Testing 16 disk configurations takes how long?

[andrew bezella]

On Sun, 2013-03-17 at 19:34 +0100, Thomas Neumann wrote:
[...] 
> @Andrew: I'm wondering if your configs are reliable, Let me explain:
> 
>   disk_config /dev/disk/by-path/pci-0000:00:1f.2-scsi-0:0:0:0 disklabel:gpt-bios align-at:1M
>   [..]
>   disk_config /dev/disk/by-path/pci-0000:00:1f.2-scsi-1:0:0:0 sameas:sda
> 
> What happens if for some reason 'sda' is something else then [...]-0:0:0:0?

hello -

while not bullet-proof, they're reliable in our environment.  our
hardware config is 2 internal drives for a mirrored os (w/no archival
data), and then either 24 or 36 hot-swappable data drives.  the
disk/by-path enumeration for each server class should be the same[1],
and the paths for the internal and external drives are quite different
(iirc external disk/by-path for our hw actually stopped working sometime
between jaunty and lucid).  anyway, if both internal drives don't show
up where i expect them by-path, fai will error out (which is the
behavior i want).

> Shouldn't the second line read
>   disk_config /dev/disk/by-path/pci-0000:00:1f.2-scsi-1:0:0:0 sameas:/dev/disk/by-path/pci-0000:00:1f.2-scsi-0:0:0:0 
> instead? (I wonder if it is actually supported?)

iirc it wasn't supported, and that's why i went with "sameas:sda."  as
you mention, there is an outside chance that the wrong disk will be sda.
but in our env that'd just mean that an expendable os-disk gets blown
away (and it was already going to be reinstalled).

so yeah, not bullet-proof (and sameas: support for disk/by-* would be
nice) but works for me.

thanks...

andy

[1] my (certainly limited and possibly outdated) understanding is that
the by-path listings (udev's ENV{ID_PATH}) are not dependent on the
order devices are seen by the kernel.

-- 
andrew bezella <abezella at archive.org>
Internet Archive