How to Profile Flask Applications with py-spy Without Adding Code Instrumentation

Mar 15, 2026

py-spy Flask profiler no instrumentation: Flask app under load (200req/s) → high CPU? Traditional cProfile/decorator slows prod. py-spy: Zero code changes, attach to live PID (ps aux | grep gunicorn), py-spy top live view (90% bottleneck in numpy loop), py-spy record -o flame.svg. Fix: vectorize → 5x speedup (1000req/s). Targets: “profile flask without code changes”, “py-spy flask gunicorn”, “flask cpu profiling live”.

Why py-spy for Flask? (Prod-Safe Sampling)

cProfile: Insert @profile → prod pollution. py-spy: OS signals (sampling 10ms), low overhead (<3% CPU), multi-thread (Gunicorn workers).

Profiler	Code Change	Overhead	Live Attach	Flamegraph
cProfile	Yes	20-50%	No	Manual
pyinstrument	Decorator	10%	No	Yes
py-spy	No	<3%	Yes	Built-in
Scalene	No	5%	Yes	Yes

Install py-spy (Rust Binary, 10s)

# Recommended: Cargo (fastest)
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
cargo install py-spy

# Or pip (wheels)
pip install py-spy

Verify: py-spy --version

Sample Slow Flask App (Gunicorn)

app.py (CPU bottleneck: naive loop):

from flask import Flask
import time
import numpy as np  # Sim heavy compute

app = Flask(__name__)

@app.route('/slow')
def slow_compute():
    result = 0
    for i in range(10**6):  # 100ms+ loop
        result += np.sin(i) * np.cos(i)
    return {'result': result}

@app.route('/health')
def health():
    return {'status': 'ok'}

if __name__ == '__main__':
    app.run()

gunicorn.conf.py:

bind = "0.0.0.0:5000"
workers = 4

Run: gunicorn -c gunicorn.conf.py app:app

Load test: locust -f locustfile.py --headless -u 50 -r 10 (or ab -n 1000 -c 50 http://localhost:5000/slow)

Step 1: Find Flask PID

ps aux | grep '[g]unicorn'
# Or: pgrep -f 'gunicorn.*app:app'
# PID 12345

Step 2: Live Top (Instant Insights)

py-spy top --pid 12345 --sort cpu

Output:

PID    CPU%   COMMAND       slow_compute()    92.3%
12345  92.3%  gunicorn      np.sin()          45.1%
12345  45.1%  gunicorn      for loop          40.2%

Bottleneck: 92% in /slow loop.

Step 3: Flamegraph (Shareable SVG)

py-spy record --pid 12345 -d 30 -o flask-profile.svg --subprocesses --rate 1000

30s sample → flask-profile.svg (open browser). Wide bars = hotspots.

Fix: Vectorize Loop (5x Speedup)

Before: 200req/s

for i in range(10**6):
    result += np.sin(i) * np.cos(i)

After:

i = np.arange(10**6)
result = np.sum(np.sin(i) * np.cos(i))

Reload Gunicorn: kill -HUP 12345

Benchmarks (M2 Mac, 4 Workers)

Endpoint	req/s Before	req/s After	CPU% Drop
/slow loop	210	1050	92→18
/health	5000	5000	0

wrk -t12 -c400 -d30s http://localhost:5000/slow

Pitfalls & Fixes

Issue	Cause	Fix
Permission denied	Non-root PID	`sudo py-spy` or `setcap`
No flamegraph	Old py-spy	`cargo update py-spy`
Threads missing	Python GIL	`--native` (C exts)
Uvicorn/FastAPI	ASGI	Works same PID
Docker	Host network	`docker exec` PID

Checklist

py-spy → Profile any Flask prod today: 5min setup, instant wins.