Open Stack

Greetings Michael,

I believe It would need to involve multiple machines at the same time.

I guess there are two different approaches that I think _could_ be
taken to facilitate this:

1) Provide a facility to use a specific volume as the "golden volume"
to boot up for IPA, and then initiate copies of that volume. The
downside that I see is the act of either copying the volume, or
presenting a snapshot of it that will be deleted a little later. I
think that is really going to depend on the back-end, and if the
backend can handle it or not. :\

2) The other possibility that I'm wondering is if we boot machines to
a read-only "golden volume", with a bootloader and a ramdisk that ends
up switching root over to a file stored on the filesystem being
mounted via the loopback. It seems evil, but I think it should work
just fine and does not involve as much network traffic as copying all
of the contents of a huge ramdisk over the wire into RAM. Of course,
then the problem is keeping kernels/drivers in sync...

I guess the question becomes, what is more appealing from an operator
standpoint, in terms of care and feeding. I suspect #2 would involve
some very specific ramdisk modifications, which may not improve the
hurdle being encountered with iterating with drivers faster.

I can't help but wonder if the headache is building an image with dib,
uploading it, then updating the node driver_info, and then deploying
to that node again.

Maybe if we could have dib write a fully formed image out to an iscsi
target... Maybe that might make the world moderately happy?

The more I think about it, the more I like #1 from a simplicity
standpoint that we could iterate upon, we would just likely have to
classify the feature as "very experimental" and stress operational
behavior issues that would/could arise, and maybe at some point
provide some tunable behavior settings as we better understand needs.

-Julia

On Wed, Nov 22, 2017 at 6:45 PM, Michael Still <mikal at stillhq.com> wrote:
> Thanks for this summary. I'd say the cinder-booted IPA is definitely of
> interest to the operators I've met. Building new IPAs, especially when
> trying to iterate on what drivers are needed is a pain so being able to
> iterate faster would be very useful. That said, I guess this implies booting
> more than one machine off a volume at once?
>
> Michael
>