Light weight Kubernetes k3s.
Shared block storage for the db.
Each “service” (farmer, harvester, wallet, Ect) running in docker in a kube cluster of say 5 physical nodes.
High availability chia? Ingress?
Run a docker container for harvester aswell in kubernetes.
So that got me thinking. What if u ran mad max on thies 5 nodes, the kluster. Would the task just be assigned to one worker or would they share resources like a normal cluster….
High availability plotter?!
Then what happens if u scale that out. 50 plotters. As long as u had enough ssd space for plotting, this is conceivable. And highly available.
Wich I think would be really good for a chia farm. 0 downtime. Persistent configs…. O yeah
could be fun.
Anyone with knowledge in the marrwr?
Chia is already a pretty durable service: spreading it out over 5 systems sounds fun to set up… but I don’t think it will gain you much. Distributed plotting is also cool, but since you can just buy it cheap… it’s also probably something you set up for fun.
Makes 0 sense if you ask me. Kubernetes also just runs on a server with x cores, y GB RAM, z TB SSD space and usually the provider is gonna charge you for the resources you use… Also, normally you willl have a resource limit set by the admin of the cluster…
Only think that makes sense would be running a node + farmer, so that it is 24/7 online. Then let your local harvesters connect to that node.
I Self host a cluster of 5 physical arm machines. And have combined thier power together with kubernetes
The hosting with cloud provider is a good idea tho. For a simple farmer
I can pass any docker container to kubernetes. And it will distribute the work among my 5 machines.
So what if I gave my cluster a mad max docker image, is what I’m wondering
I’m asking questions before attempting anything myself.
I just have no idea what im doing with kubernettees
If you want to do it for fun sure… But K8s isnt really meant for this kinda of thing, it is meant for 12-factor app instances which Chia is not. You can definitely force it into a docker container and get the results you want though. Personally after many years of Ops life I find it is better to go with the simplest solution possible. Adding K8s layer is just adding more complexity while adding little value.
simple is truly the best now with chia. i handle my instillation delicately. more functional. reliable. iv done everything i could to scale back to a single chia node. even tho harvesters supposedly make your farm faster. at this point im just investing in higher capacity drives over quantity. less machines to manage.
in my testing i come to find i really dont like kuburenetees lol. tricky to get going. but i did get a plot created. hours n hours later. from my lil arm cluster. but its just silly. hard to keep everthing straight on the line.
using docker containers inside of physical k3s hosts. communicating and managiing both clusters from portainers gui. thats hosted in the cluster itself, obviously
say that 5 times
but alas it was only a test
my real intention behind proving this possible. was to unite computers across networks to work for a single purpose… plot creation. or just running a instance of chia itself from the cli- collectivly with failover. and HA… developing that… and studying the implications of a sort of chia virus… . . . … . that would infect a chia node… with another instance… while blocking out the other instance… giving your instance to everyone… and blocking out nodes… leaving your node to win everyhting… and using docker inside of k3s harvesters… to accomplish this… both collectively bot netting… to run chia… and slowly create and move around plots across its network of chia dead hosts…swapping out your plots storage for its plots…
i think its totally doable… right??
or something. i dunno tho i get distracted alot.
+-chia dev team git at me bru +_–+=–
weakness out here
Haha sure sounds fun, and you sound like a hobbyist like me. What a time to be alive. Personally I just run a newer intel NUC with USB3 hard drive encloser for my farmer/harvester. I am a small fish compared to most on here ~200TB. NUC is running Fedora 37/python3.10 venv. Upgrades are pretty straight forward, stop farmer, pip install chia-blockchain==1.6.2 into the venv, start farmer. For plotting im using an old 2xXeon with 24C/48T ~160GB Memory and Mad Max Plotter of course. Transferring 100GB files over the network is slow, so I just move the physical drive over to the hard drive enclosure. All my bays are full now and Im seriously considering getting another used server with like 32 bays and expanding. Waiting to see what Plot 2.0/GPU farming is about first. I dont mind the HW costs as I use the rigs for other things besides Chia so not too concerned with a quick return. K8s is cool and everything but seems overkill/complex for Chia.