kolla-ansible / kolla-kubernetes neutron/tasks/config.yml mismatch

[Andrew Bruce]

Just deployed Kolla following guide at https://docs.openstack.org/developer/kolla-kubernetes/deployment-guide.html

Problem is that neutron-dhcp-agent, neutron-l3-agent, and neutron-metadata-agent all fail to start because the ml2-conf.ini key is missing from the generated configMap.

The cause is a mismatch between kolla-ansible and kolla-kubernetes. Both contain a file named 'ansible/roles/neutron/tasks/config.yml' and the two files are different. Specifically, see services_need_ml2_conf_ini (file https://github.com/andybrucenet/kolla-ansible/blob/dec8fbd7a463ab58d5d5455fcfc45b0302c1a707/ansible/roles/neutron/tasks/config.yml)

Check out the fork at https://github.com/andybrucenet/kolla-ansible. I'll submit a pull request after I see that this bug has been reviewed.

[Andrew Bruce (andybrucenet)]

Change is specific to kolla-ansible

[Eduardo Gonzalez (egonzalez90)]

In kolla-ansible we removed ml2 files from those services since it is a file that they don't use.
AFIK, kolla-kubernetes now generate its own config file.
I think this issue not affect kolla-ansible.
Change removing ml2 file from various services: https://review.openstack.org/#/c/447992/
Related bug for the change: https://bugs.launchpad.net/kolla-ansible/+bug/1677163
Regards

[Andrew Bruce (andybrucenet)]

Thank you for the quick response!

While it's fine with me that the kolla-ansible team decided that ml2_conf.ini is not needed (and I did see some IRC conversation on this), this change introduced breaking compatibility problems between kolla-ansible and kolla-kubernetes.

I understand these are two different efforts. However, keep in mind that for us end-consumers there is only *one* Kolla Bare Metal Deployment Guide (https://docs.openstack.org/developer/kolla-kubernetes/deployment-guide.html). And in that Guide, we find instructions to use kolla-ansible as part of the Kubernetes deployment.

As it is - the Kolla bare-metal process as currently written will fail when the Neutron agent containers (l3-agent, dhcp-agent, etc.) try to start. They get stuck in "Init" phase. The underlying Kubernetes logs clearly show that a configMap key is missing.

Specifically: the Bare Metal Deployment Guide instructs us:

>Generate the default configuration:

> sudo kolla-ansible genconfig

So kolla-ansible - according to the bare-metal build guide - is used to generate files which are later used by kolla-kubernetes.

Under the kolla-kubernetes file system, we find:

helm/microservice/neutron-dhcp-agent-daemonset/templates/dhcp-agent-daemonset.yaml

 21 apiVersion: extensions/v1beta1
 22 kind: DaemonSet
 23 metadata:
 24 name: neutron-dhcp-agent
 25 labels:
 26 component: neutron
 27 system: dhcp-agent
...
117 - name: neutron-dhcp-agent-configmap
118 configMap:
119 name: neutron-dhcp-agent
120 items:
121 - key: neutron.conf
122 path: neutron.conf
123 - key: config.json
124 path: config.json
125 - key: ml2-conf.ini
126 path: ml2_conf.ini
127 - key: dhcp-agent.ini
128 path: dhcp_agent.ini
129 - key: dnsmasq.conf
130 path: dnsmasq.conf

Plus, other commands for neutron-dhcp-agent throughout this file reference (and assume the existence of) ml2_conf.ini.

So, please advise me how I can notify the correct persons of this discrepancy. As a consumer, I simply want the deployment to work.

[Steven Dake (sdake)]

The deployment guide is wrong on the genconfig step.

Instead run this command from kolla-kubernetes:
ansible-playbook -e ansible_python_interpreter=/usr/bin/python -e @/etc/kolla/globals.yml -e @/etc/kolla/passwords.yml -e CONFIG_DIR=/etc/kolla ansible/site.yml

kolla-kubernetes has created its own copy of genconfig.

Thanks for the bug report!

If you want to submit a PR to fix the deployment documentation, that would be fantastic!

[Kevin Fox (kevpn)]

I think the deploy documentation is now wrong. kolla-ansible should not be used to generate the config or else these issues will keep poping up.

[Andrew Bruce (andybrucenet)]

Heard from the kolla-kubernetes team and they agree. I am going to create a PR on their side to remove all kolla-ansible dependencies and verify that Kolla deploys correctly.

Closing this bug. See https://bugs.launchpad.net/kolla-kubernetes/+bug/1693281 for the related bug in kolla-kubernetes.

[Andrew Bruce (andybrucenet)]

See https://bugs.launchpad.net/kolla-kubernetes/+bug/1693281 - problem is on the kolla-kubernetes side.