Recent Discussions
New into minecraft server hosting (Kubernetes, Multipaper, Velocity)
Unanswered
Ant posted this in #questions
51 messages
0 views
AntOP
Hi everyone!
I'm planning a larger Minecraft network project and want to build it the "modern way" from the ground up. I'm moving away from traditional setups and heading toward fully declarative, containerized infrastructure.
The Vision:
A "Global Lobby" leading to three distinct Survival worlds.
Each Survival world should be a MultiPaper cluster to allow horizontal scaling.
The Tech Stack:
OS: Talos OS (for immutability and security)
Orchestration: Kubernetes (Sidero/K3s style)
Storage: Longhorn for persistent data, Local Path Provisioner for worker performance
Proxy: Velocity with modern forwarding
What I'd love your input on:
Talos OS & Minecraft:
Has anyone here run Minecraft (specifically MultiPaper) on Talos?
Are there specific kernel tweaks or sysctl parameters you'd recommend in the MachineConfig for lower tick latency?
MultiPaper Scaling:
How does the MultiPaper master handle dynamic pod churn in a Kubernetes environment?
Are there best practices for HPA (Horizontal Pod Autoscaling) based on TPS or player count instead of just CPU usage?
Networking:
For a 16-node bare-metal cluster, which CNI would you recommend for the lowest possible overhead?
Is Flannel sufficient, or is Cilium worth the extra configuration effort at this scale?
I know this is a deep-end approach for a Minecraft network, but I want to build it as hands-off and resilient as possible.
Thanks for any insights, tips, or "don't do it this way" warnings!
I'm planning a larger Minecraft network project and want to build it the "modern way" from the ground up. I'm moving away from traditional setups and heading toward fully declarative, containerized infrastructure.
The Vision:
A "Global Lobby" leading to three distinct Survival worlds.
Each Survival world should be a MultiPaper cluster to allow horizontal scaling.
The Tech Stack:
OS: Talos OS (for immutability and security)
Orchestration: Kubernetes (Sidero/K3s style)
Storage: Longhorn for persistent data, Local Path Provisioner for worker performance
Proxy: Velocity with modern forwarding
What I'd love your input on:
Talos OS & Minecraft:
Has anyone here run Minecraft (specifically MultiPaper) on Talos?
Are there specific kernel tweaks or sysctl parameters you'd recommend in the MachineConfig for lower tick latency?
MultiPaper Scaling:
How does the MultiPaper master handle dynamic pod churn in a Kubernetes environment?
Are there best practices for HPA (Horizontal Pod Autoscaling) based on TPS or player count instead of just CPU usage?
Networking:
For a 16-node bare-metal cluster, which CNI would you recommend for the lowest possible overhead?
Is Flannel sufficient, or is Cilium worth the extra configuration effort at this scale?
I know this is a deep-end approach for a Minecraft network, but I want to build it as hands-off and resilient as possible.
Thanks for any insights, tips, or "don't do it this way" warnings!
Connecticut Warbler
would heavily recommend against multipaper
its not stable, not updated anymore, doesnt perform good, has dupes iirc
Connecticut Warbler
and its a "old" mc version not really what people want from survival
@Ruddy Ground-Dove is the kube guy here tho
@Ant Hi everyone!
I'm planning a larger Minecraft network project and want to build it the "modern way" from the ground up. I'm moving away from traditional setups and heading toward fully declarative, containerized infrastructure.
The Vision:
A "Global Lobby" leading to three distinct Survival worlds.
Each Survival world should be a MultiPaper cluster to allow horizontal scaling.
The Tech Stack:
OS: Talos OS (for immutability and security)
Orchestration: Kubernetes (Sidero/K3s style)
Storage: Longhorn for persistent data, Local Path Provisioner for worker performance
Proxy: Velocity with modern forwarding
What I'd love your input on:
Talos OS & Minecraft:
Has anyone here run Minecraft (specifically MultiPaper) on Talos?
Are there specific kernel tweaks or sysctl parameters you'd recommend in the MachineConfig for lower tick latency?
MultiPaper Scaling:
How does the MultiPaper master handle dynamic pod churn in a Kubernetes environment?
Are there best practices for HPA (Horizontal Pod Autoscaling) based on TPS or player count instead of just CPU usage?
Networking:
For a 16-node bare-metal cluster, which CNI would you recommend for the lowest possible overhead?
Is Flannel sufficient, or is Cilium worth the extra configuration effort at this scale?
I know this is a deep-end approach for a Minecraft network, but I want to build it as hands-off and resilient as possible.
Thanks for any insights, tips, or "don't do it this way" warnings!
I'm planning a larger Minecraft network project and want to build it the "modern way" from the ground up. I'm moving away from traditional setups and heading toward fully declarative, containerized infrastructure.
The Vision:
A "Global Lobby" leading to three distinct Survival worlds.
Each Survival world should be a MultiPaper cluster to allow horizontal scaling.
The Tech Stack:
OS: Talos OS (for immutability and security)
Orchestration: Kubernetes (Sidero/K3s style)
Storage: Longhorn for persistent data, Local Path Provisioner for worker performance
Proxy: Velocity with modern forwarding
What I'd love your input on:
Talos OS & Minecraft:
Has anyone here run Minecraft (specifically MultiPaper) on Talos?
Are there specific kernel tweaks or sysctl parameters you'd recommend in the MachineConfig for lower tick latency?
MultiPaper Scaling:
How does the MultiPaper master handle dynamic pod churn in a Kubernetes environment?
Are there best practices for HPA (Horizontal Pod Autoscaling) based on TPS or player count instead of just CPU usage?
Networking:
For a 16-node bare-metal cluster, which CNI would you recommend for the lowest possible overhead?
Is Flannel sufficient, or is Cilium worth the extra configuration effort at this scale?
I know this is a deep-end approach for a Minecraft network, but I want to build it as hands-off and resilient as possible.
Thanks for any insights, tips, or "don't do it this way" warnings!
Ruddy Ground-Dove
I've run Talos heavily and still do.
Personally I would use Cilium, if you're starting at Kube then Flannel likely will bottleneck you fairly quickly.
You shouldn't need to do much in MachineConfig to optimise for Minecraft, just make sure you have a decent CNI and CSI. Minecraft demands a lot of disk IOPS so be aware that you generally want something that can handle in the high tens of thousands continuously if you're doing several hundred players concurrent.
As for HPA? Generally don't hope for HPA in Minecraft, it's not a microservice or database. You'll gain more power by doing VPA with custom forks such as UniverseSpigot or whichever one is fastest at the time.
A suggestion I do have is embrace chaos engineering and Git drive everything with ArgoCD and the AoA pattern, when you learn how you want to be able to do blue/green deployments and fully automated testing.
One thing I have is a system where 100% of my Talos cluster is defined purely by files so I can spin a copy up in minutes and test everything to make sure it works with clever CI/CD use, this has cut down on failures multiple times since I can test the entire cluster at scale of I need to.
Personally I would use Cilium, if you're starting at Kube then Flannel likely will bottleneck you fairly quickly.
You shouldn't need to do much in MachineConfig to optimise for Minecraft, just make sure you have a decent CNI and CSI. Minecraft demands a lot of disk IOPS so be aware that you generally want something that can handle in the high tens of thousands continuously if you're doing several hundred players concurrent.
As for HPA? Generally don't hope for HPA in Minecraft, it's not a microservice or database. You'll gain more power by doing VPA with custom forks such as UniverseSpigot or whichever one is fastest at the time.
A suggestion I do have is embrace chaos engineering and Git drive everything with ArgoCD and the AoA pattern, when you learn how you want to be able to do blue/green deployments and fully automated testing.
One thing I have is a system where 100% of my Talos cluster is defined purely by files so I can spin a copy up in minutes and test everything to make sure it works with clever CI/CD use, this has cut down on failures multiple times since I can test the entire cluster at scale of I need to.
@Connecticut Warbler and its a "old" mc version not really what people want from survival
AntOP
So Folia or something similar + maybe sharding would be the better choice?
@Ant So Folia or something similar + maybe sharding would be the better choice?
Key West Quail-Dove
I believe that's what donutsmp does
AntOP
But does it work good? I want to create a kind of Freebuild Economy and i could imagine that building near or even on a shard border is a problem
Key West Quail-Dove
No it's not seamless, I haven't seen any decent seamless sharding implementations yet
Folia gets you most of the way though I think
@Ruddy Ground-Dove I've run Talos heavily and still do.
Personally I would use Cilium, if you're starting at Kube then Flannel likely will bottleneck you fairly quickly.
You shouldn't need to do much in MachineConfig to optimise for Minecraft, just make sure you have a decent CNI and CSI. Minecraft demands a lot of disk IOPS so be aware that you generally want something that can handle in the high tens of thousands continuously if you're doing several hundred players concurrent.
As for HPA? Generally don't hope for HPA in Minecraft, it's not a microservice or database. You'll gain more power by doing VPA with custom forks such as UniverseSpigot or whichever one is fastest at the time.
A suggestion I do have is embrace chaos engineering and Git drive everything with ArgoCD and the AoA pattern, when you learn how you want to be able to do blue/green deployments and fully automated testing.
One thing I have is a system where 100% of my Talos cluster is defined purely by files so I can spin a copy up in minutes and test everything to make sure it works with clever CI/CD use, this has cut down on failures multiple times since I can test the entire cluster at scale of I need to.
Personally I would use Cilium, if you're starting at Kube then Flannel likely will bottleneck you fairly quickly.
You shouldn't need to do much in MachineConfig to optimise for Minecraft, just make sure you have a decent CNI and CSI. Minecraft demands a lot of disk IOPS so be aware that you generally want something that can handle in the high tens of thousands continuously if you're doing several hundred players concurrent.
As for HPA? Generally don't hope for HPA in Minecraft, it's not a microservice or database. You'll gain more power by doing VPA with custom forks such as UniverseSpigot or whichever one is fastest at the time.
A suggestion I do have is embrace chaos engineering and Git drive everything with ArgoCD and the AoA pattern, when you learn how you want to be able to do blue/green deployments and fully automated testing.
One thing I have is a system where 100% of my Talos cluster is defined purely by files so I can spin a copy up in minutes and test everything to make sure it works with clever CI/CD use, this has cut down on failures multiple times since I can test the entire cluster at scale of I need to.
AntOP
Thanks for this response, i appreciate it. How is it with talos in a production environment? Do i need to pay ?
HPA is than maybe the best way for my website?
HPA is than maybe the best way for my website?
@Ant Thanks for this response, i appreciate it. How is it with talos in a production environment? Do i need to pay ?
HPA is than maybe the best way for my website?
HPA is than maybe the best way for my website?
Ruddy Ground-Dove
Talos is extremely production stable and it's also free. Where Sidero makes their money is with Omni which is paid if you go into high tiers. It's well worth paying for though since it makes managing the cluster simple if you're still new.
Ultimately if you're talented you can just declaratively drive everything with Terraform or whatever, it depends on what skill you have.
Ultimately if you're talented you can just declaratively drive everything with Terraform or whatever, it depends on what skill you have.
@Ruddy Ground-Dove Talos is extremely production stable and it's also free. Where Sidero makes their money is with Omni which is paid if you go into high tiers. It's well worth paying for though since it makes managing the cluster simple if you're still new.
Ultimately if you're talented you can just declaratively drive everything with Terraform or whatever, it depends on what skill you have.
Ultimately if you're talented you can just declaratively drive everything with Terraform or whatever, it depends on what skill you have.
AntOP
Can I expect any advantages if I use Kubernetes instead of, for example, Proxmox?
@Ant Can I expect any advantages if I use Kubernetes instead of, for example, Proxmox?
Ruddy Ground-Dove
Proxmox is just a hypervisor, it's a different class of OS that tries to do different things.
The amount of advantages Talos has over Proxmox is honestly insane, it's just harder to learn and use.
@Key West Quail-Dove Folia gets you most of the way though I think
AntOP
My concern is that there will be performance issues; players are supposed to be able to found cities and build shops (which could bring many players into a shard).
@Ruddy Ground-Dove The amount of advantages Talos has over Proxmox is honestly insane, it's just harder to learn and use.
AntOP
As long as the advantages for server network operation are worthwhile, it is also worth embarking on the steep learning curve.
@Ant As long as the advantages for server network operation are worthwhile, it is also worth embarking on the steep learning curve.
Ruddy Ground-Dove
I'll put it this way, Kubernetes on any platform is going to outcompete traditional servers to an obscene degree.
It's the most advanced method humans have to orchestrate servers, exactly nothing keeps up.
It's a paradigm shift and if you want to learn how it works then it will reward you.
It's the most advanced method humans have to orchestrate servers, exactly nothing keeps up.
It's a paradigm shift and if you want to learn how it works then it will reward you.
There are not many server software types that don't scale better on Kube, that's just how it works.
@Ruddy Ground-Dove I'll put it this way, Kubernetes on any platform is going to outcompete traditional servers to an obscene degree.
It's the most advanced method humans have to orchestrate servers, exactly nothing keeps up.
It's a paradigm shift and if you want to learn how it works then it will reward you.
It's the most advanced method humans have to orchestrate servers, exactly nothing keeps up.
It's a paradigm shift and if you want to learn how it works then it will reward you.
AntOP
Can I send you a DM if I need help with Talos/Kubernetes in the future?
Loading...
Loading...