"Are we there yet?" (or testing d-i with qemu)
I'm on a bit of a mission to get the RAID support in d-i to allow the creation of a degraded array. So this afternoon, I thought I'd refresh myself with how mdcfg currently did it's thing by installing a RAID1 system in qemu.
Last time I played with qemu and d-i, it was all good up until it came to the network, which ended up being more of a notwork. Today, I tried the approach I used about 18 months ago with User Mode Linux. Using bridge-utils, I rejigged the host to be using a bridge device, br0, and added eth0 and the tun0 device to the bridge. (I had to load the tun kernel module and because I was running qemu as myself, I also made /dev/net/tun mode 666 for good measure). This immediately gave network access to d-i running in qemu, and DHCP worked flawlessly.
I created two 1GB empty files by dd'ing /dev/zero, one for hda and one for hdb and kicked off an install. Something like 3 hours later, the install completed. It was sooo sloooow. I was reminded of a driver reviver TV commercial, where a family of koalas are driving somewhere, and the kid koala in the back is going "Are we there yet?" every five minutes, because I felt like I was constantly doing that, jumping between virtual consoles to check the installation's progress.
But it wasn't in vain. I spent time poking around the innards of mdcfg, and it should be relatively easy to hack it so that, for example, if you go to make a RAID1 device, and only select 1 partition, it'll create a degraded mirror. It seems to do something of this ilk already. If you specify that there are a number of hot spare devices, and then fail to select that many partitions, it assumes the remainder are missing.
All in all, I'm very impressed with how d-i has progressed in terms of block device flexibility. We've come from boot floppies, supporting only vanilla partitions, to partman, that until recently supported LVM and RAID, to the situation today, where it'll finally do LVM on RAID. That's the icing on the cake for me. If I can help get to the point where you can create a degraded array, so you can add disks post-install, I think we'll be at the pinnacle of partitioning flexibility.