Open Stack

On 07/08/2014 03:28 AM, Robert Collins wrote:
> We got to the current place thusly:
> 
>  - The initial ramdisk code in Nova-baremetal needed a home other than
> a wikipage.
>  - And then there was a need/desire to build ramdisks from a given
> distro != current host OS.
>  - Various bits of minor polish to make it more or less in-line with
> the emerging tooling in DIB.
> 
> I've no attachment at all to the ramdisk implementation though -
> there's nothing to say its right or wrong, and if dracut is better,
> cool.
> 
> The things I think are important are:
>  - ability to build for non-host arch and distro.

I'm still building ramdisks in a dib chroot, so this shouldn't be a problem.

>  - ability to build working Ironic-deploy ramdisks today, and
> Ironic-IPA ramdisks in future

I'll double-check, but for the moment I'm keeping all the init fragment
functionality untouched, so anything we can build today should also work
in my proposed dracut ramdisks.

For anyone who's interested, I've got a PoC for this up in Gerrit:
https://review.openstack.org/105275  It obviously needs some work, but I
think it shows a workable way forward with this.

I'll get a spec written up so we can have a formal discussion over
requirements and implementation.

Thanks.

-Ben

> 
> -Rob
> 
> 
> 
> On 4 July 2014 15:12, Ben Nemec <openstack at nemebean.com> wrote:
>> I've recently been looking into using dracut to build the
>> deploy-ramdisks that we use for TripleO.  There are a few reasons for
>> this: 1) dracut is a fairly standard way to generate a ramdisk, so users
>> are more likely to know how to debug problems with it.  2) If we build
>> with dracut, we get a lot of the udev/net/etc stuff that we're currently
>> doing manually for free.  3) (aka the self-serving one ;-) RHEL 7
>> doesn't include busybox, so we can't currently build ramdisks on that
>> distribution using the existing ramdisk element.
>>
>> For the RHEL issue, this could just be an alternate way to build
>> ramdisks, but given some of the other benefits I mentioned above I
>> wonder if it would make sense to look at completely replacing the
>> existing element.  From my investigation thus far, I think dracut can
>> accommodate all of the functionality in the existing ramdisk element,
>> and it looks to be available on all of our supported distros.
>>
>> So that's my pitch in favor of using dracut for ramdisks.  Any thoughts?
>>  Thanks.
>>
>> https://dracut.wiki.kernel.org/index.php/Main_Page
>>
>> -Ben
>>
>> _______________________________________________
>> OpenStack-dev mailing list
>> OpenStack-dev at lists.openstack.org
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> 
> 
>