Open Stack

On 10/18/2015 04:52 PM, Robert Collins wrote:
> On 11 October 2015 at 01:17, Jeremy Stanley <fungi at yuggoth.org> wrote:
>> On 2015-10-09 18:51:51 +0000 (+0000), Tim Bell wrote:
>>> There is a need to distinguish between server side py26 support
>>> which is generally under the control of the service provider and
>>> py26 support on the client side. For a service provider to push
>>> all of their hypervisors and service machines to RHEL 7 is under
>>> their control but requiring all of their users to do the same is
>>> much more difficult.
>> [...]
>>
>> Agreed, this is why the master branches of clients/libraries opted
>> to continue testing with Python 2.6 throughout the supported
>> lifetime of the stable/juno branches. But much like for the service
>> providers and users, continuing to run older Linux distributions
>> (CentOS 6.x, Ubuntu 12.4.x) in our test infrastructure comes at a
>> significant cost and at some point we need to free up our limited
>> sysadmin resources to focus on better support for more recent distro
>> releases.
>>
>> Once new patches to OpenStack projects are no longer actively tested
>> on these older platforms, we have to expect that the ability to run
>> new versions of the projects on them will quickly vanish. However,
>> users on those older distro releases can opt to continue using
>> whatever versions last supported them or perhaps fall back on the
>> versions provided by their distro's package manager.
>
> For pbr I'd actually value keeping 2.6 testing for a bit: because in
> setup_requires we're not capped anywhere - and those folk that *are*
> still using 2.6 shouldn't be broken by us - pbr is basically the last
> thing that can sanely move.
>
> I'd be happy with deadsnakes PPA on ubuntu, or a docker container or
> whatever - just having the confidence that we haven't accidentally
> broken anything until well past the point that the last release of a
> higher level library is no longer accepting bug reports - that would
> be the goal.

Maybe a non-voting job that adds the deadsnakes PPA and installs 2.6 in 
the job setup and then runs pbr tests with it.

I say non-voting because we'd be introducing a direct-internet consuming 
test, which increases the risk of failures - but given that the pbr team 
is already careful (given the nature of the project) an informational 
job should be good enough, no?