Add caching to JWT decoding

[moritz89]

The decoding of JWTs is 'relatively' expensive and called about 5 times per request. Caching the request results in almost a 10x speed up. Caching the call also reduces the total call by almost 20%. Therefore, it would be impactful to add a cache to the function. In order not to cache an expired token, the cache would be given a time-to-live (TTL). Here, even a 60 second cache time would have significant benefits.

I would propose adding the ttl_cache decorator from cachetools.

Graph of the cache timings:

Zoomed graph without the first requests:

Data and graph source:

timings.ods

• Note about the missing first entry. Loading the JWKS from Keycloak takes about 10 ms and distorts the actual measurement which is why it was removed.

The timings were recorded with the following function:

from functools import wraps
from time import time

def timing(f):
    @wraps(f)
    def wrap(*args, **kw):
        ts = time()
        result = f(*args, **kw)
        te = time()
        print(f"took: {(te - ts) * 1000} ms")
        return result

    return wrap