The Homelab Argument Nobody Wins

Someone posted their homelab setup this week: a repo full of Kubernetes config, KEDA doing the heavy lifting, nine self-hosted apps scaling to zero overnight and waking on the first request. Neat idea. The comments turned it into a small war about whether any of it was worth doing, which is the most homelab thing that could possibly happen.

I’ve got a much less impressive setup at home. A Raspberry Pi doing Pi-hole duty, a NAS that mostly holds photos of my daughter as a toddler that I’ll never look at again, and a habit of reading about people’s elaborate Kubernetes clusters instead of building one myself. So I read this thread the way you’d watch someone rebuild a car engine in their driveway: genuinely interested, no intention of trying it myself, quietly judging the bloke two doors down who says it’s a waste of time.

The core argument was about whether scaling containers to zero actually saves anything meaningful. One person did the maths and landed on something like 61 kilowatt-hours a year, about twenty-five bucks in this country. Someone else pointed out you’d burn more running a dryer. Both true. And also both beside the point, I think.

Because nobody doing this is optimising for the twenty-five dollars. They’re optimising for the feeling of having built something that behaves the way they think it should behave. There’s a difference between “this saves money” and “this is correct,” and homelab people, in my limited experience, care a lot more about the second one. I get that. I spent a whole weekend once writing a bash script to automatically back up config files nobody was ever going to need, purely because the manual way offended me. Twenty minutes of manual copying, saved forever, at the cost of six hours of scripting. The maths never worked. I did it anyway.

The more interesting bit of the thread, buried under the power-bill argument, was the table someone posted about what should and shouldn’t scale to zero. DNS: no, obviously, you’ll break your own network. Home Assistant: no, it does things without being asked. A forum for mates: sure, why would that need to sit warm all night waiting for nobody. That’s the actual skill on display, not the YAML, but the judgment about what deserves to sleep and what doesn’t. Same instinct you’d want in someone managing infrastructure for actual paying customers, funnily enough. It’s the same conversation we have at work about what needs five-nines uptime and what can go down for ten minutes without the world ending, except at work someone’s paying you to have that argument and at home you’re just doing it for fun on a Tuesday night.

There’s a genuine tension in the thread that nobody resolved, because it doesn’t resolve. Spinning drives down and up more often can shorten their life. Leaving them spinning burns power continuously. Both camps had people with actual failure statistics. I don’t know who’s right, and I suspect it depends on the drive, the workload, and a bit of luck. Anyone giving you a confident answer either side is guessing with extra steps.

What I liked most, honestly, wasn’t the tech. It was watching someone build a public repo, invite strangers to pick it apart, and take the “why bother” comments in stride instead of getting precious about it. Half the internet can’t handle someone saying “I don’t see the point of this” without treating it as an attack. He just answered the question. That’s rarer than it should be.

I’m not going to scale my Pi-hole to zero. It’s already sipping about as little power as a thing can sip. But I might finally write down which of my little home scripts actually earn their keep and which ones I built because I was bored on a Sunday and wanted to feel clever. I suspect the honest answer is most of them are the second thing, and I’ve made peace with that.