DANGER!!! Problems with 10.04 installer (RAID devices *will* get corrupted)

[Dave Howorth]

Alvin Thompson wrote:
> On 04/22/2010 12:18 AM, Christopher Chan wrote:
>> You mentioned md0 aka a Linux software raid array and therefore do not
>> have any 'raid' devices setup in BIOS. The 'raid' bios will not,
>> therefore, attempt to do anything on your disk. Otherwise, it would have
>> trashed your md0 a long time ago. Please STOP adding comments to those
>> bug reports because you have just completely confused the issues.
>> #191119 (sorry Xander - missed the part where it says Alvin made the
>> comment) has NOTHING to do with what you are currently seeing.
> 
> I don't know how many different ways I can explain this, but here's 
> another try.  This is the relevant quote from the bug report:
> 
>   Select partition on alternate, non raid, drive to install Ubuntu on.

I hesitate to join this thread, since it can get a bit heated. But I
think that there are probably two separate bugs and I'd like to explain why.

#191119 sounds like a problem I have had with a different distro and
different hardware config and different partitioning. Basically what
happens is that grub overwrites blocks that it shouldn't. In my case it
is a true hardware RAID with LVM occupying the whole disk (i.e. not in a
partition and I now know why this is not recommended! ). The LVM
metadata occupies the start of the RAID virtual disk. grub got confused
about the boot order of the machine and wrote to that disk instead of
the separate boot disk. #191119 sounds like the same situation, with
grub writing to one of the disks in the fakeRAID (and I assume, but
don't know, that the fakeRAID has control information at the start of
the disk). So that bug would be down to the well-known difficulties grub
sometimes has in its relationship with the BIOS.

#568183 seems like a completely different situation, since it involves
partitions rather than raw disks and since Alvin says it depends on how
the installer detects partition types. It's really not clear to me
exactly what the installer wrote into that ghost partition though. It
seems very difficult to create a bug that would write to an existing
partition when instructed not to.

I hope this doesn't fan any flames. I'm just trying to say that I think
the two bugs should NOT be marked as duplicates until a developer
establishes that they really are. Otherwise one of the problems won't
get investigated.

Cheers, Dave