The RAID1+LVM HowTo shows how to create an array as follows

mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sda1 missing
mdadm --manage --add /dev/md0 /dev/sdc1

in stead of (what I would expect)

mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sda1 /dev/sdc1

Would someone explain the difference?

Regards,
DJ

two steps instead of one?

I don't immediately understand why they would want to create a two-disk raid1 with a missing member and then add the 2nd member (sdc1) after, when you can do it all in a single step. Either way it will synch...

I think you can create a raid array with a missing unit, copy the data to the disk and see everything works well, then add the second disk to the array if there weren't any problems...

If you are configuring RAID on a pre-existing machine, you want to ensure that you don't loose your data, therefore it makes sense to create the raid set with one disk, copy the data and then add the second disk to the raid set.

But I would have thought that the initial install would be using /dev/sda rather than /dev/sdc

I have never had much luck with software raid personally and always prefer to use hardware raid where possible

Ahhh, could it be that this way you determine the direction of the sync? Always from existing --> added? You wouldn't need to manually copy and you wouldn't run the risk of the array "syncing" an empty disk over your precious data.

That's my hypothesis for now.

Regards,
DJ

No, you cannot create an array with a mounted volume (to prevent accidental data loss)

Therefore if your machine has two drives, and you want to have a RAID 1 setup, you first create the raid on the unmounted (presumably empty) disk, then copy the data to the RAID disk, then change the fstab entries to use the raid set remount the / partition and add the original disk to the raid set.

I would always suggest trying this out on a testlan before doing this in a production environment (and always have a backup before you start)

1 members found this post helpful.