Logging/Debugging failed requests

[nicholaswmin]

Summary

Loading a custom model but a little unsure about the configuration; =
point is I have no way of debugging this w/o looking at the failed request's parameters.

or maybe a zed --debug that prints to STDERR

  "language_models": {
    "openai": {
      "version": "1",
      "api_url": "https://api-inference.huggingface.co/v1/",
      "available_models": [
          {
            "name": "codellama/CodeLlama-34b-Instruct-hf",
            "display_name": "CodeLlama-34b-Instruct",
            "max_tokens": 128000
          }
        ]
    }
  },

  "assistant": {
    "enabled": true,
    "default_model": {
      "provider": "openai",
      "model": "codellama/CodeLlama-34b-Instruct-hf"
    },
    "version": "2",
    "button": true,
    "default_width": 480,
    "dock": "right"
  },

Actual Behavior:

• displays 404 NOT Found.
• logs nothing

Expected Behavior:

• displays 404 NOT Found.
• logs the failed request URL, HEADERS and body

[notpeter]

I haven't ever used Zed via api-inference.huggingface.co perhaps someone else can chime in with their experience.

Can you try removing the trailing slash from your api_url?

The version keys are only for zed settings and have no relation to the v1 in the url.

[nicholaswmin]

@notpeter I've changed the gist of this thread to focus around the lack of error logging,
does it make sense to reocnvert it to an issue again?

[ViggoC]

@nicholaswmin Mac M4
And you have mistaken the positions of "expected" and "actual".

I think this is worth reconverting into an issue. Are you interested in fixing this issue?

[nicholaswmin]

I got carried away with building a MITM CLI that logs requests; i got an m2 and got it working and was gonna ask if you want it, completely forgetting that your first post was a fix suggestions :) oh well..

I'd love to - but I've never touched Rust
so no

[ViggoC]

@notpeter I added a http client wrapper to log the failed request. Could you take a look? #25851.

[ViggoC]

@maxdeviant Thanks for you feed back in #25851. Do you think it's a good idea to add logging in each module that initiates requests? This would make modifications more dispersed, but each module could independently control which information to log.

[DerEchteKoschi]

Hi!
Has anyone have a temporary solution for this? I, too, am trying to integrate a custom model and it is quite difficult without having the raw request data. I cannot simply get the logs from the servers and the only other way i found is using a proxy, which is a pain in the neck to set up as we're using TLS encryption.
For me, logging all HTTP requests but only when a certain variable, lets say "DEBUG_HTTP: VERBOSE" is set, would absolutely be an option without cluttering all the logs with all requests.
Everything I could find up to now was just logging the error message, but not the actual request.

[notpeter]

the only other way i found is using a proxy, which is a pain in the neck to set up as we're using TLS encryption.

I wonder whether it would be useful to create documentation for how to do this or a script to make it easy. In a previous job, we had a script which made it possible to launch the application with a MITM proxy https unwrapping proxy and made sure the fake CA was exported in the correct places (node_extra_ca_certs, etc).

For me, logging all HTTP requests but only when a certain variable, lets say "DEBUG_HTTP: VERBOSE" is set, would absolutely be an option without cluttering all the logs with all requests.

I'd like to focus on the dev-build use case at first. I would like to avoid the footgun where you temporarily set an environment variable and later you find out your Stable release daily-driver Zed has been logging requests/secrets to disk for months.