This shows that storage instance startup times for S3 are much higher than other backends. This is due to some connection resource setup that the S3boto backend does.

s3-storage-startup-timing.py

import random
import string
import timeit

from django.conf import settings
from django.core.files.storage import get_storage_class


iterations = 100
path = "pdfs/docs/latest/docs.pdf"
storage = get_storage_class(settings.RTD_BUILD_MEDIA_STORAGE)()


def test_const_only():
    return get_storage_class(settings.RTD_BUILD_MEDIA_STORAGE)()


def test_url_only():
    new_path = ''.join([random.choice(string.ascii_lowercase) for _ in range(30)])
    return storage.url(new_path)


def test_const_plus_url():
    storage = get_storage_class(settings.RTD_BUILD_MEDIA_STORAGE)()
    return storage.url(path)


# ~sub 0.1ms per iteration
iter_time = timeit.timeit(test_const_only, number=iterations) / iterations
print(f"Constructor Only: {iter_time:.6f} seconds")

# ~1.5ms per iteration -- actually the first iteration is expensive and others are much faster
iter_time = timeit.timeit(test_url_only, number=iterations) / iterations
print(f"URL Calculation Only: {iter_time:.6f} seconds")

# ~50ms per iteration
iter_time = timeit.timeit(test_const_plus_url, number=iterations) / iterations
print(f"Constructor + URL: {iter_time:.6f} seconds")