Test operating capacity
This guide addresses how to test performance by simulating multiple users accessing an app concurrently.
Load testing means testing the performance of the entire Panel application and the server(s) running it.
This kind of testing is really useful if you want to
Before you get started ensure you have installed the required dependencies:
pip install panel pytest locust pytest-playwright pytest-asyncio loadwright
and ensure playwright sets up the browsers it will use to display the applications:
Create the app
Lets create a simple data app for testing. The app sleeps 0.5 seconds (default) when loaded and when the button is clicked.
Create the file app.py and add the code below:
:::{card} app.py
import time
import panel as pn
import param
class App(pn.viewable.Viewer):
run = param.Event(doc="Runs for click_delay seconds when clicked")
runs = param.Integer(doc="The number of runs")
status = param.String(default="No runs yet")
load_delay = param.Number(default=0.5)
run_delay = param.Number(default=0.5)
def __init__(self, **params):
super().__init__(**params)
result = self._load()
self._time = time.time()
self._status_pane = pn.pane.Markdown(self.status, height=40, align="start", margin=(0,5,10,5))
self._result_pane = pn.Column(result)
self._view = pn.Column(
pn.Row(pn.widgets.Button.from_param(self.param.run, sizing_mode="fixed"), self._status_pane),
self._result_pane
)
def __panel__(self):
return self._view
def _start_run(self):
self.status = f"Running ..."
self._time = time.time()
def _stop_run(self):
now = time.time()
duration = round(now-self._time,3)
self._time = now
self.runs+=1
self.status=f"Finished run {self.runs} in {duration}sec"
@param.depends("run", watch=True)
def _run_with_status_update(self):
self._start_run()
self._result_pane[:] = [self._run()]
self._stop_run()
@param.depends("status", watch=True)
def _update_status_pane(self):
self._status_pane.object = self.status
def _load(self):
time.sleep(self.load_delay)
return "Loaded"
def _run(self):
time.sleep(self.run_delay)
return f"Result {self.runs+1}"
if pn.state.served:
pn.extension(sizing_mode="stretch_width")
App().servable()
:::
Serve the app via panel serve app.py and open http://localhost:5006/app in your browser.
The conftest.py file should be placed alongside your tests and will be loaded automatically by pytest. It is often used to declare fixtures that allow you declare reusable components. It will:
Create the file conftest.py and add the code below.
:::{card} conftest.py
"""Shared configuration and fixtures for testing Panel"""
import panel as pn
import pytest
PORT = [6000]
@pytest.fixture
def port():
PORT[0] += 1
return PORT[0]
@pytest.fixture(autouse=True)
def server_cleanup():
"""
Clean up server state after each test.
"""
try:
yield
finally:
pn.state.reset()
:::
For more inspiration see the Panel conftest.py file
Test the initial load with Locust
Locust can help you test the behaviour of users that load (i.e. request) your Panel app. Locust provides many useful performance related statistics and charts.
Create the file locustfile.py and add the code below.
:::{card} locustfile.py
from locust import HttpUser, task
class RequestOnlyUser(HttpUser):
@task
def goto(self):
self.client.get("/app")
:::
Start the Panel server:
Start the Locust server
locust --host http://localhost:5006
Open http://localhost:8089. Keep the default settings and click the Start swarming button. This should look like the below:
The median response time is ~530ms when one user requests the page every second. If you try to increase to 10 simultaneous users you will see a median response time of ~5300ms. If this is a likely scenario, you will have to look into how to improve the performance of your app.
Test advanced interactions with Loadwright
Loadwright is a young load testing framework built on top of Playwright and Panel.
Lets replicate a user that:
Create the file test_loadwright.py and add the code below:
:::{card} test_loadwright.py
import param
import pytest
from loadwright import LoadTestRunner, LoadTestViewer, User
import panel as pn
from app import App
HEADLESS = False
class LoadAndClickUser(User):
"""A custom User that loads a page and clicks a button n_clicks times"""
n_clicks = param.Integer(
default=1, bounds=(0, None), doc="The number of times to click the button"
)
async def run(self):
with self.event(name="load", user=self.name):
await self.page.goto(self.url)
await self.page.get_by_role("button", name="Run").wait_for()
await self.sleep()
for click_index in range(self.n_clicks):
with self.event(name="interact", user=self.name):
await self.page.get_by_role("button", name="Run").first.click()
await self.page.get_by_text(f"Finished run {click_index+1}").wait_for()
await self.sleep()
@pytest.mark.asyncio
async def test_component_2(port=6001):
"""We can run the LoadTestRunner with 5 users each clicking 5 times"""
async with LoadTestRunner.serve(App, port=port) as host:
runner = LoadTestRunner(host=host, headless=HEADLESS, user=LoadAndClickUser(n_clicks=5), n_users=5)
await runner.run()
# You can access the data of the run via runner.logger.data
assert runner.logger.data.duration.mean()<=10
if pn.state.served:
import panel as pn
pn.extension(sizing_mode="stretch_width")
viewer = LoadTestViewer(data="test_results/loadwright.csv")
pn.template.FastListTemplate(
site="LoadWright",
title="Load Testing with Playwright and Panel",
main=[viewer]
).servable()
:::
Run the tests with pytest
pytest test_loadwright.py
View the results with Panel
$ panel serve test_loadwright.py
...
2023-01-01 15:53:03,396 Bokeh app running at: http://localhost:5006/test_loadwright
.
You will find an archive of test results in the tests_results/archive folder.
