Atomic database path
Business rules live in PostgreSQL stored procedures so debits, credits, limits, and balances resolve inside one transactional boundary.
Rinha de Backend 2024/Q1 · Python pressure lab
Slow is Smooth.
Smooth is Fast.
A Flask/Gunicorn/PostgreSQL implementation tuned for a banking API challenge: two Python APIs, one NGINX edge, PostgreSQL atomic procedures, and a 1.5 CPU / 550MB envelope.
- Python 3.14
- Flask 3.1
- Gunicorn
- PostgreSQL procedures
- NGINX
- k6 CI
LanguagePython 3.14
APIFlask + Gunicorn
DatabasePostgreSQL procedures
EdgeNGINX least_conn
Load testk6 reports
source-backed constraints
The whole stack fits inside the contest envelope.
class ChallengeEnvelope:
# Rinha de Backend 2024/Q1
TOTAL_CPU = 1.5
TOTAL_RAM_MB = 550
# Request path
EDGE = "NGINX"
API_REPLICAS = 2
API_RUNTIME = "Flask + Gunicorn"
DATABASE = "PostgreSQL procedures"
# Validation
STRESS_TOOL = "k6"
REPORTS_ARCHIVED = 33implementation mechanisms
Not async theater. A measured WSGI and Postgres path.
Two WSGI API replicas
The Python service runs as two Flask/Gunicorn containers behind NGINX, tuned for the challenge CPU and memory split.
Constraint-first validation
The site links directly to committed k6 HTML reports and the release workflow artifacts so performance claims stay tied to evidence.
benchmark dossier
Numbers stay attached to run artifacts.
k6 reports are published as static artifacts. Treat these values as historical proof points, then open the reports to inspect the exact run output before making performance decisions.
latest artifact
2026-04-01 · 07:36 UTC
Open the newest archived k6 HTML report, then compare it with the workflow that generated it.
Recent k6 reports
33 archived stress test runs. Latest first.