mirror of
https://github.com/optim-enterprises-bv/terraform-talos.git
synced 2025-10-29 17:42:47 +00:00
Terraform examples to launch Talos.
I store here the terraform code to launch Talos in the clouds. I wouldn't use the terrafrom modules from internet. The goal is to create all cloud services from scratch.
Ideas
First, I will create separate clusters on each cloud provider, test them thoroughly, and bring them close to production readiness. When I merge these separate Kubernetes clusters into one, they will have a single control plane.
Why is it so important?
Having a single Kubernetes control plane that spans multiple cloud providers can offer several benefits:
- Improved resilience and availability: By using multiple cloud providers, you can reduce the risk of downtime due to a single point of failure.
- Flexibility: A single control plane allows you to easily move workloads between different cloud providers, depending on your needs.
- Cost savings: You can take advantage of the different pricing models and discounts offered by different cloud providers to save on costs.
- Improved security: By using multiple cloud providers, you can implement a defense-in-depth strategy to protect your data and reduce the risk of a security breach.
- Decrease the time to recovery (TTR)
Clouds
| Platform | Checked Talos version | Addons | Setup type | Nat |
|---|---|---|---|---|
| Azure | 1.3.0 | CCM,CSI,Autoscaler | many regions, many zones | ✓ |
| Exoscale | 1.3.0 | CCM,Autoscaler | many regions | ✗ |
| GCP | 1.3.0 | CCM,CSI,Autoscaler | one region, many zones | ✓ |
| Hetzner | 1.3.0 | CCM,CSI,Autoscaler | many regions | ✗ |
| Openstack | 1.3.0 | CCM,CSI | many regions, many zones | ✓ |
| Oracle | 1.3.0 | CCM, |
one region, many zones | ✓ |
| Scaleway | 1.3.0 | CCM,CSI | one region | ✓ |
Common
- cilium network with vxlan tunnels.
- ingress-nginx (daemonsets) runs on
webrole nodes. It useshostNetworkports 80,443 for optimizations. It helps me to tweak the kernel on a host and apply it to ingress controller. And I can disable conntrack too. - coredns-local (daemonsets) uses dummy interface on al nodes and has ip
169.254.2.53It decrease the dns response (all traffic does not leave the node). It makes sense in multi-cloud setup. Kubernets still does not have geo-based load balancer capabilities (alfa). - rancher.io/local-path as default storage class.