[libvirt] Storage volume issues when creating an LVM or RAID partition

[Henrik Persson E]

Hi everyone,

I have currently some patches coming up for storage pools and volumes
based on physical disks. However I have an issue that I would like to
hear your opinion about.
 
When it comes to storage pools and volumes based on physical disks, the
name given does not really have any real meaning, the underlying
disk/partition will at the end decide the name of the pool/volume (as
soon as a refresh on the pool is made, the user defined names will be
gone). This is quite confusion but at the end it still works. But there
are cases where it does not work. 
To be able to create a partition type such as LVM or raid, parted (which
is used as the disk management backend) does not support setting this
partition type when creating the partition but instead by setting a flag
on the partition after it has been created. As it is not possible to
decide at which position in the partition table the new partition should
be created, parted will just choose the next available number. To be
able to set this flag on the partition, we need to know which partition
we just created, and suddenly the name of the volume is important, but
only if it has followed the rules that parted applied when finding the
next available slot. If the name of the volume is incorrect we will end
up with a volume created with incorrect type since we can't do 2
operation towards parted in a transaction manner (e.g. Volume name is
sdb2 but the newly created partition is sdb1).

I see two possible solutions on this problem:

1. We add volume name constraints in this case that makes sure that when
a volume is created by a user (by checking the pool and scan all ready
existing volumes in that pool), it has to create a name that is correct
according to parted's rules. In that case we can use the volume name to
find the partition just created and we can set the flag on the partition
in a safe way

2. we extend the libvirt_parthelper and uses the libparted instead to
create a partition, which then can be made in a more transaction like
manner. However, in this case we have to more or less add all the
support that parted has for creating a partition which is quite
extensive.

I put my vote on solution 1 but I like to know your opinion and also if
any one knows that the initial plans were.

/Henrik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20090622/5f23d2eb/attachment-0001.htm>