Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Providers for cheaply running a HA Kubernetes cluster
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Providers for cheaply running a HA Kubernetes cluster

fartfart Member
edited January 2023 in General

I've been thinking of learning Kubernetes and perhaps the best way to learn about something is to run some real-life workloads, I'm looking for a cheap provider that:

  1. Preferably has an internal network, otherwise you need to set up tricky firewall rules allowing internal traffic between the nodes but not others.
  2. Won't suspend me for the continuous activity on the cores. Since the idea behind setting up a cluster is to consolidate workloads, I expect the CPU to not be completely idle, though I have no real world numbers.
  3. Somehow run 2 masters and 2 workers at < $25/m, with 2 cores and >4GB RAM on the workers. I can commit annually.

I know the pricing I'm asking for is super tight, probably more so given that I'm looking at dedicated vCPU offerings? But, given that I don't come from a particularly well off country, that figure is around the ballpark of what I can do.

For providers who may like to offer something on this thread, I should remind that I may not buy immediately (or at all).


  • vsys_hostvsys_host Member, Patron Provider

    Such a solution, as a rule, is paid only for the amount of the resource used. You can try our Kubernetes -

    Thanked by 1treesmokah
  • AC_FanAC_Fan Member
    edited January 2023

    GreenCloudVPS Budget Sale - 4GB@25, 8GB@45 per year.
    Advin Servers - Multiple options at 16GB@8 and 32GB@16 per month.
    VirMach Special Offers - 4GB@27 per year.

    You might need nested virtualisation on Advin Servers, which they'll probably provide. All are decent, but probably don't have internal network. You need to worry about CPU more than RAM, as K8s tends to hog CPU, often without reason.

  • what about running microk8s? paying 'a guru cloud' account for their lab environments would be more cost effective too.

  • Hi, I might be biased but I recommend Hetzner Cloud with my tool

    It's the quickest and easiest way to create fully functional clusters in Hetzner Cloud. The tool provisions the servers, the private network (so all traffic between the nodes is private), the firewall (you can restrict access to both the Kubernetes API server and to the nodes via SSH to specific IP ranges) and a load balancer if you create a cluster with highly available control plane (multiple masters).

    Once the infrastructure is provisioned, the tool deploys Kubernetes (k3s specifically) as well as

    • Cloud Controller Manager, so you can provision load balancers for your apps right away
    • CSI driver, so you can provision persistent volumes out of the box
    • Cluster autoscaler - yes! it even supports autoscaling, like managed Kubernetes services
    • System Upgrade Controller, to make upgrading to a new version ridiculously easy and quick

    You can also create multiple node pools in multiple locations for higher availability.

    Give it a try. I am always available on Github if you run into any issue. As you may have guessed I am the author :)

    Thanked by 1treesmokah
  • BTW if you use my referral link we both get credits. Win-win :D

    Thanked by 1treesmokah
  • @vsys_host said:
    Such a solution, as a rule, is paid only for the amount of the resource used. You can try our Kubernetes -

    I really like the page design with the cow and cat images :)

  • @vsys_host said:
    Such a solution, as a rule, is paid only for the amount of the resource used. You can try our Kubernetes -

    vouch, based provider for any task(that is legal in Ukraine).

  • Just my comments!

    Let me add on, picking up k8s (higher abstraction layer - or new IaaS) is not going to be easy. Having a fast environment is critical too.

    etcd are going to attempt to do sync operations rapidly, and is highly resource intensive. Imagine each worker node with 2vCPU & 2GB memory & 20GB storage. Multiply by 3. And if you are managing the control plane (CP) that's another 3 more CP nodes.

    And if you want to run enterprise alike, that's another set of 3 CP + 3 Worker Nodes.

    Running actual workloads to learn k8s is not necessary, but if you're still keen on going down this rabbit hole, it's going to cost you money.

Sign In or Register to comment.