Build a Lightweight Private Cloud with Harvester, K3s, and Traefik // Traefik Labs
HTML-код
- Опубликовано: 29 июн 2022
- Although we’re in a cloud native era, virtualization is still present in most organizations and isn’t going anywhere. The truth is that VMs and containers will coexist in most enterprise infrastructures indefinitely.
At SUSE, we’re bringing virtual machines and containers even closer together with Harvester. Harvester is a modern open source and lightweight HCI (Hyperconverged Infrastructure) solution based on Kubernetes, Longhorn, and KubeVirt, providing a platform that creates a solid foundation for deploying clusters on-prem. Harvester helps organizations modernize on-prem infrastructure or edge computing by combining containers and VMs.
In this session, we introduce Harvester and demonstrate how to deploy a K3s cluster on Harvester using its integration with Rancher. We also deploy a sample service using Traefik Proxy as the ingress controller.
You will learn:
- What Harvester is and how HCI solutions work
- The underlying technologies (SUSE Linux, RKE2, Longhorn, and KubeVirt)
- The main use cases and advantages (On-prem and Edge HCI, Hybrid Cloud, Harvester & Rancher Integration)
- How to deploy K3s in Harvester using Rancher along with Traefik Proxy
Speaker
Andrés Valero, Technical Marketing Manager, SUSE
ABOUT TRAEFIK LABS
Traefik Labs is the creator of the simplest, most comprehensive cloud-native stack to help enterprises manage their entire network across data centers, on-premises servers and public clouds all the way out to the edge. Traefik Proxy, it's flagship open source product, is trusted in projects big and small and has 2B+ downloads, 30K+ stars on GitHub, and 500+ contributors.
Learn more about Traefik Labs here: traefik.io/ 
Наука
Very nice demo. Provides a great overview of the product.
I'm not sure I understand the benefit over running RKE2 on baremetal + LongHorn + Rancher + KubeVirt. Is it the networking support, and is that going to help that much over CNI networking policies? Is it being able to put things in distributed datacenters on the same VLan on different clusters and have them act as the same network? Or what am I missing?
16 cores for all or for each installation? If I have blade system with 2 x e5-2609 xeons on each of 16 blades?
16 cores would be the minimum for an installation. You can deploy a two-node installation with 16 cores for each node or two separate deployments with 16 cores and a single node. Whatever fits better your needs.
Any reason Rancher and Harvester arent combined into one product? It seems like the obvious solution. VMs and containers in the same place... Harvester uses k8s under the hood anyway so, i dont really understand why its done this way.
What if i'm starting with one physical machine and i dont want to do unecessary nested virtualization?
Starting with nested virtualization makes it easier for you you to scale in the future if you need to.
Harvester has a UI based on Rancher, so you can use the UI embedded in Harvester without Rancher. However, suppose you are using Rancher and want to manage multiple clusters (Harvester, Kubernetes, or both). In that case, you probably already have a management cluster with software running and where you probably want to have Rancher installed. That management cluster can be inside Harvester in a VM or in any K8s cluster you already have running. It is a matter of offering choices and flexibility.
Harvester being released by the Rancher dudes (seems an appropriate designation for something to do with a ranch) was really exciting for me. Finally a proper solution for hosting Rancher. Sure, some VMs might be nice someday, but the promise of a one stop shop where I could install harvester and spin up Rancher instances using Longhorn as a common underlying storage foundation. What could be more perfect?
What we go instead was yet another VM solution which, let's be honest, we all have VM solutions, we certainly don't need yet another one. There are two established ways of managing VMs, one is to waterfall everything for people who just love doing everything the hard way. The other is to use Terraform/Ansible/Powershell DSC. Adding K8s/KubeVirt really just doesn't offer any real advantages. In 2020-2023 we don't replace one VM hypervisor with another, we redesign our apps to run on containers. After all, if we need to redesign our apps and management scripts for a new platform, why invest the time and money doing it. And frankly, Rancher's support agreements are expensive enough that if you go that route, why not just stay on the existing VM platform and launch a K8s environment elsewhere.
Harvester has one single purpose in existence. It's to provide a single pane of glass for managing containers and the infrastructure beneath them. When someone installs Harvester, it should provide a simple one step process of spinning up a complete Rancher cluster preconfigured to use the underlying Longhorn instance. Then there should be a single Rancher UI for managing not just the harvester, but using RBAC, it should allow management of all systems involved. For disaster recovery, there should be a clear plan for backing up and restoring the full Longhorn storage.
I think that we're at the point where unless the folks at Rancher can't start delivering a platform which integrates their own products in a meaningful fashion, there is simply no value to running their platform and I'd probably be just as well off running Proxmox and/or OpenShift. They're both developed in a way that appears to have a clear path for what they are meant to accomplish.
Honestly, I think Harvester was a truly amazing idea. I even went to the bosses to make them empty their pockets for buying official support for a harvester/rancher cluster, but after extensive testing, I've come to the realization that Harvester and Rancher are being developed to provide some sort of solution which I can't imagine what it is. There is only a single thing which Harvester should be absolutely perfect at doing and that's hosting Rancher. I should be able to next/next/next/finish building a harvester cluster, then be able to deploy a full resilient Rancher RKE cluster with either one kubectl command or one click and it should use the Harvester that's there already for storage. That's all.
Maybe they're hoping people will spend millions moving away from other VM solutions and use theirs instead and they're just giving up on containers?
Min. 16 core is a Lightweight system? I am using Proxmox with 4 core CPU. I have 10 LXC containers running in Proxmox. No problems at all. Harvester uses 70% of CPU, with no VM or container.
Can you "Build a Lightweight Private Cloud with Harvester, K3s, and Traefik" without Rancher ???
Sure you can. It depends on your needs. If you don't need a Virtualization layer, you can use K3s, Traefik, and Rancher.