LVM2 mirroring is still somewhat under development. The log volume is used to track which extents are up-to-date on each half of the mirror when one of the halves fails.
I guess that placing the log volume on a separate disk is a temporary solution that makes RAID1 debugging easier and allows to side-step the problems of adding the log to the existing volume(s) when it is mirrored: allocating some disk space from the volumes themselves would require shrinking the filesystem a little first, and on-line shrinking is not supported with most filesystems.
Another possibility would be to pre-allocate some space for RAID1 log from each LV as they're created, just in case they will be used for RAID1 at some point of their life. Of course this would require adding support for the pre-allocation feature and some way to migrate the existing LVs where this pre-allocation is not yet done...
To recover from the loss of log volume, a full resync of the mirror set is needed.
You can use the "--mirrorlog core" option with the lvconvert command to place the log in RAM. This is useful if the mirror is short-lived, for example when performing an on-line data migration from an old storage system to a new one. In this situation it is typical to remove the "old" half of the RAID1 set as soon as all data is successfully duplicated to the "new" half.
MK
MK
