[top-posting] While the flattening / moving work has made huge progress (thanks Dan for driving this effort!), I think it's time to agree if whether or not we backport the deployment directory for now. For new services added in Stein, which need to be backported in stable versions, they'll have to be backported with merge conflict & dealt with it. For bug fixes in existing services, we'll have to backport to stable branches with small merge conflict but if we backported the deployment directory, I think the conflicts would be bigger since stable branches would miss all the flattening/moving work (IMHO too big to backport). So I propose that we - do not backport deployment directory - deal with small merge conflicts when backporting patches in existing services - for new stein services, and when backport needed, put them in either docker or puppet directories in rocky and backport to queens eventually. Feedback is open. Thanks, On Thu, Oct 25, 2018 at 6:06 PM Dan Prince <dprince@redhat.com> wrote:
On Wed, Oct 17, 2018 at 11:15 AM Alex Schultz <aschultz@redhat.com> wrote:
Time to resurrect this thread.
On Thu, Jul 5, 2018 at 12:14 PM James Slagle <james.slagle@gmail.com>
wrote:
On Thu, Jul 5, 2018 at 1:50 PM, Dan Prince <dprince@redhat.com> wrote:
Last week I was tinkering with my docker configuration a bit and was
a
bit surprised that puppet/services/docker.yaml no longer used puppet to configure the docker daemon. It now uses Ansible [1] which is very cool but brings up the question of how should we clearly indicate to developers and users that we are using Ansible vs Puppet for configuration?
TripleO has been around for a while now, has supported multiple configuration ans service types over the years: os-apply-config, puppet, containers, and now Ansible. In the past we've used rigid directory structures to identify which "service type" was used. More recently we mixed things up a bit more even by extending one service type from another ("docker" services all initially extended the "puppet" services to generate config files and provide an easy upgrade path).
Similarly we now use Ansible all over the place for other things in many of or docker and puppet services for things like upgrades. That is all good too. I guess the thing I'm getting at here is just a way to cleanly identify which services are configured via Puppet vs. Ansible. And how can we do that in the least destructive way possible so as not to confuse ourselves and our users in the process.
Also, I think its work keeping in mind that TripleO was once a multi- vendor project with vendors that had different preferences on service configuration. Also having the ability to support multiple configuration mechanisms in the future could once again present itself (thinking of Kubernetes as an example). Keeping in mind there may be a conversion period that could well last more than a release or two.
I suggested a 'services/ansible' directory with mixed responces in our #tripleo meeting this week. Any other thoughts on the matter?
I would almost rather see us organize the directories by service name/project instead of implementation.
Instead of:
puppet/services/nova-api.yaml puppet/services/nova-conductor.yaml docker/services/nova-api.yaml docker/services/nova-conductor.yaml
We'd have:
services/nova/nova-api-puppet.yaml services/nova/nova-conductor-puppet.yaml services/nova/nova-api-docker.yaml services/nova/nova-conductor-docker.yaml
(or perhaps even another level of directories to indicate puppet/docker/ansible?)
Personally, such an organization is something I'm more used to. It feels more similar to how most would expect a puppet module or ansible role to be organized, where you have the abstraction (service configuration) at a higher directory level than specific implementations.
It would also lend itself more easily to adding implementations only for specific services, and address the question of if a new top level implementation directory needs to be created. For example, adding a services/nova/nova-api-chef.yaml seems a lot less contentious than adding a top level chef/services/nova-api.yaml.
It'd also be nice if we had a way to mark the default within a given service's directory. Perhaps services/nova/nova-api-default.yaml, which would be a new template that just consumes the default? Or perhaps a symlink, although it was pointed out symlinks don't work in swift containers. Still, that could possibly be addressed in our plan upload workflows. Then the resource-registry would point at nova-api-default.yaml. One could easily tell which is the default without having to cross reference with the resource-registry.
So since I'm adding a new ansible service, I thought I'd try and take a stab at this naming thing. I've taken James's idea and proposed an implementation here: https://review.openstack.org/#/c/588111/
The idea would be that the THT code for the service deployment would end up in something like:
deployment/<service>/<specific-service-element>-<implementation>.yaml
A matter of preference but I can live with this.
Additionally I took a stab at combining the puppet/docker service definitions for the aodh services in a similar structure to start reducing the overhead we've had from maintaining the docker/puppet implementations seperately. You can see the patch https://review.openstack.org/#/c/611188/ for an additional example of this.
Please let me know what you think.
I'm okay with it in that it consolidates some things (which we greatly need to do). It does address my initial concern in that people are now putting Ansible services into the puppet/services directory albeit a bit heavy handed in that it changes everything (rather than just the new Ansible services). Understood that you are also eliminating the "inheritance" from the docker/services to the puppet/services files... by simply eliminating the Puppet varients. I had hoped to implement a baremetal Packstack like installer (still somewhat popular) with t-h-t but there doesn't appear to be anyone in the Packstack camp speaking up for the idea here. So much for consolidation. There doesn't seem to be anyone else in TripleO (aside from myself) who wants baremetal so I guess be gone with it. Opportunity to test on baremetal lost! Like we discussed we could have easily kept both and used jinja templates to avoid the "inheritance" across the baremetal(puppet) and docker/services and thus also minimized our Heat resources which have grown. There *are* ways to implement this addressing all concerns and keeping both features. We are choosing not do this however.
I see people are already +2'ing the patches so it appears to have been decided. But, Are we going to commit to trying to move all the existing services to this new format within a release? There are 200 or so t-h-t patches up for review that would likely need rebased. And yes this will make backports more difficult as well.
In all this is we are changing our entire directory structure but not really removing any dependencies. I suppose if we are going to do the work I'd rather see us drop some dependencies in the process too :/. Where destructive meets productive is our best investment I think.
Dan
Thanks, -Alex
-- -- James Slagle --
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions) Unsubscribe: OpenStack-dev-request@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions) Unsubscribe: OpenStack-dev-request@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: OpenStack-dev-request@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
-- Emilien Macchi