Back to browse
GitHub Repository

DuckLake on Hetzner for under 15 euros a month.

44 starsPython

Deploy a DuckLake data lakehouse on Hetzner for under €10/mo

by deezypls·Feb 16, 2026·3 points·0 comments

AI Analysis

●●SolidSolve My ProblemNiche Gem

One-command DuckLake deployment on Hetzner, but orchestration of existing tools.

Strengths
  • Real labor savings: reduces 'quite a few steps' to 'make all', eliminating friction for DuckLake tryouts
  • Cost transparency and honest caveats (cx33 stock issues, production security notes) build credibility
Weaknesses
  • Orchestrates OpenTofu + PyInfra + DuckDB/PostgreSQL/S3—no novel infrastructure pattern, just wiring
  • Limited audience: niche within niche (DuckLake early adopters + Hetzner price-seekers)
Target Audience

Data engineers and analysts wanting to self-host DuckLake on budget infrastructure

Similar To

Self-hosted data stack templates (general) · Neon serverless Postgres (different approach)

Post Description

DuckLake launched recently and I wanted to try it out.

It's DuckDB, but with a proper catalog on PostgreSQL with S3 for storage. It still takes quite a few steps to set up, so I put together a setup that runs on Hetzner.

It sets up a VPS with PostgreSQL, an S3 bucket, and a DuckDB init script that connects everything. One 'make all' command and it works. Used OpenTofu for the infra, PyInfra for server config.

Runs me about €9/mo with 1TB of storage. The VPS is a cx33 (4 vCPU, 8GB RAM) because Hetzner never has the cheaper tier in stock, maybe you'll get lucky.

Nothing fancy, but if you'd like to try out DuckLake, it should help you get going quickly!

Similar Projects

AI/ML●●Solid

Quantlix – Deploy AI models instantly without managing infrastructure

Instant CLI deploys, a tiny REST surface and SDKs for Python/Node make spinning up an inference endpoint trivial — you can go from pip/npm to curl in minutes. The landing page leans into useful operational promises (EU hosting, scale-to-zero, <50ms inference) and clear pricing tiers, but the space is crowded (Replicate/Hugging Face/etc.) and the page lacks independent benchmarks or details about supported model formats, versioning, or runtime guarantees.

SlickSolve My Problem
jensenjesper
213mo ago