I am wondering what the advantages and disadvantages of the following two scenarios are.
LVM physical volume directly on SAN volume
pvcreate /dev/sda
LVM physical volume on a single partition, which is spanning the whole SAN volume
parted /dev/sda -s -- mklabel gpt mkpart primary 0 -1
pvcreate /dev/sda1
I read in the LVM-Manual, that PV on the whole disk is not recommended because of management issues with other OSes that don't understand LVM-labels. But I am not sure, whether this really applies to SAN volumes in practice.
Furthermore I think partitioning adds another layer with possible problems, like device name changes of mapped devices on the partition device.
What is better practice?
Either method will work just fine and equivalently stable.
The partition-method is somewhat more intelligible by recovery utilities, though that is a minor concern. Anything based on a Linux (and probably *BSD) core will know how to handle a full-disk LVM setup, which includes pretty much every recovery tool I'd consider using on a broken Linux box. Tools designed for Windows recovery won't be able to handle full-disk LVM, though.
The SAN volume thing does require some consideration, but the big question to ask yourself is:
Unless you're creating a cluster, the answer to that is probably "no" except in dire recovery circumstances. If you have to rebuild the boot/system volumes for some reason, chances are near certain you'll use the same OS you started with.
Full-disk LVM is just fine.
So is One-Partition LVM.
Use whichever feels better to you.