Open Stack

On Mon, Mar 20, 2017 at 12:10 PM, gordon chung <gord at live.ca> wrote:
>
>
> On 20/03/17 11:37 AM, Thomas Goirand wrote:
>
>>
>> I really don't understand why the Telemetry team insists in being
>> release-independent, out of big tent and such, when the reality is that
>> all of released Telemetry components are *very tightly* bound to a
>> specific versions of OpenStack. IMO, it doesn't make sense upstream, or
>> downstream of Telemetry.
>
> i believe the tightly coupled perception between gnocchi+ceilometer is a
> misconception. ceilometer can be configured to output to various targets
> that are not gnocchi. based on dev questions in irc, this is a common
> workflow that people are actively leveraging. aodh and panko are
> definitely more bound to ceilometer as they don't have any other sources
> (currently).
>
>>
>> Now, having Gnocchi out of the OpenStack infra is to me a step in the
>> wrong direction. We should aim at full integration with the rest of
>> OpenStack, not getting out.
>>
>
> i should re-iterate, this won't change our testing or integration.
> ceilometer has a gate that ensures compatibility with gnocchi as a
> target. this will remain and the auto-scaling
> aodh+ceilometer+gnocchi+heat use case will continue to be validated. not
> sure how we can quantify/qualify 'full integration' but we remain
> committed to ensuring gnocchi+ceilometer works.
>
> the use case for gnocchi is generic. if you have to store a bunch of
> timestamp+value data, use gnocchi. the use case definitely fits
> openstack's requirement, but i believe you can see it isn't just limited
> to that.
>
> i'm glad we have your opinion here, i had previously asked jd about
> effects on packaging and while i think Red Hat has a plan already, it'd
> be interesting to get your feedback on how this will affects other distros.

Keep in mind, that OpenStack inside of Debian is just Thomas for a
variety of reasons. Others have tried to help and are trying to help
and aren't really able to stick around.

The effects on downstreams shouldn't be significant. People packaging
Ceilometer likely already package Gnocchi. How those packagers choose
to consume deliverables is what will change. In a similar vein,
OpenStack Infra has a signing key for each release cycle that Gnocchi
is likely currently signed with when tarballs are released. You may
receive complaints that new releases aren't verifiable in the same
way, but that and needing a decent MANIFEST.in (assuming you're also
dropping your usage of PBR) will probably be the largest problems
(beyond where downstreams get the actual deliverable).

In the end, I think this should inform your decision but not make it.
-- 
Ian Cordasco