We cut our Mongo DB costs by 90% by moving to Hetzner

Original Article ↗ Hacker News Discussion ↗

Cost Savings vs Reliability Trade-off

  • Core move: from a 3-node MongoDB Atlas cluster to a single Hetzner bare-metal box, saving ≈$3k/month.
  • Many point out the comparison is not like-for-like: Atlas delivered multi-AZ redundancy; the new setup is a single-server SPOF.
  • Some argue this is fine if the database is non-critical analytics/ML data and occasional downtime is acceptable; others see it as reckless for anything customer-facing.
  • Commenters warn that once “90% savings” is celebrated publicly, it can be politically hard to get budget back for replicas later.

Cloud vs Bare Metal and Provider Experiences

  • Several say a simple dedicated machine can be more reliable in practice than complex cloud stacks, which fail in surprising ways.
  • Others counter that even a single EC2 instance often benefits from hyperscaler-level hardware management and live migration.
  • Hetzner experiences are mixed: some report years of excellent uptime; others describe recent flakiness, null-routing under “abuse” suspicions, and weak support. OVH and other low-cost hosts are mentioned as alternatives.
  • Bandwidth/egress pricing is a major pain point with AWS/Atlas; Hetzner’s cheap or “unlimited” traffic is a key factor in the savings.

Operational Complexity and Security

  • Critics stress that self-hosting adds responsibilities: backups, restore testing, monitoring, patching, and securing services (including network/firewall and disk encryption).
  • Some believe this complexity is overstated and comparable to wrangling managed-cloud setups; others cite real incidents where DIY infra or home‑rolled “S3” led to security failures.
  • Hetzner does not provide at-rest encryption by default; several recommend LUKS and off-provider backups.

MongoDB Atlas Pricing and Lock-in

  • Broad agreement that Atlas is expensive, often multiples of self-managed MongoDB on EC2 or bare metal, especially once storage, backups, and cross-AZ/network traffic are accounted for.
  • People mention opaque backup pricing, sharding limits, and replication traffic costs.
  • Some note Percona Server for MongoDB and community editions offering many “enterprise” features without Atlas fees.

Why MongoDB at All?

  • One camp questions MongoDB entirely, preferring Postgres (often with JSONB) for cost, maturity, and tooling.
  • Defenders cite schemaless documents, change streams, and ease of scaling/replication as fitting their domain models and speeding development.