stagDB Engineering

Notes from building production-adjacent dev tools: real benchmarks, AWS storage internals, and what it takes to give every developer production-scale data.

Aug 7, 2026 Copy-on-write development databases: ten clones at 1.12× the storage
Per-developer copies of production data are O(size × developers), which is why teams settle for seed data and shared staging. Copy-on-write collapses the multiplication: measured storage amplification on FSx for OpenZFS.
Aug 7, 2026 Clone latency on FSx for OpenZFS: from 10 minutes to 16 seconds
A ZFS clone is a metadata operation. On FSx for OpenZFS the provisioning call takes about two minutes and serializes under concurrency. How we got from 600 s to 16 s with a pre-warmed clone pool.
Aug 7, 2026 Benchmarking FSx for OpenZFS for Postgres: throughput vs. provisioned IOPS
8× the provisioned throughput bought +29% Postgres TPS. 3× the IOPS bought 2.5×. Benchmark data for sizing FSx for OpenZFS under a transactional workload.