[openstack-dev] [tripleo] POC patch for using tripleo-repos for repo setup

Ben Nemec openstack at nemebean.com
Wed Nov 23 16:07:50 UTC 2016



On 11/22/2016 08:18 PM, Emilien Macchi wrote:
> Even if I was part of those who approved this feature, I still have
> some comments, inline:
>
> On Tue, Nov 22, 2016 at 1:30 PM, Alex Schultz <aschultz at redhat.com> wrote:
>> On Tue, Nov 22, 2016 at 11:06 AM, Ben Nemec <openstack at nemebean.com> wrote:
>>>
>>>
>>> On 11/21/2016 05:26 PM, Alex Schultz wrote:
>>>>
>>>> On Mon, Nov 21, 2016 at 2:57 PM, Ben Nemec <openstack at nemebean.com> wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> I have a POC patch[1] up to demonstrate the use of the tripleo-repos tool
>>>>> [2] as a replacement for most of tripleo.sh --repo-setup.  It has now
>>>>> passed
>>>>> all of the CI jobs so I wanted to solicit some feedback.
>>>>>
>>>>> There are a few changes related to repo naming because the tool names
>>>>> repo
>>>>> files based on the repo name rather than always calling them something
>>>>> generic like delorean.repo.  I think it's less confusing to have the
>>>>> delorean-newton repo file named delorean-newton.repo, but I'm open to
>>>>> discussion on that.
>>>>>
>>>>> If no one has any major objections to how the tool looks/works right now,
>>>>> I'll proceed with the work to get it imported into the openstack
>>>>> namespace
>>>>> as part of TripleO.  We can always iterate on it after import too, of
>>>>> course, so this isn't a speak now or forever hold your peace situation.
>>>>> :-)
>>>>>
>>>>
>>>> Why a python standalone application for the management of specifically
>>>> (and only?) tripleo repositories.  It seems we should be trying to
>>>> leverage existing tooling and not hiding the problem behind a python
>>>> app.  It's not that I enjoy the current method described in the spec
>>>> (3 curls, 1 sed, 1 bash thing, and a yum install) but it seems like to
>>>> write 586 lines of python and tests might be the wrong approach.
>>>> Would it be better to just devote some time to rpm generation for
>>>> these and deliver it in discrete RPMs?  'yum install
>>>> http://tripleo.org/repos-current.rpm' seems way more straight forward.
>>>
>>>
>>> That's essentially trading "curl ..." for "yum install ..." in the docs.
>>> The repo rpm would have to be part of the delorean build so you'd still have
>>> to be pointing people at a delorean repo.  It would also still require code
>>> changes somewhere to handle the mixed current/current-tripleo setup that we
>>> run for development and test. Given how specific to tripleo that is I'm not
>>> sure how much sense it makes to implement it elsewhere.
>>>
>>
>> I'm asking because essentially we're delivering basically static repo
>> files.  Which should really be done via RPM. Upgrades and cleanups are
>> already well established practices between RPMs.  I'm not seeing the
>> reasoning why a python app.  I thought about this further and I'm not
>> sure why this should be done on the client side via an app rather than
>> at repository build/promotion time.  As long as we're including the
>> repo rpm, we can always create simple 302 redirects from a webserver
>> to get the latest version.  I don't see why we should introduce a
>> client tool for this when the action is really on the repository
>> packaging side.   This seems odd doing system configuration via a
>> python script rather than a configuration management tool like
>> ansible, puppet or even just packaging.
>>
>>> There are also optional ceph and opstool repos and at least ceph needs to
>>> match the version of OpenStack being installed.  Sure, you could just
>>> document all that logic, but then the logic has to be reimplemented in CI
>>> anyway so you end up with code to maintain either way.  At least with one
>>> tool the logic is shared between the user and dev/test paths, which is one
>>> of the primary motivations behind it.  Pretty much every place that we have
>>> divergence between what users do and what developers do becomes a pain point
>>> for users because developers only fix the things they actually use.
>>>
>>
>> Yes we should not have a different path for dev/test vs operational
>> deployments, but I'm not convinced we need to add a custom python tool
>> to handle this only for tripleo.  There are already established
>> patterns of repository rpm delivery and installation via existing
>> tools.  What are we getting from this tool that can't already be done?
>
> That is true, here are some of them:
> - centos-release-ceph-(hammer|jewel) rpm that deploys repos.
> - since we are moving TripleO CI to use tripleo-quickstart, we could
> handle repository with Ansible, directly in the roles.

This is exactly what I'm trying to avoid here.  I want us to be using 
the same thing for repo management in CI and dev and real user 
environments.  Unless we're making quickstart the new API for TripleO 
this basically defeats the whole purpose of redoing the repo setup IMHO.

>
>>> There are other benefits too - the tool cleans up old repos so you don't
>>> have to worry about accidentally being left with a stray repo file that
>>> could cause problems.  You can also install the repos to a non-system path
>>> so you can make use of them without actually installing them on the host
>>> system.  I'm not entirely clear on the use case for that, but it's something
>>> that comes up from time to time.
>>>
>>
>> If it's an rpm, the upgrades should cleanup after themselves.  Do we
>> have specific and documented use cases where dropping the repo
>> information into an alternative location is required?  Even with rpm,
>> can't we just rpm --prefix=/some/other/location?
>>
>> Thanks,
>> -Alex
>>
>>
>>>
>>>>
>>>> Thanks,
>>>> -Alex
>>>>
>>>>> 1: https://review.openstack.org/#/c/395813
>>>>> 2:
>>>>>
>>>>> https://specs.openstack.org/openstack/tripleo-specs/specs/policy/tripleo-repos.html
>>>>> (note that this spec was mistakenly submitted as a policy, it will be
>>>>> moving
>>>>> to the ocata directory soon)
>>>>>
>>>>> Thanks.
>>>>>
>>>>> -Ben
>
> I don't have a strong opinion right now, as Ben already wrote the
> tool.  Though I would be interested to investigate in the future how
> we could maybe integrate this repo managing directly in Ansible within
> quickstart, as part of a regular task in the roles.
> I'm fine with starting to use this tool as a first iteration of
> "making our repo management better", but I share Alex thoughts here
> and we might want to iterate again later on this topic.
>

I'm fine with someone investigating an alternative approach, but I still 
think the other proposals here don't fully address the problems I'm 
trying to solve, and in some cases just move code to places we don't 
gate and may not be appropriate for tripleo-specific code.



More information about the OpenStack-dev mailing list