Load Modeling

User Behavior Modeling

A load test that only hammers a single endpoint is not realistic. Real users follow flows: they log in, browse, add to cart, pay, log out.

Realistic Flow Example

from locust import HttpUser, task, between, SequentialTaskSet


class CheckoutFlow(SequentialTaskSet):
    def on_start(self) -> None:
        resp = self.client.post("/auth/login", json={
            "email": "user@example.com",
            "password": "secret",
        })
        self.token = resp.json()["token"]

    @task
    def browse(self) -> None:
        self.client.get("/products", headers={"Authorization": f"Bearer {self.token}"})

    @task
    def add_to_cart(self) -> None:
        self.client.post("/cart", json={"product_id": 1, "qty": 1},
                         headers={"Authorization": f"Bearer {self.token}"})

    @task
    def checkout(self) -> None:
        self.client.post("/orders",
                         headers={"Authorization": f"Bearer {self.token}"})
        self.interrupt()


class ShopUser(HttpUser):
    tasks = [CheckoutFlow]
    wait_time = between(1, 2)

SequentialTaskSet enforces task order — the user cannot checkout before browsing.

---

Load Patterns

Constant Load

Flat number of users for the full test duration. Used for baseline validation.

Users: ─────────────────────── (constant)
Time:  0          30s         60s

Ramp-Up

Gradually increase users to find the saturation point without shocking the system.

Users:        ___________
             /
            /
Time:  0  ramp  sustained

Spike

Sudden jump and drop. Tests system's ability to handle and recover from bursts.

Users:    |
          |
──────────┘└────────────
Time:  0  spike  0

Step Load

Increase users in discrete steps. Each step stabilizes before the next increment.

Users:          ___
          _____|
    _____|
___|
Time:  0   step1  step2  step3

---

Think Time

Think time is the delay between a user's actions — simulating reading, scrolling, filling forms.

Without think time: - Unrealistic burst load - 100 users generate 10,000+ RPS - Numbers don't reflect production

With realistic think time: - 100 users may generate 50–200 RPS (realistic) - Matches actual user behavior - Results are comparable to production monitoring

from locust import between, constant, constant_throughput

wait_time = between(1, 5)            # random: 1–5 seconds
wait_time = constant(2)              # always 2 seconds
wait_time = constant_throughput(2)   # maintain 2 tasks/sec/user

Choose constant_throughput when you need to target a specific RPS regardless of how long individual requests take.