[🐛 Bug]: Ensuring Drivers close

[titusfortner]

Description

Selenium closes the driver as part of the quit method, but it is possible for Selenium to start the driver, but error when creating the session, which prevents the driver process from being closed.

I see this for sure in Ruby, but I think this applies to other languages as well, so we should check.

Reproducible Code

n/a

Debugging Logs

n/a

[cgoldberg]

Python handles this correctly. If you try to instantiate a browser and the session is not created, it will raise SessionNotCreatedException and stop the driver process.

For example, the following code will raise an exception, stop execution, and no chromedriver processes will remain:

from selenium import webdriver
from selenium.webdriver.chrome.options import Options

options = Options()
options.add_argument("--user-data-dir=/no/such/location")
driver = webdriver.Chrome(options=options)

[titusfortner]

self.service.stop() is only called from the quit() method of each driver class. If the driver isn't instantiated, quit can never be called, and that method isn't explicitly executed. It's possible the system might kill child processes when the python process ends? Actually, I should figure out why that isn't happening in Ruby. But I think this is the issue with #15632 and I think it makes sense to explicitly call self.service.stop() when the self.start_session(capabilities) has a session creation error.

[cgoldberg]

It's possible the system might kill child processes when the python process ends?

Not the system, but our code is killing the child processes when the Python interpreter exits.

There is a __del__ method in the selenium.webdriver.common.service.Service class that all drivers use. When the Python interpreter exits, it calls __dell__ on every object. This method calls self.stop(), just like driver.quit() does, so all the child processes get killed.

I think it makes sense to explicitly call self.service.stop() when the self.start_session(capabilities) has a session creation error.

Now that I think about it more, that makes sense. If you continuously create drivers, and handle the exception without exiting the program (like a test framework might do), the drivers wouldn't get cleaned up until the end.

I'll see if I can add that.

[cgoldberg]

I think this is all that is needed for Python: #15636

[cgoldberg]

This is completed for Python via #15636