Skip to content

Latest commit

 

History

History
98 lines (69 loc) · 3.75 KB

README.md

File metadata and controls

98 lines (69 loc) · 3.75 KB
Raw

KVM Lab Ansible Playbook

Build a set of virtual machines on a single KVM host specifically for deploying a K8s home lab.

My goal was to have a virtual infrastructure allowing to integrate a multi-node K8s cluster with:

  • metallb with BGP based routing and ECMP load balancing. This enables routable VIPs for services of type LoadBalancer such as an ingress controller.
  • external-dns setup to publish ingress and service names to an authoritative DNS system.

A router VM separates the K8s worker segment from the KVM host and has the bits needed to peer with metallb and also runs a powerdns instance to work with external-dns.

All the guest VM images are unmodified cloud images from CentOS 7 and are configured via cloud init (by a config drive)

System requirements:

  • A KVM host with at least 8GB of RAM (16BG better)
    • Ansible >=2.7
    • Libvirtd with the virt-install and genisoimage utilities
  • 2 KVM networks defined to simulate a public and private segments (see sample XML files)
  • The guest OS cloud image under the libvirt image directory. I.e. CentOS-7-x86_64-GenericCloud.qcow2

Usage

See the defaults for the VMs defined under ansible/roles/kvm-infra/defaults/main.yml. This builds the dual-homed router VM and 3 more KVM guests to be used as a K8s controller (node1) and 2 worker nodes (nodes[2-3])

Make sure to update the SSH key in there as this is the only way to SSH in after deploying the playbook. Then build the infrastructure with:

 ansible-playbook configure-kvm-infra.yml -i inventory.ini

Once the virtual infrastructure is up, I suggest to deploy your K8s cluster easily with the excellent kubeadm-ansible project. A sample host.ini matching the nodes you have just setup would look like:

[master]
192.168.200.31

[node]
192.168.200.[32:33]

[kube-cluster:children]
master
node

Then install helm/tiller on your cluster, metallb and external-dns (sample provided should work unmodified).

helm install stable/metallb -n metallb --namespace=metallb-system -f k8s/metallb/values.yaml

See BGP daemon log on router VM has established peering with the 2 worker nodes

[root@routerng ~]# docker logs bgp
time="2019-01-03T14:21:14Z" level=info msg="gobgpd started"
time="2019-01-03T14:21:14Z" level=info msg="Finished reading the config file" Topic=Config
time="2019-01-03T14:21:14Z" level=info msg="PeerGroup k8s-workers is added"
time="2019-01-03T14:21:14Z" level=info msg="Add a peer group configuration" Name=k8s-workers Topic=Peer
time="2019-01-03T14:21:14Z" level=info msg="Dynamic Neighbor 192.168.200.0/24 is added to PeerGroup k8s-workers"
time="2019-01-03T17:53:19Z" level=info msg="Peer Up" Key=192.168.200.33 State=BGP_FSM_OPENCONFIRM Topic=Peer
time="2019-01-03T17:53:22Z" level=info msg="Peer Up" Key=192.168.200.32 State=BGP_FSM_OPENCONFIRM Topic=Peer

Create service LB type. Here as an example for the k8s dashboard

$ kubectl expose service  kubernetes-dashboard -n kube-system --name lb-dash --type=LoadBalancer --target-port=8443 --port=443
service/lb-dash exposed

Router has ECMP routing setup for the service LB IP

ip route
...
198.51.100.0 proto zebra
	nexthop via 192.168.200.33 dev eth0 weight 1
	nexthop via 192.168.200.32 dev eth0 weight 1

Now you can curl it

Note that if you install an ingress controller and wish external-dns to properly register its LB service IP for each ingress resource, it must be started with:

For nginx_ingress:

helm install stable/nginx-ingress -n nginx-ingress --namespace=ingress-system \
 --set rbac.create=true --set controller.publishService.enabled=true

or for traefik:

helm install stable/traefik -n traefik --namespace=traefik-system \
  --set rbac.enabled=true --set kubernetes.ingressEndpoint.useDefaultPublishedService=true