Bluefin Server Alpha 2 | Bluefin

[giscus[bot]]

Bluefin Server Alpha 2 | Bluefin

Alright, here it is, Bluefin Server, this one actually boots. This is the base of the Bluefin Factory, the thing you'll build your homelab on. And in turn, your desktops or whatever you want. Kubernetes for the home.

https://docs.projectbluefin.io/blog/bluefin-server/

[fumosus]

I am so here for this! Thank you for driving me to learn about this stuff. It's so rad dude!

[castrojo]

Anyone got bare metal to spare? I made some unfortunate mistakes at MicroCenter and I got some new kit. 😄 But would love to get more feedback on this!

Here's an idea for kernels, would love feedback.

Since this is a DDI system we can swap kernels. My plan is to do a bst from source build of flatcar's kernels and patches for our LTS kernel. This would "lock" us to the same version they use, but built from scratch, and would then let us use Flatcar sysexts. So this means we could support ZFS as a toggle, no different image/kernels needed. This would be huge for us!

And then a source build of the Fedora CoreOS kernel and it's patches for our newer kernel. This pair would be the defaults, "a love letter to the CoreOS family", which is awesome given the history of both projects!

It would also not be hard to do an Ubuntu kernel, etc. This appeals to me because ideally I want a pure upstream kernel as the default. And people are going to ask for weird wifi modules, etc. It's easier to just use the ubuntu stuff for that and keep our default kernel the same as fedora coreOS, which has TONS of users. The Fedora version the Cloud Native Factory uses this as the default OS it's all common ground.

I also want to ensure we're pushing the latest cloud native tech in here. Unraid just made a very important decision. All of this stuff is going to make their users happy. People are going to keep talking about distros: "coreos, ucore, bluefin server, flatcar, what a mess!" but the reality is when you say "I want immich deployed and import my photos" it's the exact same on all of them.

We also reserved the sauropods for the server product. Most people don't know that plant eating animals kill more humans per year than predators. Water Buffalos lol, no thank you! These represent the sheer massive force that Kubernetes and cloud native represent. It's a different form of "power" than a T-Rex. This is life on the most massive scale our planet has ever seen. Ironically so we can burn it all down, but that's a different post lol.

I've almost got it booting into this but the idea is:

• install, reboot
• firstboot setup, then the banner shows you the weburl
• go to weburl, install this relay thing via brew on your client since you're doing cluster management.
• create local account, associate google, github, etc if you want.
• load kubestellar's dashboard.
• go to zfs section, make your array, load in the container section and start deploying jellyfin and your entire self hosted setup.
• turn on your bootc factory, pick bluefin as your base in the ux, clickety click packages/branding for your custom chicken.

Get 2nd machine, repeat installation

• This time at the kubestellar step you get a banner:
• "New node detected, join cluster?"
• You say yes, do the auth dance.
• Your workloads start to even out over your double servers, for important ones that you designate they will run 2 copies, one on each node, behind a load balancer.
• You've amped up your digital sovereignty score by deploying more and more of your services yourself. The local model on your lab at night has been working on a plan to migrate your stuff of.

Get 3nd machine, repeat installation

• Now the lab is shredding, the USB-4 daisy chain of three Strix Halo machines is at full capacity. All the bst builds for your entire digital life are being built across three machines. You are the new Gentoo god. Remember there is 0 network latency, those nvme drives, RAM, and CPU/APU are being shared unlike anything I've been able to pull off with home gear.

You now have the dream cluster. Any one can go down, you can keep one control node or split among three. If you need to move one you drain the node with one command, turn it off, and then do what you need, and then turn it back on. All of your machines push ffmpeg transcoding to this cluster, it's all distributed. Idle macs doing nothing with that sweet encoding hw? Easy to join to the cluster.

With this setup local models running your home life is doable today. k8sgpt keeps your cluster healthy, everything is on network. homeassistant becomes much easier - you direct consume their container.

Maintenance is automated, when the OS needs to be updated the individual nodes will take turns rebooting, each node draining, cordoning itself off, then rebooting, rejoining the cluster and getting the workloads back. You lose track of system uptime, only your service uptime matters. When there's a jellyfin update it will do so in turn behind the load balancer, so if an update bricks the thing it never gets served to you.

I am doing all of this today and I hope this becomes the new baseline for home servers (which is just using k8s lol)! Ideas welcome!

Major Marketing Message: It's modern Linux. Enterprise and Home are the same thing. That divide is now over thanks to k8s and agents. The best part is all this exists already, this isn't a "sell the idea" problem it's "let's assemble it" problem.

[fumosus]

I have have several business PCs sitting idle but they are quite old now (circa '16 I think). Do you have a minimum vintage you're aiming for?

[castrojo]

Should be enough! If they boot it should work! I am setting up my bare metal install today so if you see errors report them please!

[ericrocha97]

Congratulations to the whole team, this looks like an awesome project! I’m really excited to see where Bluefin Server goes and I’m looking forward to using it on my home server once it’s ready. Great work!

[IvanVeloz]

I have a trio of Haswell machines that I was looking to turn into a cluster with CoreOS. Now I'm planning to try this instead. I'll keep you posted.

I also have a Super6C Pi cluster. But those Pis are only 8GB eMMC and 2GB RAM. I have a bunch of 16GB Optane M.2s I could use. Pis can have an UEFI installed on them.