Picture by Creator
# Introduction
Stress testing is essential for understanding how your utility behaves beneath heavy load. For machine learning-powered APIs, it’s particularly essential as a result of mannequin inference may be CPU-intensive. By simulating numerous customers, we will establish efficiency bottlenecks, decide the capability of our system, and guarantee reliability.
On this tutorial, we might be utilizing:
- FastAPI: A contemporary, quick (high-performance) internet framework for constructing APIs with Python.
- Uvicorn: An ASGI server to run our FastAPI utility.
- Locust: An open-source load testing software. You outline consumer conduct with Python code, and swarm your system with tons of of simultaneous customers.
- Scikit-learn: For our instance machine studying mannequin.
# 1. Undertaking Setup and Dependencies
Arrange the venture construction and set up the required dependencies.
- Create
necessities.txtfile and add the next Python packages: - Open your terminal, create a digital atmosphere, and activate it.
- Set up all of the Python packages utilizing the
necessities.txtfile.
fastapi==0.115.12
locust==2.37.10
numpy==2.3.0
pandas==2.3.0
pydantic==2.11.5
scikit-learn==1.7.0
uvicorn==0.34.3
orjson==3.10.18
python -m venv venv
venvScriptsactivate
pip set up -r necessities.txt
# 2. Constructing the FastAPI Software
On this part, we are going to create a file for coaching the Regression mannequin, for pydantic fashions, and the FastAPI utility.
This ml_model.py handles the machine studying mannequin. It makes use of a singleton sample to make sure just one occasion of the mannequin is loaded. The mannequin is a Random Forest Regressor educated on the California housing dataset. If a pre-trained mannequin (mannequin.pkl and scaler.pkl) does not exist, it trains and saves a brand new one.
app/ml_model.py:
import os
import threading
import joblib
import numpy as np
from sklearn.datasets import fetch_california_housing
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
class MLModel:
_instance = None
_lock = threading.Lock()
def __new__(cls):
if cls._instance is None:
with cls._lock:
if cls._instance is None:
cls._instance = tremendous().__new__(cls)
return cls._instance
def __init__(self):
if not hasattr(self, "initialized"):
self.mannequin = None
self.scaler = None
self.model_path = "mannequin.pkl"
self.scaler_path = "scaler.pkl"
self.feature_names = None
self.initialized = True
self.load_or_create_model()
def load_or_create_model(self):
"""Load current mannequin or create a brand new one utilizing California housing dataset"""
if os.path.exists(self.model_path) and os.path.exists(self.scaler_path):
self.mannequin = joblib.load(self.model_path)
self.scaler = joblib.load(self.scaler_path)
housing = fetch_california_housing()
self.feature_names = housing.feature_names
print("Mannequin loaded efficiently")
else:
print("Creating new mannequin...")
housing = fetch_california_housing()
X, y = housing.information, housing.goal
self.feature_names = housing.feature_names
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
self.scaler = StandardScaler()
X_train_scaled = self.scaler.fit_transform(X_train)
self.mannequin = RandomForestRegressor(
n_estimators=50, # Lowered for quicker predictions
max_depth=8, # Lowered for quicker predictions
random_state=42,
n_jobs=1, # Single thread for consistency
)
self.mannequin.match(X_train_scaled, y_train)
joblib.dump(self.mannequin, self.model_path)
joblib.dump(self.scaler, self.scaler_path)
X_test_scaled = self.scaler.rework(X_test)
rating = self.mannequin.rating(X_test_scaled, y_test)
print(f"Mannequin R² rating: {rating:.4f}")
def predict(self, options):
"""Make prediction for home value"""
features_array = np.array(options).reshape(1, -1)
features_scaled = self.scaler.rework(features_array)
prediction = self.mannequin.predict(features_scaled)[0]
return prediction * 100000
def get_feature_info(self):
"""Get details about the options"""
return {
"feature_names": checklist(self.feature_names),
"num_features": len(self.feature_names),
"description": "California housing dataset options",
}
# Initialize mannequin as singleton
ml_model = MLModel()
The pydantic_models.py file defines the Pydantic fashions for request and response information validation and serialization.
app/pydantic_models.py:
from typing import Listing
from pydantic import BaseModel, Area
class PredictionRequest(BaseModel):
options: Listing[float] = Area(
...,
description="Listing of 8 options: MedInc, HouseAge, AveRooms, AveBedrms, Inhabitants, AveOccup, Latitude, Longitude",
min_length=8,
max_length=8,
)
model_config = {
"json_schema_extra": {
"examples": [
{"features": [8.3252, 41.0, 6.984, 1.024, 322.0, 2.556, 37.88, -122.23]}
]
}
}
app/principal.py: This file is the core FastAPI utility, defining the API endpoints.
import asyncio
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException
from fastapi.responses import ORJSONResponse
from .ml_model import ml_model
from .pydantic_models import (
PredictionRequest,
)
@asynccontextmanager
async def lifespan(app: FastAPI):
# Pre-load the mannequin
_ = ml_model.get_feature_info()
yield
app = FastAPI(
title="California Housing Worth Prediction API",
model="1.0.0",
description="API for predicting California housing costs utilizing Random Forest mannequin",
lifespan=lifespan,
default_response_class=ORJSONResponse,
)
@app.get("/well being")
async def health_check():
"""Well being examine endpoint"""
return {"standing": "wholesome", "message": "Service is operational"}
@app.get("/model-info")
async def model_info():
"""Get details about the ML mannequin"""
strive:
feature_info = await asyncio.to_thread(ml_model.get_feature_info)
return {
"model_type": "Random Forest Regressor",
"dataset": "California Housing Dataset",
"options": feature_info,
}
besides Exception:
increase HTTPException(
status_code=500, element="Error retrieving mannequin data"
)
@app.put up("/predict")
async def predict(request: PredictionRequest):
"""Make home value prediction"""
if len(request.options) != 8:
increase HTTPException(
status_code=400,
element=f"Anticipated 8 options, received {len(request.options)}",
)
strive:
prediction = ml_model.predict(request.options)
return {
"prediction": float(prediction),
"standing": "success",
"features_used": request.options,
}
besides ValueError as e:
increase HTTPException(status_code=400, element=str(e))
besides Exception:
increase HTTPException(status_code=500, element="Prediction error")
Key factors:
lifespansupervisor: Ensures the ML mannequin is loaded throughout utility startup.asyncio.to_thread: That is essential as a result of scikit-learn’s predict technique is CPU-bound (synchronous). Working it in a separate thread prevents it from blocking FastAPI’s asynchronous occasion loop, permitting the server to deal with different requests concurrently.
Endpoints:
/well being: A easy well being examine./model-info: Offers metadata concerning the ML mannequin./predict: Accepts an inventory of options and returns a home value prediction.
run_server.py: It accommodates the script that’s used to run the FastAPI utility utilizing Uvicorn.
import uvicorn
if __name__ == "__main__":
uvicorn.run("app.principal:app", host="localhost", port=8000, staff=4)
All of the information and configurations can be found on the GitHub repository: kingabzpro/Stress-Testing-FastAPI
# 3. Writing the Locust Stress Take a look at
Now, let’s create the stress take a look at script utilizing Locust.
checks/locustfile.py: This file defines the conduct of simulated customers.
import json
import logging
import random
from locust import HttpUser, job
# Scale back logging to enhance efficiency
logging.getLogger("urllib3").setLevel(logging.WARNING)
class HousingAPIUser(HttpUser):
def generate_random_features(self):
"""Generate random however reasonable California housing options"""
return [
round(random.uniform(0.5, 15.0), 4), # MedInc
round(random.uniform(1.0, 52.0), 1), # HouseAge
round(random.uniform(2.0, 10.0), 2), # AveRooms
round(random.uniform(0.5, 2.0), 2), # AveBedrms
round(random.uniform(3.0, 35000.0), 0), # Population
round(random.uniform(1.0, 10.0), 2), # AveOccup
round(random.uniform(32.0, 42.0), 2), # Latitude
round(random.uniform(-124.0, -114.0), 2), # Longitude
]
@job(1)
def model_info(self):
"""Take a look at well being endpoint"""
with self.consumer.get("/model-info", catch_response=True) as response:
if response.status_code == 200:
response.success()
else:
response.failure(f"Mannequin data failed: {response.status_code}")
@job(3)
def single_prediction(self):
"""Take a look at single prediction endpoint"""
options = self.generate_random_features()
with self.consumer.put up(
"/predict", json={"options": options}, catch_response=True, timeout=10
) as response:
if response.status_code == 200:
strive:
information = response.json()
if "prediction" in information:
response.success()
else:
response.failure("Invalid response format")
besides json.JSONDecodeError:
response.failure("Did not parse JSON")
elif response.status_code == 503:
response.failure("Service unavailable")
else:
response.failure(f"Standing code: {response.status_code}")
Key factors:
- Every simulated consumer will wait between 0.5 and a couple of seconds between executing duties.
- Creates reasonable random characteristic information for the prediction requests.
- Every consumer will make one health_check request and three single_prediction requests.
# 4. Working the Stress Take a look at
- To judge the efficiency of your utility beneath load, start by beginning your asynchronous machine studying utility in a single terminal.
- Open your browser and navigate to http://localhost:8000/docs. Use the interactive API documentation to check your endpoints and guarantee they’re functioning appropriately.
- Open a brand new terminal window, activate the digital atmosphere, and navigate to your venture’s root listing to run Locust with the Internet UI:
- Within the Locust internet UI, set the full variety of customers to 500, the spawn fee to 10 customers per second, and run it for a minute.
- Through the take a look at, Locust will show real-time statistics, together with the variety of requests, failures, and response occasions for every endpoint.
- As soon as the take a look at is full, click on on the Charts tab to view interactive graphs displaying the variety of customers, requests per second, and response occasions.
- To run Locust with out the net UI and routinely generate an HTML report, use the next command:
Mannequin loaded efficiently
INFO: Began server course of [26216]
INFO: Ready for utility startup.
INFO: Software startup full.
INFO: Uvicorn operating on http://0.0.0.0:8000 (Press CTRL+C to stop)
locust -f checks/locustfile.py --host http://localhost:8000
Entry the Locust internet UI at http://localhost:8089 in your browser.
locust -f checks/locustfile.py --host http://localhost:8000 --users 500 --spawn-rate 10 --run-time 60s --headless --html report.html
After the take a look at finishes, an HTML report named report.html might be saved in your venture listing for later assessment.
# Last Ideas
Our app can deal with numerous customers as we’re utilizing a easy machine studying mannequin. The outcomes present that the model-info endpoint has a better response time than the prediction, which is spectacular. That is the best-case state of affairs for testing your utility regionally earlier than pushing it to manufacturing.
If you want to expertise this setup firsthand, please go to the kingabzpro/Stress-Testing-FastAPI repository and comply with the directions within the documentation.
Abid Ali Awan (@1abidaliawan) is a licensed information scientist skilled who loves constructing machine studying fashions. At present, he’s specializing in content material creation and writing technical blogs on machine studying and information science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students scuffling with psychological sickness.
