The controller installed fine. Built a new 2.6.16 (2.6.16.53 now, the previous was 2.6.16.29). Booted... works! Good.

Shutdown, connect new disk #1, bootup. Good.

The disk still lies beside the server... since yesterday. Because...

Okay, there's this old lvm1 volume group (VG). Which I used since way back, it was created on a 2.4.x kernel. And still worked in 2.6.

Now, I created a new physical volume (PV) on a partition of the new disk. Tried to extend the VG. Failed, because the VG is lvm1 (metadata format) and the new PV is lvm2 format.

Okay, remake the PV in lvm1 format. Add to VG. Works.

Now... pvmove the physical extents (the actual data) from one of the old PVs to the new one. Oh, yes, now I remember this won't work for a lvm1-format VG. Great.

No problem, I think, vgconvert to new lvm2 format. Success. Now do pvmove... oops: "Metadata Too Large For Circular Buffer"

It appears I have too many LVs. 117 in fact.

Now, does it not seem strange that the "limited to 255 LVs" lvm1 metadata format can handle these fine, while the "virtually unlimited LVs" lvm2 metadata format fails?

Found some (limited) help on IRC. I could convert back to lvm1 format, but that's about it.

And here I am, stuck between a rock and a hard place. Do I keep lvm1, but no pvmove (and none of the mythical (to me) advantages of lvm2 which I had no need for in the past). Or move to lvm2 (and a whole new VG), but need some guesswork as to how big a metadata area I need.