[openstack-dev] [kolla][kubernetes] One repo vs two

Michal Rostecki michal.rostecki at gmail.com
Mon May 2 14:27:58 UTC 2016

It seems for me that we have a dilemma between security (abstaining from 
creating a core group which may overuse their rights in kolla repo) and 
usability (not having multiple repos, which we experienced badly in the 
kolla-mesos era).

I don't find the argument about having k8s ecosystem in separate repo 
solid. Of course creating separate python package, being the standalone 
CLI, is not problematic. But the deployment of k8s itself for the 
development environment (i.e. Vagrant) or CI may be painful.

When developing kolla-mesos, we created ansible playbooks for deploying 
Mesos. We had to duplicate the kolla-ansible script and kolla_docker 
Ansible module. We also duplicated a lot from Vagrantfile. That was 
because any officially available method of deploying Mesos wasn't 
elastic enough to meet the requirements of kolla-mesos development. And 
I'm almost sure that the similar situation will be with k8s - I don't 
see any way to reuse official kube-up scripts or Salt manigests for the 
kolla-k8s needs.

Second thing, we implemented our own Dockerfiles for Mesos. I don't know 
whether it will be needed for k8s, maybe yes, maybe not. But if yes, 
then handling the build of underlay infra containers from kolla repo for 
the needs of kolla-kubernetes looks overcomplicated, exactly as it was 
between kolla and kolla-mesos.

Therefore, I'm in favor of having one repository. I prefer to monitor 
actions of kolla-k8s cores in the short period of time than begin with 
duplicates and technical debt since the beginning of development.


More information about the OpenStack-dev mailing list