From Localhost to Hetzner: A Love Story in Four Acts

Or: How I Became a Hosting Provider Connoisseur Without Meaning To

Act I: The Humble Beginning (localhost)

It all started innocently enough. I built an image optimizer API in Go, slapped together a Next.js frontend, and ran it on my laptop while sipping coffee and feeling like a real developer. Life was simple. The app was fast. The costs were zero (well, except for the electricity bill and the therapy needed after debugging CORS for the third time).

But then came the inevitable question: “Can other people use this?”

Narrator: They could not use this.

Act II: The Render Romance (render.com)

“Deploy to Render!” the internet shouted. “It’s free! It’s easy! It has a cute dashboard!”

And it was! For about two weeks. Then I discovered that:

Free tier goes to sleep after inactivity (rude)
Waking up takes 30-60 seconds (very rude)
Users don’t appreciate waiting a minute to optimize a tiny PNG

So I did what any rational developer would do: I panic-searched “fast hosting” at 2 AM and found…

Act III: The Fly.io Fling (fly.io)

Fly.io promised edge computing, global distribution, and vibes. The deployment was chef’s kiss. The CLI was beautiful. The architecture diagrams in their docs made me feel smart.

Everything was perfect until one fateful day when a user uploaded a 10,766×9,913 pixel sprite sheet.

Fly.io: “Out of Memory” Me: “But I gave you 2GB!” Fly.io: “Yeah, about that…”

Turns out, processing massive images with libvips while auto-resizing and adding padding requires… more than 2GB of RAM. Who knew? (Spoiler: Everyone who’s ever processed images knew.)

The Pricing Revelation

I looked at upgrading to 4GB of RAM on Fly.io:

2GB VM: ~$12/month
4GB VM: ~$24/month

Then I looked at Hetzner:

4GB VPS (CX22): €5.83/month (~$6.50)

Record scratch sound

Wait. Let me get this straight. I can pay:

More for less RAM and keep the cool kids hosting
Less for more RAM but have to do actual DevOps

My wallet made the decision for me.

Act IV: The Hetzner Homecoming (hetzner.com)

And so began the great migration. Armed with a 900-line migration guide (yes, I wrote it myself, yes, I’m fun at parties), I embarked on the journey from managed cloud utopia to “I own a server now, help.”

What I Learned

Things that were harder than expected:

Remembering that bimg requires CGO so you can’t just build on macOS and yeet the binary to Linux
Creating /var/log/caddy before Caddy tries to write to it (who knew permissions mattered?)
Explaining to myself why I needed to type systemctl so many times

Things that were easier than expected:

Caddy automatically getting SSL certificates (thank you, Let’s Encrypt overlords)
Ubuntu 22.04 just… working
The existential satisfaction of running htop and seeing YOUR processes on YOUR server

The Results

Before (Fly.io):

2GB RAM
OOM errors on large images
$12/month
Cool deployment story at meetups

After (Hetzner):

4GB RAM (2x!)
No OOM errors
€5.83/month (~$6.50)
Slightly less cool deployment story, significantly cooler AWS bill… wait, wrong cloud

Savings: $66/year (that’s like… 13 fancy coffees, or 66 regular coffees, or infinite water)

The Stack (For the Nerds)

Frontend: GitHub Pages Status: Never touched it, still works, 10/10

Backend: Hetzner Cloud CX22

Ubuntu 22.04 (because I’m basic)
Go 1.22 with libvips (for the image magic)
Caddy (the reverse proxy that just works™)
Systemd (because apparently I run a real server now)

Deployment Process:

rsync source-code hetzner-server:/tmp/
ssh hetzner-server
go build
mv binary /usr/local/bin/
systemctl restart image-optimizer

Future me reading this: “Why didn’t you just use Docker?” Current me: “BECAUSE WE’RE ITERATING, OKAY?”

Lessons Learned

Hosting loyalty is a myth - Use what works, migrate when it doesn’t
Always have 2x the RAM you think you need - Images are sneaky memory hogs
Write migration docs DURING migration - Future you will thank past you
Managed services are worth it… until they’re not - The line is different for everyone
Europe has better VPS pricing - Sorry, American data centers

What’s Next?

Now that I’m no longer hemorrhaging money on RAM I don’t have, I can focus on the important things:

Actually use the extra 2GB of RAM
Write more blog posts about migrations (kidding)
Build features instead of moving servers
Resist the urge to migrate to another provider just for fun
Set up monitoring before something breaks

The Real Moral of the Story

Sometimes the best solution isn’t the shiniest one. Sometimes it’s the boring VPS from a German company you’ve never heard of that just quietly works and costs half as much.

Also, if you’re processing images: buy more RAM than you think you need. This is not a suggestion. This is a warning from someone who learned the hard way.

Acknowledgments

Fly.io: Thanks for the good times and the fancy CLI
Render.com: You tried, buddy
Localhost: You’ll always be my first
Hetzner: Welcome to the family
Let’s Encrypt: You’re doing the lord’s work
The 10,766×9,913 pixel sprite: You monster

P.S. - If you’re reading this in 2026 and I’ve migrated again, no you didn’t.

P.P.S. - If I ever write “From Hetzner to [X]”, someone please stage an intervention.

Technical specs for the curious:

Migration time: ~2 hours (including fixing my own documentation bugs)
Downtime: ~5 minutes (DNS propagation)
Lines of migration docs: 900+
Bugs encountered: 3 (CGO cross-compilation, log permissions, missing version variables)
Coffees consumed: Yes

Want to see the migration guide? It’s probably on GitHub with more systemd commands than anyone should ever need.

Have your own hosting horror story? I’d love to hear it. Misery loves company.

Update 2025-10-20: It’s been 3 hours since migration. No OOM errors. Living the dream.