Linux – Convert Raid 1 to Raid 10 with no new disks

Ok, silly me. I figured it out.

After running mdadm --assemble --scan which finds and assembles the container, running mdadm -I /dev/md/imsm0 assembles the arrays inside the container. D'oh!

# mdadm --assemble --scan
mdadm: Container /dev/md/imsm0 has been assembled with 3 drives
# mdadm -I /dev/md/imsm0
mdadm: Started /dev/md/NMWVolume with 3 devices
# ls -l /dev/md*
brw-rw---- 1 root disk   9, 126 Jan 24 02:22 /dev/md126
brw-rw---- 1 root disk 259,   0 Jan 24 02:22 /dev/md126p1
brw-rw---- 1 root disk   9, 127 Jan 24 02:22 /dev/md127

/dev/md:
total 0
lrwxrwxrwx 1 root root 8 Jan 24 02:22 imsm0 -> ../md127
lrwxrwxrwx 1 root root 8 Jan 24 02:22 NMWVolume -> ../md126

/dev/md126p1 has my RAID5'd 1TB NTFS volume.

This is very simple - you have two 250GB disks that are presented to any OS as two disks, that you'd configured Windows to mirror them is nice but irrelevant as they were always two disks. ESXi is just seeing the same two disks and as it doesn't support software RAID is just showing you the two disks as they were presented to it.

Of course there's nothing to stop you creating two .vmdk's, one per datastore, attaching them to a VM and having that VM's OS mirroring them internally - at least then it will handle the failure of the disk that's not got ESXi running on it. Bit pointless and you should just get a RAID-capable and HCL-approved controller but there you go.