Install vexxhost magnum-cluster-api driver

About this repository

This repository includes playbooks and roles to deploy the Vexxhost magnum-cluster-api driver for the OpenStack Magnum service.

The playbooks create a complete deployment including the control plane k8s cluster which should result in a ready-to-go experience for operators.

The following architectural features are present:

  • The control plane k8s cluster is an integral part of the openstack-ansible deployment, and forms part of the foundational components alongside mariadb and rabbitmq.

  • The control plane k8s cluster is deployed on the infra hosts and integrated with the haproxy loadbalancer and OpenStack internal API endpoint, and not exposed outside of the deployment

  • SSL is supported between all components and configuration is possible to support different certificate authorities on the internal and external loadbalancer endpoints.

  • Control plane traffic can stay entirely within the management network if required

  • The magnum-cluster-api-proxy service is deployed to allow communication between the control plane and workload clusters when a floating IP is not attached to the workload cluster.

  • It is possible to do a completely offline install for airgapped environments

The magnum-cluster-api driver for magnum can be found here https://github.com/vexxhost/magnum-cluster-api

Documentation for the Vexxhost magnum-cluster-api driver is here https://vexxhost.github.io/magnum-cluster-api/

The ansible collection used to deploy the controlplane k8s cluster is here https://github.com/vexxhost/ansible-collection-kubernetes

The ansible collection used to deploy the container runtime for the controlplane k8s cluster is here https://github.com/vexxhost/ansible-collection-containers

These playbooks require Openstack-Ansible Caracal or later.

Highlevel overview of the Magnum infrastructure these playbooks will build and operate against.

OSA Magnum Cluster API Architecture

Pre-requisites

  • An existing openstack-ansible deployment

  • Control plane using LXC containers, bare metal deployment is not tested

  • Core openstack services plus Octavia

OpenStack-Ansible Integration

The playbooks are distributed as an ansible collection, and integrate with Openstack-Ansible by adding the collection to the deployment host by adding the following to /etc/openstack_deploy/user-collection-requirements.yml under the collections key.

collections:
  - name: vexxhost.kubernetes
    source: https://github.com/vexxhost/ansible-collection-kubernetes
    type: git
    version: 896d67092706d01b6fb9d7f4856750819f8898cf
  - name: osa_ops.mcapi_vexxhost
    type: git
    version: master
    source: https://opendev.org/openstack/openstack-ansible-ops#/mcapi_vexxhost

The collections can then be installed with the following command:

cd /opt/openstack-ansible
openstack-ansible scripts/get-ansible-collection-requirements.yml

The modules in the kubernetes collection require an additional python module to be present in the ansible-runtime python virtual environment. Specify this in /etc/openstack_deploy/user-ansible-venv-requirements.txt

docker-image-py
kubernetes

OpenStack-Ansible configuration for magnum-cluster-api driver

Specify the deployment of the control plane k8s cluster in /etc/openstack_deploy/env.d/k8s.yml

---
component_skel:
  k8s_capi:
    belongs_to:
      - k8s_all

container_skel:
  k8s_container:
    belongs_to:
      - cluster-api_containers
    contains:
      - k8s_capi

physical_skel:
  cluster-api_containers:
    belongs_to:
      - all_containers
  cluster-api_hosts:
    belongs_to:
      - hosts

Define the physical hosts that will host the controlplane k8s cluster in /etc/openstack_deploy/conf.d/k8s.yml. This example is for an all-in-one deployment and should be adjusted to match a real deployment with multiple hosts if high availability is required.

cluster-api_hosts:
  aio1:
    ip: 172.29.236.100

Integrate the control plane k8s cluster with the haproxy loadbalancer in /etc/openstack_deploy/group_vars/k8s_all/haproxy_service.yml

---
# Copyright 2023, BBC R&D
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

haproxy_k8s_service:
  haproxy_service_name: k8s
  haproxy_backend_nodes: "{{ groups['k8s_all'] | default([]) }}"
  haproxy_ssl: false
  haproxy_ssl_all_vips: false
  haproxy_port: 6443
  haproxy_balance_type: tcp
  haproxy_balance_alg: leastconn
  haproxy_interval: '15000'
  haproxy_backend_port: 6443
  haproxy_backend_rise: 2
  haproxy_backend_fall: 2
  haproxy_timeout_server: '15m'
  haproxy_timeout_client: '5m'
  haproxy_backend_options:
    - tcplog
    - ssl-hello-chk
    - log-health-checks
    - httpchk GET /healthz
  haproxy_backend_httpcheck_options:
    - 'send hdr User-Agent "osa-haproxy-healthcheck" meth GET uri /healthz'
  haproxy_backend_server_options:
    - check-ssl
    - verify none
  haproxy_service_enabled: "{{ groups['k8s_all'] is defined and groups['k8s_all'] | length > 0 }}"

k8s_haproxy_services:
  - "{{ haproxy_k8s_service | combine(haproxy_k8s_service_overrides | default({})) }}"

Configure the LXC container that will host the control plane k8s cluster to be suitable for running nested containers in /etc/openstack_deploy/group_vars/k8s_all/main.yml

lxc_container_config_list:
  - "lxc.apparmor.profile=unconfined"

lxc_container_mount_auto:
  - "proc:rw"
  - "sys:rw"

Set up config-overrides for the magnum service in /etc/openstack_deploy/user_variables_magnum.yml. Adjust the images and flavors here as necessary, these are just for demonstration. Upload as many images as you need for the different workload cluster kubernetes versions.

Attention must be given to the SSL configuration. Users and workload clusters will interact with the external endpoint and must trust the SSL certificate. The magnum service and cluster-api can be configured to interact with either the external or internal endpoint and must trust the SSL certificiate. Depending on the environment, these may be derived from different certificate authorities.

---
# Copyright 2020, VEXXHOST, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# install magnum-cluster-api and kubernetes python package into magnum venv
magnum_user_pip_packages:
  - git+https://github.com/vexxhost/magnum-cluster-api@main#egg=magnum-cluster-api
  - kubernetes

# ensure that the internal VIP CA is trusted by the CAPI driver
# TODO - this is too hardwired somehow
magnum_config_overrides:
  drivers:
    openstack_ca_file: '/usr/local/share/ca-certificates/ExampleCorpRoot.crt'
  capi_client:
    ca_file: '/usr/local/share/ca-certificates/ExampleCorpRoot.crt'
    endpoint: 'internalURL'
  cluster_template:
    kubernetes_allowed_network_drivers: 'calico'
    kubernetes_default_network_driver: 'calico'
  certificates:
    cert_manager_type: x509keypair

# use the openstack_resources role in the capi functional test rather than os_magnum
# to create the test resources
magnum_glance_images: []
magnum_cluster_templates: []
magnum_flavors: []

Set up config-overrides for the control plane k8s cluster in /etc/openstack_deploy/user_variables_k8s.yml` These variables integrate the control plane k8s deployment with the rest of the openstack-ansible deployment.

# wire OSA group, host and network addresses into k8s deployment
kubernetes_control_plane_group: k8s_container
kubelet_hostname: "{{ ansible_facts['hostname'] }}"
kubelet_node_ip: "{{ management_address }}"
kubernetes_hostname: "{{ internal_lb_vip_address }}"
kubernetes_non_init_namespace: true

# Pick a range of addresses for cilium that do not collide with anything else
cilium_ipv4_cidr: 172.29.200.0/22

# Set this manually, or kube-proxy will try to do this - not possible
# in a non-init namespace and will fail in LXC
openstack_host_nf_conntrack_max: 1572864

# OSA containers dont run ssh by default so cannot use synchronize
upload_helm_chart_method: copy

Run the deployment

For a new deployment

Run the OSA playbooks/setup.yml playbooks as usual, following the normal deployment guide.

Run the magnum-cluster-api deployment

openstack-ansible osa_ops.mcapi_vexxhost.k8s_install

For an existing deployment

Ensure that the python modules required for ansible are present:

./scripts/bootstrap-ansible.sh

Alternatively, without re-running the bootstrap script:

/opt/ansible-runtime/bin/pip install docker-image-py

Add the magnum-cluser-api driver to the magnum service

openstack-ansible playbooks/os-magnum-install.yml

Create the k8s control plane containers

openstack-ansible playbooks/lxc-containers-create.yml --limit k8s_all

Run the magnum-cluster-api deployment

openstack-ansible osa_ops.mcapi_vexxhost.k8s_install

Optionally run a functional test of magnum-cluster-api

This can be done quickly using the following playbook

openstack-ansible osa_ops.mcapi_vexxhost.functional_test

This playbook will create a neutron public network, download a prebuilt k8s glance image, create a nova flavor and a magnum cluster template.

It will then deploy the workload k8s cluster using magnum, and run a sonobouy “quick mode” test of the workload cluster.

This playbook is intended to be used on an openstack-ansible all-in-one deployment.

Use Magnum to create a workload cluster

Upload Images

Create a cluster template

Create a workload cluster

Optional Components

Deploy the workload clusters with a local registry

TODO - describe how to do this

Deploy the control plane cluster from a local registry

TODO - describe how to do this

Use of magnum-cluster-api-proxy

TODO - describe what this is for

Troubleshooting

Local testing

An OpenStack-Ansible all-in-one configured with Magnum and Octavia is capable of running a functioning magnum-cluster-api deployment.

Sufficient memory should be available beyond the minimum 8G usually required for an all-in-one. A multinode workload cluster may require nova to boot several Ubuntu images in addition to an Octavia loadbalancer instance. 64G would be an appropriate amount of system RAM.

There also must be sufficient disk space in /var/lib/nova/instances to support the required number of instances - the normal minimum of 60G required for an all-in-one deployment will be insufficient, 500G would be plenty.