Neon Storage

Neon Storage is an open-source storage architecture that separates compute from persistence for PostgreSQL workloads. By splitting stateless Postgres compute nodes from a distributed Rust-based storage layer, it enables copy-on-write database branching and rapid scaling down to zero. Teams run the engine on Linux or Kubernetes clusters backed by S3-compatible object storage to operate serverless Postgres infrastructure without recurring cloud platform fees.

Why Neon Storage instead of Neon

Neon operates serverless Postgres as a managed cloud service, billing teams for compute runtime, active connection time, and per-gigabyte storage. For continuous integration pipelines or test suites spinning up hundreds of preview branches, managed consumption costs accumulate quickly. Neon Storage provides the core engine behind that platform, letting DevOps teams deploy pageservers and safekeepers on their own hardware. Storage expenses drop to raw S3, Ceph, or MinIO capacity, turning database branching from a metered expense into an internal utility.

The difference centers on operational overhead. Neon provides turnkey provisioning, a web administrative console, automated failover, and global connection pooling. Operating Neon Storage in-house requires platform engineers to run Kubernetes workloads, monitor write-ahead log consensus across safekeepers, and manage pageserver compaction. Compared to proprietary engines like Amazon Aurora, self-hosting Neon Storage eliminates AWS lock-in and high Aurora Serverless compute minimums, but shifts multi-tenant security and backup orchestration onto the engineering team.

What it does well

• Instant copy-on-write branching creates isolated database environments from any point in write-ahead log history without duplicating physical table files.
• Scale-to-zero compute provisioning detaches stateless Postgres instances when idle, freeing RAM and CPU across staging environments.
• Decoupled storage hierarchy shifts long-term page history to S3-compatible object stores while caching active pages in local NVMe and RAM.
• Quorum-based write replication streams write-ahead log records across safekeeper nodes via Paxos before confirming commits, guaranteeing durability without single-disk dependencies.
• Standard PostgreSQL compatibility preserves standard SQL execution, allowing standard drivers, ORMs, and extensions like pgvector to connect unmodified.