I'm nicking this for my next standup. But seriously, has anyone actually tried OpenTofu with the Vultr provider? Curious if the fork is keeping up better or if it's the same maintainers under a different hat.
Can confirm, their block storage attachment syntax changed in March and broke my state. But here's the thing: Vultr has a São Paulo region now, so pauloserver at least gets low latency for the pain.
For LLM stuff I just keep AMI-style snapshots ready. API breaks, I nuke and pave in a new region. 48GB A40s are 48GB A40s wherever they live.
And I pay 21% VAT on top of USD because they bill from Delaware but my card is Argentine. The API drama is the cheap part.
Hetzner won't take my money at all, they blocked AR cards last year. So I'm stuck with whoever takes me, schema changes and all.
Member OP
Yes! This is why I was looking at them. My RackNerd boxes are all in LA, ping to SP is 180ms. Vultr SP is 15ms for my users.
But the provider deprecation scared me. If I build everything on Vultr and they do same thing...
I run a small hosting co-op in Barcelona. We got burned by a provider API change last year, now we maintain our own libcloud fork for the three providers we use.
It's not pretty but it's ours. When Vultr changed their firewall rule format in June we patched in two days instead of waiting for upstream.
Upstream is just another dependency that ghosts you.
Same maintainers, different hat. I checked. The Vultr provider on OpenTofu registry is literally the same HashiCorp-published binary with the license stripped.
You're not escaping anything, you're just downloading from a different URL.
Pulumi at least has their own provider written in Go, separate codebase. Still vendor lock but different lock.
Member
Hetzner takes my PLN no problem. Their auction servers are 34 EUR for i7-6700 with 64GB, I have three. Falkenstein ping to Wrocław is 18ms.
But their API is minimal. No terraform provider drama because there is barely API to break. You want fancy automation? Hetzner is not for you. You want cheap iron that stays up? Tak.
Member
This is the real point everyone skips. I manage 40 client sites across Madrid and Barcelona. The providers with the richest APIs break the most because they ship the most features.
My Time4VPS box in Vilnius? API does start, stop, reinstall. That's it. Three endpoints, zero drama in four years.
Complexity is the enemy. You want reliability, find a provider that moves slow.
In Japan we have Sakura Cloud, very stable API, but no GPU. For GPU I use Vultr Tokyo or Oracle. Oracle API is nightmare, Vultr is "only" bad dream.
I now use packer to build images, then terraform only for "launch instance from image". If API break, I can still launch manually from web console and fix terraform later.
Terraform becomes documentation, not execution.
Ah so OpenTofu is halal wrapper around haram binary? I cancel my migration plan then.
Back to my three-API module. Is ugly but I understand ugly. I do not understand ghost.
