On Tue, 2006-08-15 at 15:21 -0400, Bill Nottingham wrote:
No, I mean you'll get disparate events:
- hey, I added hda1!
- Looks like it might be part of VG <foo>...
- hey, I added hdb1!
- Looks like it might be part of VG <foo>...
<kernel takes a nap, five seconds later>
- hey, I added hdc1!
- Looks like it might be part of VG <foo>...
When do you decide to assemble?
Each PV got enough metadata to tell what the entire LV looks like - e.g. actually each PV has metadata saying 1) I'm part of LV "foo"; and 2) LV foo consists of PV's "bar", "baz" and "bat".
So we just teach HAL to pick up this metadata (like we already do for 20 + file systems) and then export it as volume.linux.lvm.* properties and then g-v-m can do the right thing when it sees all the PV's. It's actually pretty simple.
Same thing for is true for RAID. However, for RAID, we might allow the user to start a RAID mirror set (probably from g-d-u) even if some backing volumes are missing but we probably shouldn't do this automatically. Except for maybe on boot but that's initramfs / initscripts business. That's what I thought you were asking :-)
David