswap doesn’t ask before overwriting. Pointed at the wrong LV - say lv_apps instead of lv_swap - it prints warning: wiping old xfs signature, then does it anyway, and the data on that LV is gone. Check the path twice. The one case it does refuse is a swap device that’s currently active.

4. Turn it on and look

sudo swapon /dev/vg_data/lv_swap
swapon --show
free -h

NAME      TYPE       SIZE USED PRIO
/dev/dm-0 partition 1024M   0B   -1
/dev/dm-2 partition    2G   0B   -1
               total        used        free
Swap:          3.0Gi          0B       3.0Gi

Both swap devices are active, and the total went from 1 GiB to 3 GiB. This only lasts until the next reboot - nothing records it yet.

5. Make it persistent in /etc/fstab

echo "/dev/vg_data/lv_swap  none  swap  defaults,pri=10  0 0" | sudo tee -a /etc/fstab
  • none in the mount point column - swap isn’t mounted anywhere.
  • swap as the type is what tells the system to activate it with swapon at boot instead of mounting it.
  • pri=10 - see section 6.
  • The LV path instead of UUID=... is deliberate - section 7 shows why.

Now test the entry. The usual fstab test is the wrong one here:

sudo swapoff /dev/vg_data/lv_swap
sudo mount -a
swapon --show

NAME      TYPE       SIZE USED PRIO
/dev/dm-0 partition 1024M   0B   -1

mount -a skips swap entries completely. No error, no output, and the new swap is still off - the entry could contain a typo and mount -a would never show it. The command that reads the swap lines from fstab is swapon -a:

sudo swapon -a
swapon --show

NAME      TYPE       SIZE USED PRIO
/dev/dm-0 partition 1024M   0B   -1
/dev/dm-2 partition    2G   0B   10

Then let systemd pick up the new entry and check the file as a whole:

sudo systemctl daemon-reload
findmnt --verify                                 # Success, no errors or warnings detected
systemctl status dev-vg_data-lv_swap.swap        # Loaded: loaded (/etc/fstab; generated)

At boot, systemd turns each swap line in fstab into a .swap unit like this one. findmnt --verify without the daemon-reload first reports a warning that systemd is still using the old version of fstab.

6. Swap priority - which device gets used first

The PRIO column above changed from -1 to 10. The kernel fills swap devices in order of priority, highest first. Devices without an explicit priority get a negative one. Devices with the same priority are used in parallel, round-robin.

With pri=10, the new LV is used first, and the installer’s swap only after it fills up. That’s useful when the two sit on different disks and one is faster. Leave pri= out to keep both at the same level.

Note that pri=10 only applies when the swap is turned on through fstab - by swapon -a or at boot. A manual swapon /dev/vg_data/lv_swap ignores fstab and gives it the default priority - which is why every re-activation in section 7 goes through swapon -a.

7. Growing the swap later

Growing a filesystem LV is one command - lvextend -r. For swap, it isn’t:

sudo lvextend -r -L +512M /dev/vg_data/lv_swap

  File system extend is not supported (swap).

The command fails, and the LV isn’t extended at all - lvs still shows 2.00g. Without -r, the LV does grow, even while the swap is active:

sudo lvextend -L +512M /dev/vg_data/lv_swap
lvs vg_data/lv_swap          # 2.50g
swapon --show                # still 2G

The kernel keeps using the size written in the swap signature when mkswap ran. Even turning the swap off and on again doesn’t change that:

sudo swapoff /dev/vg_data/lv_swap
sudo swapon -a
swapon --show                # still 2G - the LV is bigger, the swap isn't

The extra space only becomes swap after a new mkswap, which has to run while the swap is off:

sudo swapoff /dev/vg_data/lv_swap
sudo mkswap /dev/vg_data/lv_swap
sudo swapon -a
swapon --show

mkswap: /dev/vg_data/lv_swap: warning: wiping old swap signature.
Setting up swapspace version 1, size = 2.5 GiB (2684350464 bytes)
NAME      TYPE       SIZE USED PRIO
/dev/dm-0 partition 1024M   0B   -1
/dev/dm-2 partition  2.5G   0B   10

Every mkswap generates a new UUID. In this run, the UUID changed from 43308a47-... to 0f830b70-.... An fstab line written as UUID=43308a47-... would now point at nothing:

swapon: cannot find the device for UUID=43308a47-5f98-4142-8eac-1b485b70ea27

That’s why section 5 uses /dev/vg_data/lv_swap: the LV path stays the same through any number of mkswap runs. If an existing fstab uses the UUID, either update it after mkswap, or keep the old one with mkswap -U <old-uuid> /dev/vg_data/lv_swap.

Verification - prove it survives a reboot

sudo reboot

After the system is back:

swapon --show                                  # both devices, lv_swap at 2.5G with PRIO 10
free -h                                        # Swap total 3.5Gi
systemctl status dev-vg_data-lv_swap.swap      # active, loaded from /etc/fstab
lsblk /dev/sdd                                 # disk -> vg_data-lv_swap -> [SWAP]

If lv_swap is missing after the reboot, check journalctl -b -u dev-vg_data-lv_swap.swap - the unit’s log says whether the device wasn’t found or the swap signature was rejected.

Safety notes

  • mkswap overwrites whatever is on the target without a prompt, other than a warning printed after the decision is already made. Double-check the LV path before running it, every time.
  • Test swap entries in fstab with swapon -a, not mount -a. mount -a ignores them, so a broken swap line passes that test silently.
  • swapoff moves everything in that swap device back into RAM. On a system under memory pressure, check free -h first - if what’s swapped out doesn’t fit in free RAM, swapoff can hang or wake the OOM killer.
  • Don’t run mkswap on a swap device that’s in use. It refuses with is mounted; will not make swapspace - turn it off with swapoff first.
  • After every mkswap, the UUID is new. Anything that refers to the swap by UUID - fstab, or a resume= option for hibernation on the kernel command line - has to be updated, or use the LV path instead.
  • More swap isn’t a fix for a system that’s constantly short on memory - it turns an out-of-memory crash into a very slow system. Watch the si/so columns in vmstat (see the disk check cheat sheet’s swap section) to tell occasional swap use from constant paging.