[CentOS-devel] Building a downstream CentOS Atomic Host

Thu May 7 03:54:29 UTC 2015
Ian McLeod <imcleod at redhat.com>

On 05/06/2015 07:53 AM, Ian McLeod wrote:
> On 05/05/2015 08:36 AM, Ian McLeod wrote:
>> On 05/05/2015 08:08 AM, Karanbir Singh wrote:
>>> On 05/05/2015 01:46 PM, Colin Walters wrote:
>>>> On Tue, May 5, 2015, at 08:31 AM, Ian McLeod wrote:
>>>>>
>>>>> I will doublecheck with Colin on this, but as best I can tell, the
>>>>> "rpm-ostree-toolbox installer" input allows only repo URLs and package
>>>>> name based excludes. 
>>>>
>>>> Yes, unfortunately.  This is actually a lorax restriction in that it
>>>> only takes repo URLs as arguments, not repo files that would
>>>> have `excludes=` processed.
>>>
>>> not a problem, i will get the repo setup today. If we can make sure the
>>> PR's are all in, I can also run through the build and post results.
>>>
>>> - KB
>>>
>>>
>>
>> So, I did a quick experiment and added explicit NVR package excludes
>> using the "lorax_exclude_packages" pass-through option.  It's a bit of a
>> hack, and it will stop working and/or need to be refreshed when a newer
>> anaconda appears in the upstream CentOS repos, but by that point our
>> 7.1.1 downstream should be out the door and done.
>>
>> Here is the change:
>>
>> https://github.com/CentOS/sig-atomic-buildscripts/pull/30/files#diff-38589f3a5b1e8e2eca835f956b204e0eR20
>>
>> It seems to have worked.
> 
> Can we quickly establish a location for the candidate composed trees and
> lorax install trees, and then do some test runs with the content in the
> "downstream" tree?
> 
> It's really impossible to do an end to end test of the repo content if
> there's no established location (and ref name) to put into the kickstart
> used to generate the image.

Some quick further updates here:

I have been unable to generate a working install tree using the current
mirror content combined with the "cah" repo.  The lorax runs complete,
but the resulting Anaconda fails during installs in one way or another.

I initially hit some python-blivet version mismatch issues.  Even after
fixing these, I end up with an install that hangs when creating the
initial filesystems.  At this point, I attempted to dial back the kernel
version pulled in by lorax, to match the somewhat older version present
in the RHELAH installer.  This, it seems, is impossible using the
existing mirror content, which is now irreversibly newer.

If, however, I use the composed ostree generated from the current
"downstream" branch of the Atomic SIG repo, combined with the Anaconda
used to generate the SIG images, all is well.  Using local runs of Image
Factory, I've generated a vagrant/libvirt box that boots successfully
into Atomic and this system contains the "cah" repo packages.  To be
clear, I'm talking about the install tree hosted here:

http://buildlogs.centos.org/centos/7/atomic/x86_64/atomic-anaconda-nightly/latest/

So, again, if we can get the composed tree generated from the
"downstream" branch of the SIG repo hosted on buildlogs, it should be
possible to use the CBS to generate both cloud and vagrant images that
qualify, roughly speaking, as a downstream rebuild of the 7.1.1 RHELAH
release.

Thoughts?

> 
>>
>> -Ian
>> _______________________________________________
>> CentOS-devel mailing list
>> CentOS-devel at centos.org
>> http://lists.centos.org/mailman/listinfo/centos-devel
>>
>>
> 
> _______________________________________________
> CentOS-devel mailing list
> CentOS-devel at centos.org
> http://lists.centos.org/mailman/listinfo/centos-devel
> 
>