Open Stack

Agree. The motivation of pulling templates out of Magnum tree is hoping these templates can be leveraged by a larger community and get more feedback. However, it is unlikely to be the case in practise, because different people has their own version of templates for addressing different use cases. It is proven to be hard to consolidate different templates even if these templates share a large amount of duplicated code (recall that we have to copy-and-paste the original template to add support for Ironic and CoreOS). So, +1 for stopping usage of heat-coe-templates.

Best regards,
Hongbin

-----Original Message-----
From: Tom Cammann [mailto:tom.cammann at hp.com] 
Sent: June-29-15 11:16 AM
To: openstack Development Mailing List (not for usage questions)
Subject: [openstack-dev] [Magnum] Continuing with heat-coe-templates

Hello team,

I've been doing work in Magnum recently to align our templates with the "upstream" templates from larsks/heat-kubernetes[1]. I've also been porting these changes to the stackforge/heat-coe-templates[2] repo.

I'm currently not convinced that maintaining a separate repo for Magnum templates (stackforge/heat-coe-templates) is beneficial for Magnum or the community.

Firstly it is very difficult to draw a line on what should be allowed into the heat-coe-templates. We are currently taking out changes[3] that introduced "useful" autoscaling capabilities in the templates but that didn't fit the Magnum plan. If we are going to treat the heat-coe-templates in that way then this extra repo will not allow organic development of new and old container engine templates that are not tied into Magnum.
Another recent change[4] in development is smart autoscaling of bays which introduces parameters that don't make a lot of sense outside of Magnum.

There are also difficult interdependency problems between the templates and the Magnum project such as the parameter fields. If a required parameter is added into the template the Magnum code must be also updated in the same commit to avoid functional test failures. This can be avoided using "Depends-On: 
#xxxxxx"
feature of gerrit, but it is an additional overhead and will require some CI setup.

Additionally we would have to version the templates, which I assume would be necessary to allow for packaging. This brings with it is own problems.

As far as I am aware there are no other people using the heat-coe-templates beyond the Magnum team, if we want independent growth of this repo it will need to be adopted by other people rather than Magnum commiters.

I don't see the heat templates as a dependency of Magnum, I see them as a truly fundamental part of Magnum which is going to be very difficult to cut out and make reusable without compromising Magnum's development process.

I would propose to delete/deprecate the usage of heat-coe-templates and continue with the usage of the templates in the Magnum repo. How does the team feel about that?

If we do continue with the large effort required to try and pull out the templates as a dependency then we will need increase the visibility of repo and greatly increase the reviews/commits on the repo. We also have a fairly significant backlog of work to align the heat-coe-templates with the templates in heat-coe-templates.

Thanks,
Tom

[1] https://github.com/larsks/heat-kubernetes
[2] https://github.com/stackforge/heat-coe-templates
[3] https://review.openstack.org/#/c/184687/
[4] https://review.openstack.org/#/c/196505/

__________________________________________________________________________
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