Cypress interceptors slower than hitting the actual backend

[johannesstricker]

Hey 👋
In an effort to improve our Cypress test runtime, we experimented with a dynamic request cache. It works as follows:

1. Unique requests are identified by full url and method
2. The first time a request occurs, we hit the backend and save the response in memory or as a fixture file (tried both)
3. Whenever we find a request that has already been cached, we return the cached response without hitting the BE

For (2) we've tried different cache mechanisms:

1. Save the response in an in-memory cache and use interceptor functions that read the response from the in-memory cache.
2. Save the response as a fixture to disk and then, for each fixture that was found on disk, set up an interceptor that returns the fixture.

No matter what we tried, our cache solutions runs even slower than without using the cache at all. In the headful Cypress runner, we can see that the requests get intercepted and don't hit the BE (cause the green circle next to the request log is not filled). Still, in the network tab we can see that the requests take between 50ms and 150ms, which sounds like an eternity, given that it's coming from the cache.

We have a lot of outgoing requests, so I understand there is some overhead for Cypress to set this up. But I would still expect the network latency to be much worse, especially cause there are a lot of external requests involved, that don't hit localhost.

Does anyone have a clue what could be going on here?