[oslo] Specs proposal to use cotyledon and futurist.
Hi there, Further to my previous mail, I have decided to create a specs proposal[1] within oslo. The idea is to use cotyledon and futurist in oslo.service and not to deprecate the project. Please don't hesitate to send me feedback. [1] https://review.opendev.org/c/openstack/oslo-specs/+/922597 -- Daniel Bengtsson Software Engineer
On Mon, 2024-06-24 at 12:00 +0200, Daniel Bengtsson wrote:
Hi there,
Further to my previous mail, I have decided to create a specs proposal[1] within oslo. The idea is to use cotyledon and futurist in oslo.service and not to deprecate the project.
im in favour of that approch and left some suggestions regarding the mechanics of how we could do that in the review. basically make cotyledon optional in this cycle make it the default in 2025.1 and deprecate the eventlet backend remove the eventlet support in 2025.2 or 2026.1 based on the progress of the eventlet removal work in the consuming projects. oslo messaging has a configurable executor https://github.com/openstack/oslo.messaging/blob/0d3338fd1d0094150491f819553... which the calling appllication initalise https://github.com/openstack/nova/blob/master/nova/rpc.py#L220-L226 we could either take the same approch with oslo.service or add a config option in oslo.service to choose the backbend. that would allow applications to move individual executable as that executable supprot the new backend. i.e. nova-scheduler could move to the new backbend before nova-compute. if there is a config option or environmental vairable either in the lib or applications that would allow use to experiment with the new backend in the ci on a per job basis too but thats just a side effect of keeping both implementations in oslo.service for a few release.
Please don't hesitate to send me feedback.
[1] https://review.opendev.org/c/openstack/oslo-specs/+/922597
On 2024-06-24 13:22, smooney@redhat.com wrote:
On Mon, 2024-06-24 at 12:00 +0200, Daniel Bengtsson wrote: im in favour of that approch and left some suggestions regarding the mechanics of how we could do that in the review. basically make cotyledon optional in this cycle make it the default in 2025.1 and deprecate the eventlet backend remove the eventlet support in 2025.2 or 2026.1 based on the progress of the eventlet removal work in the consuming projects. Thanks a lot Sean for your feedback. That sounds good and I will update my proposal.
we could either take the same approch with oslo.service or add a config option in oslo.service to choose the backbend. I think it's a good approch. Let's do it.
that would allow applications to move individual executable as that executable supprot the new backend. i.e. nova-scheduler could move to the new backbend before nova-compute. Yes right.
if there is a config option or environmental vairable either in the lib or applications that would allow use to experiment with the new backend in the ci on a per job basis too but thats just a side effect of keeping both implementations in oslo.service for a few release. Yes that sounds good.
-- Daniel Bengtsson Software Engineer
participants (2)
-
Daniel Bengtsson
-
smooney@redhat.com