This ansible repository provides a quick way to automatically deploy a Kubernetes Cluster inspired by the "K8s the hard way" method.
The resulting cluster is meant to be ready to try different networking solutions (and much more, of course!) without layers of abstraction that prevent to understand exactly what's happening under the hood (such as kubeadm which makes everything very too easy to deploy).
We all know about most of the existing networking plugins (Flannel, Calico, Kube-router, etc..) but this is a hugely moving ecosystem and having a "ready-to-bench" cluster looks pretty convenient.
I am going to re-use the network topology deployed with the Fully-Automated-BGP-fabric to focus here on the K8s cluster.
You can optimize everything but if you need a quick test lab, you can deploy with a one-liner on a free blank Cumulus In The Cloud workbench :
wget https://raw.githubusercontent.com/jpmondet/Fully-Automated-K8s-deployment/master/deploy_all.sh ; chmod u+x deploy_all.sh ; bash deploy_all.sh
Some customization can be done by changeing values in the global_vars.yaml file but the defaults one-liner above gives you :
- Kubernetes 1.19.1
- containerd 1.4.0
- runc 1.0.0-rc92
- etcd 3.4.13
- Standard CNI type Bridge (cni-plugin v0.8.7)
- IPVS replaces IPtables for Services (Not really an 'innovation' anymore :-) )
- CoreDNS replaces Kubedns (1.13 made CoreDNS the default so it's not an 'innovation' anymore but well ;-) )
- A pure RFC5549-based L3 IPv6 network (which lets you use IPv4 above v6)
- Kubefw to leverage Network Policies