Getting a token takes nearly a second

[AzureIP]

Hi
I am trying out this library with Django 5 and python 3.12.
It works as expected, but getting a token for a user takes nearly a second. Around 950ms. Why? Is there some setting that improoves the generation time?

[mohamed666666]

up

[aperezlillo]

This is probably caused by https://forum.djangoproject.com/t/stop-increasing-default-pbkdf2-iteration-count/25539, if you change the password hasher or reduce the steps this will reduce the performance (but you will require to recreate all hashes... and theoretically the application will be less secure)

[ThirVondukr]

@aperezlillo Affecting password security to gain performance is a bad idea.
I recently stumbled upon pyjwt taking 25ms when creating a JWT, compared to around 1ms in rust/go. Fixing this could be a single line change in some cases - you have to manually load your key in case you're using RSA:

import time

import jwt
from cryptography.hazmat.primitives import serialization
from cryptography.hazmat.primitives.asymmetric import rsa

key = rsa.generate_private_key(public_exponent=65537, key_size=2048)
key_str = key.private_bytes(
    encoding=serialization.Encoding.PEM,
    format=serialization.PrivateFormat.PKCS8,
    encryption_algorithm=serialization.NoEncryption(),
)

# key = serialization.load_pem_private_key(key_str.encode(), password=None)

t = time.perf_counter()
tok = jwt.encode(payload={}, key=key, algorithm="RS256")
print((time.perf_counter() - t) * 1000)

t = time.perf_counter()
tok = jwt.encode(payload={}, key=key_str, algorithm="RS256")
print((time.perf_counter() - t) * 1000)

[ThirVondukr]

I ran some tests, while majority of time is spent on checking if password is valid:

In [4]: %timeit hashlib.pbkdf2_hmac("sha256", b"abcde", b"abcde", 600_000)
153 ms ± 665 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

generating a new access token for user could be made faster:

   ...: 
   ...: import jwt
   ...: from cryptography.hazmat.primitives import serialization
   ...: from cryptography.hazmat.primitives.asymmetric import rsa
In [3]: key = rsa.generate_private_key(public_exponent=65537, key_size=2048)
   ...: key_str = key.private_bytes(
   ...:     encoding=serialization.Encoding.PEM,
   ...:     format=serialization.PrivateFormat.PKCS8,
   ...:     encryption_algorithm=serialization.NoEncryption(),
   ...: )
   ...: 
In [4]: %timeit jwt.encode(payload={}, key=key, algorithm="RS256")
181 µs ± 6.21 µs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)
In [5]: %timeit jwt.encode(payload={}, key=key_str, algorithm="RS256")
23.9 ms ± 152 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

[aperezlillo]

Hi @ThirVondukr, thx for your answer, I was just answering the original question.

I agree with you that changing this configuration is not recommended, but when you know the root cause of the problem you can make trade-offs, or as you made, trying to improve that piece of code. As you figured out, the main problem is with the password validation, not the token creation that is almost negligible. Depending on the hardware pbkdf2_hmac can take up
to 1 second and depending on the use case this could be too much.

Not sure how to address the main issue, could be worth to investigate alternative hashing libraries, or configurations.

[ThirVondukr]

@aperezlillo I don't think you really can address the issue without compromising some of the security - if it's easier to validate password hash then it would be easier to crack it.
Is this really that big of an issue? User should really only login once and then his access token is simply refreshed.
By the way - I don't think a 20ms difference is negligible there (when comparing using a str or already checked token for signing a jwt)