Open Stack

On 9/29/2017 4:27 PM, Mark Voelker wrote:
> We talked about this a bit this past week…basically it boils down to:
> 
> 1.)  hodgepodge is looking into a schema update to accommodate microversion info if it’s even necessary.  If nothing else it may be informative…e.g. we can add a new capability and tests that happens to depend on a particular microversion to the Guidelines now without doing anything special, but it might be useful for folks to be able to easily see “oh hey, this requires version X.Y of the API”.
> 
> 2.)  There were four API’s mentioned in [1] that we wanted to consider.  As Rocky noted, we’ll be working on scoring for those shortly (off the top of my head I think zhipeng is playing point on Nova scoring for this cycle).  Basically that means taking a first pass at adding new capabilities to [2] and posting a patch we can debate/iterate on.  If any of those things are found to meet the necessary criteria [3] but don’t have tests yet, we should definitely write some tests (the sooner the better).  If they don’t score high enough…well, far be it from me to tell anyone to not write tests.  =)  But it’s probably less urgent from an interop point of view since that signals they turned out not currently be good candidates anyway.  If someone’s chomping at the bit to write those tests, awesome—they can do that regardless without waiting on us.  There’s some information in [4] about what constitutes a useful test for interop purposes that test writers may want to look over.
> 
> [1]https://etherpad.openstack.org/p/InteropDenver2017PTG_Microversions
> [2]http://git.openstack.org/cgit/openstack/interop/tree/working_materials/scoring.txt
> [3]https://github.com/openstack/interop/blob/master/doc/source/process/CoreCriteria.rst  < Note that a capability doesn’t need to meet*all*  of these…just enough to score 74 points.  See [2] for more some further color.
> [4]https://github.com/openstack/interop/blob/master/working_materials/interop_test_spec.rst
> 
> At Your Service,
> 
> Mark T. Voelker

Thanks for the detailed response Mark, I was able to figure things out 
enough to propose this for the compute API 2.2 microversion:

https://review.openstack.org/#/c/509955/

I put several notes in the commit message about my rationale with the 
scoring. The main stickler for me was the "required" one since 
microversions are by definition opt-in on the client side, so 
unless/until nova ever raises the minimum required microversion, a 
client can continue to still just use the v2.1 microversion which is 
backward compatible with the old v2.0 API.

I expect that the interop WG will have to figure out some changes in 
wording or weighting for 'required' when the capability depends on a 
microversion, but that's over my head.

-- 

Thanks,

Matt