4 ms·

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

25 points by saurajg 4y ago

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.

metadat 4y ago
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.
- Havoc 4y ago
  There is a pricing page in the menu
- quickthrower2 4y ago
  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.
- styren 4y ago
  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.
Havoc 4y ago
That looks good. Congrats. The smallest preset cluster looks to be two nodes. I thought 3 is minimum for consensus reasons?
- metadat 4y ago
  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.
  - GauntletWizard 4y ago
    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.
- styren 4y ago
  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.
  - Havoc 4y ago
    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
quickthrower2 4y ago
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
- KomoD 4y ago
  It's Contabo and/or DigitalOcean
  - styren 4y ago
    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.
- saurajg 4y ago
  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.
pomatic 4y ago
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/...
- oriettaxx 4y ago
  wow, that would be great!

25 points by saurajg 4y ago

metadat 4y ago

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.

Havoc 4y ago
There is a pricing page in the menu
quickthrower2 4y ago
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.
styren 4y ago
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.

Havoc 4y ago

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

metadat 4y ago
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.
- GauntletWizard 4y ago
  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.
styren 4y ago
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.
- Havoc 4y ago
  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

quickthrower2 4y ago

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

KomoD 4y ago
It's Contabo and/or DigitalOcean
- styren 4y ago
  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.
saurajg 4y ago
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.

pomatic 4y ago

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/...

oriettaxx 4y ago
wow, that would be great!