Neon Serverless Postgres is generally available

Launch timing and positioning

• Several commenters noticed GA landing almost simultaneously with another Postgres PaaS; company reps say the date was independently planned and delayed once for internal reasons.
• Neon is framed as “serverless Postgres” with focus on autoscaling, scale-to-zero, and dev‑friendly features like branching and Vercel integration.

Architecture and features

• Storage and compute are separated: Postgres runs in small VMs inside Kubernetes pods, backed by a custom Rust storage layer.
• WAL is indexed to support point‑in‑time recovery and “branches” (copy‑on‑write database history for dev/test, troubleshooting, and time travel).
• Multi‑tenant isolation is via per‑tenant Postgres VMs on shared bare‑metal nodes.
• There’s an explicit local cache between Postgres and pageservers to compensate for lack of OS filesystem cache.

Serverless behavior and scaling

• Databases scale to zero after inactivity; users can set min/max compute so production can stay warm.
• Cold starts are claimed to be a few hundred milliseconds; one user reported seconds plus a failed first request, which they fixed with connection‑pool settings.
• Architecture is said to allow effectively unbounded storage and large compute up to the biggest AWS metal node plus replicas.

Pricing and cost concerns

• Several users find storage expensive ($1.50+/GiB, plus PITR/retention bundled in), especially for light‑compute, moderate‑storage workloads.
• Some report prior 10x storage price hikes and worry about future changes.
• Neon staff say COGS are high due to multiple replicas (safekeepers, NVMe cache, dual S3 copies) and that pricing and storage/retention separation are being reconsidered.
• Confusion exists about compute‑hour accounting; staff clarify each paid tier includes enough hours to run one instance 24/7, with extras for branches.

Reliability and production readiness

• Mixed reports: some users say Neon has been “just works” reliable and cost‑effective in production; others describe multiple outages in Q4 2023 and migrated away.
• Uptime metrics were removed from the status page due to incorrect calculation; this was perceived by some as hiding poor reliability. Staff say a new, more accurate model (inspired by other vendors) is coming.
• Company representatives emphasize recent stability work, frequent internal reviews, and that GA reflects reaching a higher reliability bar.

Open source, self‑hosting, and lock‑in

• Core storage and Postgres fork are open source; the full control plane is not. A toy local control plane exists, and third‑party operators are emerging.
• Some users are self‑hosting Neon on Kubernetes (often with Ceph) to get the branching/storage model with their own cost structure.
• Concern about lock‑in is countered by claims of vanilla Postgres compatibility and the ability to fall back to standard Postgres elsewhere.

Performance and developer experience

• Several users praise very fast database provisioning compared to RDS and the convenience of branching for per‑PR or per‑dev environments.
• Others report rough early experiences: dashboard bugs, Node.js driver pool issues with serverless, and difficulties estimating costs.
• Questions remain about performance on very large datasets; no concrete benchmarks were shared, but staff point to their caching layer and NVMe/object‑storage hybrid design.

Compliance and data handling

• Neon staff state they don’t inspect user data without permission and that data is encrypted at rest (NVMe and S3).
• EU regions are supported; full details on strict EU‑only access and broader compliance posture are not fully elaborated in the thread.