libblkid ends up seeing an lvm2pv as a swap partition instead because swap is earlier in the table. lvm2pv should probably be the first thing probed instead of the last as it's a container. See attached patch
Created attachment 302779 [details] fix
Bleah, what a mess. Neither mkswap nor pvcreate does a great job of zeroing other signatures around the new signature. For some incantations, pvcreate-then-mkswap has an initial 4k on the disk which is *identical* to mkswap-then-pvcreate. :( Sent a patch upstream (and to Jeremy) to check that the swap v1 header has a valid version and page count... "pvcreate" by itself will zero those. (pvcreate -u <blah> will not, unfortunately, so there's still a small hole). But, it fixes Jeremy's case: # blkid -c /dev/null /root/md0-for-esandeen /root/md0-for-esandeen: UUID="kyOPnO-Biwa-PssU-PsXx-kJDW-36Ry-iZvqk5" TYPE="lvm2pv"
http://article.gmane.org/gmane.comp.file-systems.ext4/6038
Patch committed and built and I tested it with my reproducer, so going to close
Patch upstream. http://git.kernel.org/?p=fs/ext2/e2fsprogs.git;a=commitdiff;h=6360d129d003f25942f62fab2cdc5e1d97cc88a6