Heroku Postgres is now based on AWS Aurora

Original Article ↗ Hacker News Discussion ↗

Heroku is launching new low-cost “Essential” Postgres tiers that run on AWS Aurora, prompting debate over whether Aurora can really deliver cheaper, better-performing databases than Heroku’s previous offerings or self-managed Postgres. Commenters highlight Aurora’s strengths in reliability, autoscaling, and storage architecture, but also its high cost, opaque quirks, and dependence on IOPS pricing. The move also rekindles broader questions about Heroku’s value in 2024 versus alternatives like Render, Crunchy Data, and DIY setups on VPS or EC2, especially for startups balancing ease of use, performance, and long‑term cloud costs.

Heroku Postgres Essential tiers & pricing

  • New “Essential” plans: 1/10/32 GB storage, low connection limits (20–40), $5–$20/month.
  • Positioned as entry-level, multi-tenant Aurora-backed DBs for toy apps/MVPs and pre-prod.
  • Replace old row-limited mini/basic tiers at same prices but with storage-based limits.
  • No replication, modest uptime target (99.5%), and other “full” features omitted; clearly distinct from larger dedicated plans.
  • Larger Aurora-backed dedicated offerings are promised “relatively soon.”

Heroku pricing vs direct cloud costs

  • Several commenters say traditional Heroku Postgres has very high margins; some report 5–10x savings by moving to AWS or self-managed Postgres.
  • Others note current top-end Heroku plans are still dramatically more expensive than equivalent AWS instances.
  • At the very low end ($5 plans), some are unsure how much margin actually remains.

Aurora cost, performance, and design

  • Some report Aurora as reliable and performant (especially with large row counts and upserts), but often the dominant line item in their AWS bill.
  • Aurora’s IO-based billing can be painful; IO-optimized storage is mentioned as a newer, more predictable option.
  • Others argue RDS or self-hosted Postgres on NVMe can be cheaper and faster, while Aurora’s value is in autoscaling storage, multi-AZ, and cross-region features.
  • Aurora’s log-structured storage and separation of compute and storage are cited as genuine technical innovations.

Aurora gotchas and operational issues

  • Reports of expensive I/O for poorly tuned queries; big cost drops after switching to self-managed Postgres.
  • Complaints about opaque behavior and undocumented differences vs vanilla Postgres/MySQL (e.g., temp storage limits, non-atomic table rename semantics, global write-forwarding latency).
  • Serverless v1 could scale to zero; v2 no longer does. Typical customers reportedly don’t use Serverless.
  • Advice to use pgbouncer rather than AWS’s own proxy in some setups.

Managed vs self-hosted Postgres

  • One camp: running your own Postgres is straightforward and avoids Aurora/Heroku markups.
  • Counterpoint: production-grade clusters (HA, backups, PITR, monitoring, upgrades) are non-trivial, not core to most businesses, and require expensive specialists.
  • Broader debate about over-outsourcing infra vs under-investing in DB skills; consensus that DB fundamentals still matter even if using managed services.

Who still uses Heroku & alternatives

  • Many still use Heroku for simplicity: “git push” deploys, low-ops, easier compliance (e.g., SOC 2) for small SaaS.
  • Others feel it’s stagnated (slow on HTTP/2, gRPC, IPv6; expensive VPC peering; async HA replication that can lose data).
  • Some are actively migrating databases to specialized Postgres providers (e.g., Crunchy) citing better performance, features like logical replication, and more granular storage pricing.
  • For app hosting, alternatives mentioned include Render, Railway, Northflank, Fly.io, DigitalOcean App Platform, ECS, etc., but several people find their developer experience still inferior or buggy compared to Heroku.
  • Negative sentiment from users in India over Heroku’s handling of card-regulation changes, perceived as abandoning smaller customers.

Cloud ecosystem & credits

  • Mixed views on startup behavior: some say free credits from major clouds keep startups on AWS/GCP/Azure; others highlight growth of higher-level platforms (Vercel, Netlify, Supabase, Render, Railway).
  • One startup describes heavy use of GPU credits across clouds, planning to move to owned hardware once credits expire.

AWS developer experience (Amplify)

  • A long critique calls Amplify one of AWS’s worst services: confusing split between CLI and GUI workflows, scattered CloudFormation/IAM/Cognito resources, inconsistent UIs, and documentation gaps around configuration files.
  • The experience is contrasted unfavorably with simpler static hosting/CI solutions like Netlify or Cloudflare Pages.