Anaconda issues with LVM when using an entire disk as a LVM physical partition

[Peter Jones]

On Sat, 2006-04-01 at 13:13 -0500, Adam Gibson wrote:
> I created a second disk /dev/hdb as an LVM partition manually in FC4. 
> There are no partitions like /dev/hdb0, etc.  /dev/hdb is the partition.

You shouldn't do that.  Instead, create a partition of type 0x8e, and
make the PV on that.

> When Anaconda first starts up it says that the disk is not initialized 
> and asks if you want to initialize it(which would have hosed my LVM on 
> that disk).
> 
> As long as you answer no to the initialization when you get to disk 
> druid, it does show the LVM partition but you can not do anything with it.
> 
> I am ok that it can not edit it because it does not expect the entire 
> disk to be a partition, but the scary part was when it said the disk was 
> uninitialized and asked to initialize it.  Someone might accidentally 
> say yes to that not realizing that it is already initialized in a sense 
> that the entire disk is an LVM partition.

The disk doesn't have a partition table, and detecting LVM metadata
correctly is not simple.

There's no good way for anaconda to handle this until somebody writes a
real library to deal with LVM metadata, rather than trying to call out
to to the executable for everything.  Even then, creating a PV on a
non-partitioned drive is the wrong approach.

-- 
  Peter