[openstack-dev] [nova][api] does validation bug-fix require microversion bump?

Sylvain Bauza sbauza at redhat.com
Mon Dec 21 09:19:08 UTC 2015



Le 21/12/2015 08:25, Ken'ichi Ohmichi a écrit :
> Hi nova-api team,
>
> I'd like to get a feedback about the way to bump a microversion.
>
> Short version:
>    We found a validation bug on Nova v2.1 API.
>    To fix the bug, do we need to bump a new microversion?
>
> Long version:
> As LP bug report[1], nova v2.0 API allows a list of server-IDs on
> scheduler_hint "different_host" like
>
>      "os:scheduler_hints": {
>          "different_host": [
>              "099b8bee-9264-48fe-a745-45b22f7ff79f",
>              "99644acc-8893-4656-9481-0114efdbc9b6"
>          ]
>      }
>
> on "create a server" API.
> However, nova v2.1 API is handling this request as invalid because the
> validation implementation way is wrong now.
> Nova v2.1 API should allow the list of server-IDs for backwards compatibility.
>
> We are trying to fix this bug on
> https://review.openstack.org/#/c/259247/ , and we have a question to
> fix it.
> This fix is API change even if fixing the bug, so do we need to bump a
> microversion?
>
> The one usage of microversions is notification of API change.
> If bumping it, nova can notify the fixing with a microversion.
>
> This fix should be applied to stable branches also because of helping
> the existing users.
> So if bumping a microversion on stable branch also, the microversion
> number meanings become different between clouds which are deployed
> with different nova releases.
> So we(John, Alex, me) are guessing we should not bump a microversion
> on stable branches. but if doing that, nova cannot notify the fixing
> on stable branches.
>
> Now I am feeling this fixing will be applied without a microversion
> bump because it is nice to avoid different microversion meanings of
> master/stable branches.
> Is it fine for us?

It looks like a regression for the list, but the operator can still 
provide only one uuid if needed.
Providing a microversion for that would mean that V2.0 on v2.1 would 
have a different behaviour vs. the legacy /v2.0, which is bad IMHO.

Also, like you said, backporting the microversion to stable/liberty is 
bad too.


Those above points make me agree with you, we just need to fix the bug 
without creating a microversion IMHO.

-Sylvain


> Thanks
> Ken Ohmichi
>
> ---
> [1]: https://launchpad.net/bugs/1521928
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev




More information about the OpenStack-dev mailing list