moto s3 using AWS_ENDPOINT_URL from .env

In my Python project, I'm uploading files to AWS S3 using their boto3 library, with code that starts like:

client = boto3.client("s3", region_name=region, endpoint_url=endpoint_url)

I'm providing here the endpoint_url, as locally this points to LocalStack configured through an environment variable AWS_ENDPOINT_URL in my .env file.

To unit test this I'm using the moto library, which helps with mocking AWS services. I have a unit test method decorated with the @mock_s3 attribute and inside it I'm first creating a bucket, as per their documentation:

s3 = boto3.resource("s3")
s3.create_bucket(Bucket="test-bucket")

Because of the mocking attribute, this code is actually intercepted by moto.

I noticed that this unit test sometimes fails. Investigating this, I realized that it fails when my LocalStack Docker container is stopped, which lead me to the fact that the code is actually trying to create the bucket using the environment variable AWS_ENDPOINT_URL that I have configured in my .env file.

Now that's a serious problem, as I want the unit test to be independent and NOT use that AWS_ENDPOINT_URL.

After some debugging and things that I tried, I managed to fix this by setting that environment variable to empty string from the unit test code, so before the code above I added:

os.environ["AWS_ENDPOINT_URL"] = ""

moto do have advice along these lines in their documentation: https://docs.getmoto.org/en/latest/docs/getting_started.html#how-do-i-avoid-tests-from-mutating-my-real-infrastructure

, but this variable, AWS_ENDPOINT_URL, is not covered there.