Show HN: We built a managed Kubernetes provider at half the cost of AWS | Hacker News Reader

Show HN: We built a managed Kubernetes provider at half the cost of AWS

symbiosis.host

25 pointssaurajg15 comments

Since there’s a few discussions recently on $4 VPSes we thought you might be interested in Symbiosis, a cheaper and simpler managed k8s provider.

Supported features include external LBs, encrypted+replicated storage and autoscaler. Infrastructure can be managed with UI, CLI, Terraform or Pulumi.

Control planes are free of charge and a 1vCPU/2GB node goes for $6 per month. We’re live in Germany right now and we’re expanding into US during 2023.

Feel free to try it out and leave a comment.

Open articleView on HN

I'm a little confused by the pricing claims. They provide only one data point, a 24 CPU / 48 GB cluster. If your project can run in such a small amount of resources, is accumulating the complexity tax which comes with Kubernetes distributed system going to be a win? Would be significantly more helpful to show the price points at various compute scales.

I just looked up the cost of Oracle Kubernetes Engine on OCI (Oracle Cloud) and it's provided for free, you only pay for the compute resources you use.

On AWS you pay something like $0.10/hr/node.

Also, where do these machines run? What regions are available? What IAM controls and security policies are available? What is the disaster recovery story? For most real businesses, controlling and guaranteeing these aspects and risk factors quickly become critical. Far more important than saving a couple of hundred bucks or even a grand a month.

I understand.. it's sad to realize these things, because a better Heroku PaaS is highly appealing to the developer in me.

Perhaps there's a market segment I'm unfamiliar with which is craving this product layout. I wish you all the best.

There is a pricing page in the menu

Depends how managed it is. I feel like a cloud managed kubernetes can be simple enough to run, if you already have the experience. I dare say on par with the complexity of running a VM yourself.

We used that particular comparison of 24 cpus/48GB partly because it was a benchmark used by another competitor and partly because we felt it was in line with expected usage for a small/medium size business.

We're hosting with Hetzner and we have our one and only region in Germany right now, we'll be opening more regions during 2023. Access is granted to users on a per-cluster basis (we're planning to launch finer grained permissions), with clusters having RBAC.

Perhaps you're right in that most developers are looking for a better Heroku, and looking at our current users many of them have infra or devops background. I've used k8s in most of my recent (small & big) projects due to the fact that it's got such a large ecoystem of tools and packages I can readily deploy to it. It's perhaps not the same use-case as most of the users of Heroku or Render, but that's not necessarily the type of use-case that we're hoping to solve.

I think the smart move here would be to sell the platform, not the service. i.e. allow end-users to deploy your orchestration tools to their own Hetzner tin, and pay you for support. With source escrow or similar, that sidesteps the questions around what happens if symbiosis inc. disappear/go down the tubes/...

wow, that would be great!

Can't see anything about whether you are reselling someone else's servers (e.g. Hertzner) or running these machines yourself.

Note: One region so far: germany-1 Falkenstein, Germany

It's Contabo and/or DigitalOcean

Our status page/status monitoring service is running on Contabo since we want it off-site. We have an API gateway running on DigitalOcean that we're currently moving away from.

Our servers (in our one and only current region) are hosted with Hetzner.

We're running our service on top of Hetzner's root servers. Opening a second region in Europe soon and plans for the US later this year.

That looks good. Congrats.

The smallest preset cluster looks to be two nodes. I thought 3 is minimum for consensus reasons?

You can do it with two, but availability will be lower than even a single node because if either one disconnects the cluster enters a split-brain state and cannot safely mutate / update anything.

Their control planes are hosted, so that's not an issue. You need three nodes for the control plane's master election, but for failover app routing, two is fine.

Thanks!

The worker nodes don't need quorum, they are merely VMs that are running a daemon capable of scheduling k8s workloads onto it. The control planes can be configured for high availability (HA), resulting in 3 control planes.

ah gotcha. I'm used to running all on all at home (3x rasps) so didn't occur to me that control plane is what matters

Reply on news.ycombinator.com