Artillery load tests across FastAPI (sync), FastAPI (async), and Express, with capped CPU/RAM
This repository compares three API implementations (FastAPI synchronous, FastAPI asynchronous, and Express) against a shared PostgreSQL schema. Docker Compose profiles run one backend at a time; each service is constrained (as documented) to half a CPU core and 512 MB RAM so runs are more comparable. Artillery YAML suites cover read-heavy, write-heavy, spike, stress, soak, and breakpoint read/write patterns, with scripts to produce JSON and HTML reports.
When it is useful
You want repeatable numbers for talks, internal RFCs, or classroom exercises, not a single “winner” claim for every workload. The README also links a Looker Studio dashboard and ships example Artillery outputs for reference.
What you can do
- Start a profile (fastapi-sync, fastapi-async, or express) and run
run_artillery_test.shwith a scenario name from the docs. - Sweep all profiles and scenarios with
run_all_artillery_tests.shwhen you need a full matrix. - Inspect generated reports locally or compare against the bundled documentation snapshots; the README also points to an example Looker Studio dashboard at https://lookerstudio.google.com/reporting/270e16b6-8831-41d5-aed0-4d2352b39218/page/pqwOF.
Limits
- Results bind tightly to hardware, Docker limits, DB tuning, and Artillery config; your production stack will differ.
- Fairness still depends on equivalent app logic, connection pooling, and warm-up; interpret deltas carefully.
- This project benchmarks; it does not choose your framework for you without your own product constraints.
