Skip to content
Register

Automation question: should I trust provider APIs with my infrastructure?

General Discussion by pauloserver 24 replies 2.5K views
pauloserver

Member OP

13 #1

I need advice. My Terraform provider for RackNerd got deprecated without warning. No email, no changelog, just "this provider is no longer maintained" when I run init. Going to break my whole CI/CD pipeline if I don't fix fast.

I was going to just switch everything to Vultr provider, but my friend say I going to get burned again. Should I build abstraction layer? Use OpenTofu? Or just accept that all cloud APIs are trash and going to betray me eventually?

What you all doing for provider stability? I don't want to rewrite HCL every six months

chill infrastructure for chill people 🦫
moebig

Member

RAID is not backup

#2

Wesh pauloserver

I learned this le hard way. Now I use provider-agnostic setup, incha'allah for uptime. Basically I wrap le Vultr API, le RackNerd API, and le Hetzner API in one module. حلو, right?

If one provider die, I switch variable and redeploy. French-style redundancy:)

RAID 1: because paranoia pays
Nadia

Member

/dev/null enthusiast

#3
moebig said:
Wrap le Vultr API, le RackNerd API, and le Hetzner API

That's not agnostic, that's just three redirects to the same dependency pipe

You've built a /dev/null with extra steps. When all three providers change their schema, your wrapper gets EOF'd

I pipe everything through Pulumi with dynamic providers now. Still vendor-locked but at least the lock-in is redirected to one layer I control

/dev/null: full of good ideas
garykwh

Member

Inference on a Budget

#4

Ngl the real play is just run everything on k8s and treat nodes as cattle lol

I was doing LLM inference on RackNerd GPUs, 48GB VRAM per A40, and their API changes broke my node pool autoscaler. Migrated to Vultr with same terraform module in like 2 hours.

For AI workloads you can't afford provider loyalty. VRAM is VRAM. Lol.

CUDA cores are my love language
PacketpunkUma

Member

#5

Wrapping three APIs is not agnostic! That's just more code to break! Fire your friend!

I run everything on bare metal with IPMI scripts! No provider API! No terraform! Just curl and prayer!

Actually I lie! I use ansible! But I own the boxes! 💀

mariannoodle

Member

Debian Stable

#6

Lol @PacketpunkUma calm down Tbh I just use nixos and if the provider api breaks I redeploy somewhere else in like 10 mins. Wrote my own tiny wrapper around the Vultr api, not because I'm smart but because I got burned by RackNerd too For my homelab I use proxmox on a Time4VPS box. No api drama there tbh

declarative, reproducible, indescribably good
Hana5

Member

GPU poor

#7

Please do not trust provider API

I wrap with server abstraction, but server is liar just three API call in trenchcoatBetter to use own images and migrate fast no "senior team" can help when API change at 3amI learned this from Oracle free tier, they break API every month

8-char NTLM found in 4min 32sec
MA

Member

#8

Three APIs in one module? That's just triple the breakage surface.

RAJ

Member

#9

Same with Oracle, their changelog is a fiction novel.

your margin is my opportunity
ana_mad

Member

#10

Vultr schema changes twice a quarter, good luck wrapping that.

swimming upstream since 2019 🐟

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft