Prepare a Disk for LVM with pvcreate Without Losing Data

pvcreate writes an LVM label onto a device and wipes its first four sectors, so the wrong path erases a boot partition, not the spare disk you meant. This guide walks you through initialising a whole device or partition as an LVM physical volume (PV), checking the resulting data-area alignment, and leaving it ready for vgcreate or vgextend. The examples match LVM 2.03.16(2), from the installed lvm2 package version 2.03.16-3ubuntu3.2.

Allow about fifteen minutes, plus whatever time you need to identify the correct disk. You need a shell, the lvm2 package and root privileges for the operation. This guide assumes the device is disposable or has been backed up.

1. Confirm the installed command

This is a read-only checkpoint and needs no elevated privileges:

$ pvcreate --version
  LVM version:     2.03.16(2)
  Library version: 1.02.185

Exact build configuration lines may vary; the version is the relevant compatibility detail here. Read the local help too if this machine has a different LVM release:

$ pvcreate --help | sed -n '1,35p'
  pvcreate - Initialize physical volume(s) for use by LVM

2. Identify the exact target

List devices and their current signatures before choosing a path. This read-only command is normally unprivileged, though some columns may need access to device metadata:

$ lsblk -o NAME,PATH,SIZE,TYPE,FSTYPE,MOUNTPOINTS
NAME        PATH           SIZE TYPE FSTYPE MOUNTPOINTS
nvme0n1     /dev/nvme0n1  931.5G disk
|-nvme0n1p1 /dev/nvme0n1p1  512M part vfat   /boot/efi
`-nvme0n1p2 /dev/nvme0n1p2   931G part ext4   /
sdb         /dev/sdb       1.8T disk

• Match size, type and physical device to your storage plan before replacing /dev/sdb below with your own target.
• A whole disk and a partition are different targets. /dev/sdb and /dev/sdb1 are not interchangeable.
• Never infer the target from a changing letter in a script, and never use a mounted filesystem's device unless you have deliberately planned the outage.

3. Preview the operation

Use --test with the real target path before writing metadata. LVM disables metadata writing in this mode but still runs its checks, and returns success in cases where the real command would proceed:

$ sudo pvcreate --test /dev/sdb
  TEST MODE: Metadata will NOT be updated and volumes will not be changed.
  Physical volume "/dev/sdb" successfully created.

Output depends on the device and configuration. What matters is that the path is the intended device, the command does not report an existing volume group or incompatible signature, and it returns status 0. Capture that immediately:

$ printf 'preview status: %s\n' "$?"
preview status: 0

A successful dry run is not a backup and does not reserve the device; it also does not prove another process will not open it before the real command runs. Stop if the device holds data you have not explicitly accounted for.

4. Initialise the physical volume

$ sudo pvcreate /dev/sdb
  Physical volume "/dev/sdb" successfully created.

The default metadata type is the current lvm2 format. Without --uuid, LVM generates a random PV UUID. Do not add --force or --yes to silence a warning until you understand it: one --force overrides some checks and confirmations, while repeating it as --force --force overrides stronger protections, including checks that a PV is already in a volume group.

If the command stops with a prompt or warning, treat that as a safety signal. Do not answer yes out of habit; resolve the existing filesystem, PV or VG state first.

5. Verify the label and alignment

Ask LVM to report the new PV. Reading the device needs access to its metadata but does not change it:

$ sudo pvs -o pv_name,pv_size,pv_uuid,pe_start,mda_size,pv_attr /dev/sdb
  PV         PSize   PV UUID                                1st PE  MdaSize  Attr
  /dev/sdb   <1.82t  3w...                                  1.00m   1020.00k a--

The UUID and rounded sizes will differ. With the installed command's default settings, the first physical extent is usually aligned at 1 MiB. pe_start and mda_size are the facts worth checking, not any particular UUID.

Do not assume every disk has the same alignment: LVM may adjust the data-area start from device or MD properties reported by sysfs. If an application or array needs a specific alignment, set it deliberately with --dataalignment, then inspect pe_start afterwards.

6. Choose metadata space before creating a large VG

The metadata area holds text copies of volume-group metadata. Its size cannot grow after pvcreate, and two metadata copies must fit when the PV is used that way. For a VG expected to have unusually large or frequently changing metadata, choose a larger area during initialisation:

$ sudo pvcreate --metadatasize 8m /dev/sdb
$ sudo pvs -o pv_name,pe_start,mda_size /dev/sdb

Run that example instead of the plain initialisation in step 4, and only on a device not already initialised for the intended VG. The requested size is approximate: alignment can round it up, moving pe_start. Verify the result rather than treating 8m as the final reported size.

7. Continue or recover

Once the PV is verified, create a new volume group with vgcreate VG_NAME /dev/sdb, or add it to an existing group with vgextend VG_NAME /dev/sdb. Those commands change LVM configuration and deserve their own separate planning. Keep the PV's UUID and the output of pvs in your change record.