Chapter 8. Removing storage devices

You can safely remove a storage device from a running system, which helps prevent system memory overload and data loss.

Prerequisites

  • Before you remove a storage device, you must ensure that you have enough free system memory due to the increased system memory load during an I/O flush. Use the following commands to view the current memory load and free memory of the system: 

    # vmstat 1 100
# free

  • Red Hat does not recommend removing a storage device on a system where:

    • Free memory is less than 5% of the total memory in more than 10 samples per 100.
    • Swapping is active (non-zero si and so columns in the vmstat command output).

8.1. Safe removal of storage devices

Safely removing a storage device from a running system requires a top-to-bottom approach. Start from the top layer, which typically is an application or a file system, and work towards the bottom layer, which is the physical device.

You can use storage devices in multiple ways, and they can have different virtual configurations on top of physical devices. For example, you can group multiple instances of a device into a multipath device, make it part of a RAID, or you can make it part of an LVM group. Additionally, devices can be accessed via a file system, or they can be accessed directly such as a “raw” device.

While using the top-to-bottom approach, you must ensure that:

  • the device that you want to remove is not in use
  • all pending I/O to the device is flushed
  • the operating system is not referencing the storage device

8.2. Removing block devices and associated metadata

To safely remove a block device from a running system, to help prevent system memory overload and data loss you need to first remove metadata from them. Address each layer in the stack, starting with the file system, and proceed to the disk. These actions prevent putting your system into an inconsistent state.

Use specific commands that may vary depending on what type of devices you are removing:

  • lvremove, vgremove and pvremove are specific to LVM.
  • For software RAID, run mdadm to remove the array. For more information, see Managing RAID.
  • For block devices encrypted using LUKS, there are specific additional steps. The following procedure will not work for the block devices encrypted using LUKS. For more information, see Encrypting block devices using LUKS.
Warning

Rescanning the SCSI bus or performing any other action that changes the state of the operating system, without following the procedure documented here can cause delays due to I/O timeouts, devices to be removed unexpectedly, or data loss.

Prerequisites

  • You have an existing block device stack containing the file system, the logical volume, and the volume group.
  • You ensured that no other applications or services are using the device that you want to remove.
  • You backed up the data from the device that you want to remove.

  • Optional: If you want to remove a multipath device, and you are unable to access its path devices, disable queueing of the multipath device by running the following command: 

    # multipathd disablequeueing map multipath-device

    This enables the I/O of the device to fail, allowing the applications that are using the device to shut down.

Note

Removing devices with their metadata one layer at a time ensures no stale signatures remain on the disk.

Procedure

  1. Unmount the file system: 

    # umount /mnt/mount-point

  2. Remove the file system: 

    # wipefs -a /dev/vg0/myvol
    Note

    If you have added an entry into /etc/fstab file to make a persistent association between the file system and a mount point you should also edit /etc/fstab at this point to remove that entry.

    Continue with the following steps, depending on the type of the device you want to remove:

  3. Remove the logical volume (LV) that contained the file system: 

    # lvremove vg0/myvol

  4. If there are no other logical volumes remaining in the volume group (VG), you can safely remove the VG that contained the device: 

    # vgremove vg0

  5. Remove the physical volume (PV) metadata from the PV device(s): 

    # pvremove /dev/sdc1
    # wipefs -a /dev/sdc1

  6. Remove the partitions that contained the PVs: 

    # parted /dev/sdc rm 1
Note

Follow the next steps only if you want to fully wipe the device.

  1. Remove the partition table: 

    # wipefs -a /dev/sdc
Note

Follow the next steps only if you want to physically remove the device.

  • If you are removing a multipath device, execute the following commands:

    1. View all the paths to the device: 

      # multipath -l

      The output of this command is required in a later step.

      1. Flush the I/O and remove the multipath device: 

        # multipath -f multipath-device

  • If the device is not configured as a multipath device, or if the device is configured as a multipath device and you have previously passed I/O to the individual paths, flush any outstanding I/O to all device paths that are used: 

    # blockdev --flushbufs device

    This is important for devices accessed directly where the umount or vgreduce commands do not flush the I/O.

  • If you are removing a SCSI device, execute the following commands:

    1. Remove any reference to the path-based name of the device, such as /dev/sd, /dev/disk/by-path, or the major:minor number, in applications, scripts, or utilities on the system. This ensures that different devices added in the future are not mistaken for the current device.

    2. Remove each path to the device from the SCSI subsystem: 

      # echo 1 > /sys/block/device-name/device/delete

      Here the device-name is retrieved from the output of the multipath -l command, if the device was previously used as a multipath device.

  1. Remove the physical device from a running system. Note that the I/O to other devices does not stop when you remove this device.

Verification

  • Verify that the devices you intended to remove are not displaying on the output of lsblk command. The following is an example output: 

    # lsblk

NAME   MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sda      8:0    0    5G  0 disk
sr0     11:0    1 1024M  0 rom
vda    252:0    0   10G  0 disk
|-vda1 252:1    0    1M  0 part
|-vda2 252:2    0  100M  0 part /boot/efi
`-vda3 252:3    0  9.9G  0 part /

Additional resources

  • The multipath(8), pvremove(8), vgremove(8), lvremove(8), wipefs(8), parted(8), blockdev(8) and umount(8) man pages.