I just published a new blog post about hardware choices for home-brew servers:
@forestjohnson wow, the synchronicity of this post lol, I spent last weekend researching used dell optiplexes
I have this weird drive to build a what im calling a "junkyard kubernetes cluster". But currently don't have the storage or electrical capacity for multiple sff systems that need cooling in the summer.
If I had an unfinished basement the cold concrete would probably be enough 😂
That sounds intriguing, are you set on Kubernetes? Do you want to just do it for fun within your own home, or are you wanting to try to actually use the cluster capability for replication+failover in the context of homebrew servers?
@forestjohnson yes the only reason I can see for myself to run k8s in any capacity is to take advantage of heterogeneous compute infrastructure in a cluster-able way (if that makes any sense), and what better way to get heterogeneous compute than from electronics recyclers lol!
@fack Well I just asked on the Free Geek discord today and they said they do sometimes have those thin clients there.
But I expect ebay would still be worth it. Free Geeks said the most of the thin clients even support SATA, just only in the SOM form factor. So no HDDs, you have to buy an SSD and then take the tin cover off of it for it to fit. Like this:
@forestjohnson yuuup nice nice. It would be so nifty to just like resurrect a bunch of junk into a cluster and as shit literally bites the dust it's finally being used up to it's full potential, while you still maintain some level of fault tolerance and whatnot.
Literally seems like the main reason for a hobbyist to have any interest at all in k8s
@fack Yeah, true, I do think it's not gonna help much unless you have tons of apps tho.
But as a way to learn how to use kubernetes it would be great!! I just only have eyes for a multi-locality cluster these days because it would theoretically make the homebrew apps exponentially more reliable. I'm pretty sure Kubernetes cant help with that
Therin lies the rub:
AFAIK there is no standard solution for this right now. So I was gonna just try to make one myself 😅 Greenhouse was just the beginning of that
The biggest part of it is how to handle replication and fail-over without requiring apps to be reconfigured. VM systems can do that pretty well as I understand it, I was thinking about trying to make my own system that is more oriented around Linux containers. I made a diagram about it:
https://picopublish.sequentialread.com/files/one-size-fits-all-replication-failover.drawio.png
@forestjohnson what exists for multi-locality out of curiosity?
I know you can federate k8s clusters, so you can have per-locality-clusters, and iirc then you can wire them up to something akin to a meta-control-plane.
It's possible but complicated I'd say.