An open-source collection of reproducible Postgres benchmarks produced by OnGres.
Every benchmark in this repo is end-to-end reproducible: Terraform brings up
the infrastructure, Ansible provisions it, scripts run the workload, and
analysis tooling renders the published plots — all from a single git clone +
make invocation. Costs and AWS quota requirements are spelled out per
benchmark.
This commitment to transparency is detailed in "Benchmarking: do it with transparency or don't do it at all".
| Subfolder | Topic | Status |
|---|---|---|
is-postgres-on-s3-faster-than-nvme/ |
Postgres on local NVMe — ext4 vs ZFS; single-node vs 3-AZ HA; full_page_writes on/off. The NVMe baseline for a larger "Postgres on S3 vs Postgres on NVMe?" series. |
Part I published, Part II WIP |
- One benchmark per top-level subfolder, self-contained: each owns its own Terraform, Ansible, scripts, and analysis venv.
- Shared tooling can be promoted to the repo root once a real pattern emerges across multiple benchmarks. Until then, accept the per-benchmark duplication.
- Reproducibility over polish: code is committed in the state in which it actually produced the published results, not a refactored ideal. Refactors happen between benchmark runs, not retroactively.
Apache 2.0 — see LICENSE.