LVM Online Disk Replacement (LVM OLR)

When replacing a disk requires halting applications and

restoring data from a backup

Replacing a disk requires halting applications and restoring data from a backup under the

following circumstances:

v If the disk being replaced contains any unmirrored logical volumes

v If any data in the mirrored logical volumes on the disk being replaced has been

compromised due to simultaneous disk failures

If any of the logical volumes on the disk being replaced are not mirrored, the disk being

replaced held the only copy of the data and by definition there are no mirror copies

elsewhere. So replacing the disk requires halting applications and filesystems using the

disk and restoring the data from backup.

If all the logical volumes on the disk being replaced are mirrored and none have been

compromised due to simultaneous disk failures, it is not necessary to halt applications using

the logical volumes and there is no need to restore any data to the logical volumes from a

backup. Once the disk is replaced and attached to the volume group again, LVM will

automatically resynchronize the data on the disk.

The data in a mirrored logical volume is compromised when no remaining available disk in

the volume group contains a non-stale copy of the data. Prior to replacing a disk it is

important to ensure that for each extent in each logical volume on the disk being replaced

there is a non-stale copy on a different available disk. The lvdisplay(1M) -v command can

be employed to display each of the logical volumes and the state of all the extents within

them. If there is any extent that does not have a non-stale copy on some other available

disk, halt any applications using the data and restore the data from backup after replacing

the disk.

For example, given a logical volume is mirrored across two disks: A and B. If disk A

contains some stale extents, and disk B fails, the logical volume is compromised. The

cause was a partial failure of disk A (perhaps a few I/O requests timed-out) coincident with

a complete failure of B, before A could be re-synced from B. Displaying the logical

volume shows that although disk A is available, it contains some stale data. Although disk

B may appear to have a non-stale copy of all the data, that disk is down so the copy there is

unavailable. Then if disk B is replaced, the data on the replacement disk cannot be

automatically synced with the copy from disk A because the stale data there is unavailable.

Under these circumstances applications using the logical volume must be halted prior to

replacing the disk, and after the disk is replaced, it must be restored from backup.

It is important to note that simultaneous disk failures are a rare event provided that the disks

and I/O hardware in the volume group are properly isolated and promptly replaced at the

first sign of failure. Essentially a mirrored volume that has only a single available copy of

any of its data has already weathered one or more failures and is susceptible to being

compromised by just one subsequent disk failure.