'logging' example is not provided; no pattern provided

[sgharms]

Hello,

The https://nextjs.org/docs/middleware#examples page suggests that there are number of examples of various types in this repo

Authentication
Bot protection
Redirects and rewrites
Handling unsupported browsers
Feature flags and A/B tests
Server-side analytics
Advanced i18n routing requirements
Logging

all point to the same resource:

https://github.com/vercel/examples/tree/main/edge-functions

Keying in on Logging, searching the Readme for the string oggi yields no match. Similarly ack through the repo for /oggi/i and you will only find a mention in feature-flag-configcat/ example. Furthermore, this is logging from a client function not from a server function e.g. the middleware function.

Absent any directions, I ahem, duplicated this error vercel/next.js#13214

Further research at https://nextjs.org/docs/api-reference/edge-runtime#unsupported-apis, a leap whose connection was made my reading URL paths, suggests this is not possible. OK, I can live with that, but then it becomes all the more important that it be documented how to log server-side data in the middlewares.

I suspect most people are implementing a custom server to get 'fs' access, a practice that seems discouraged in the docs.

So is there either:

1. An example of doing server-side logging in the Next Middleware (if not...)
2. The expectation that that should be possible and, if so, how will that be possible if fs and other core modules are not available (e.g. process, util, os, buffer, et al.)

[oswaldoacauan]

I had a very frustrating evening yesterday trying to find a simple logging middleware example. The docs pointing nowhere was not very helpful also.

I see this as a basic feature for a framework that is to be production ready and same goes for Telemetry which there is no examples to be found.

[okbel]

Hey! Thanks for bringing this up. We're already working on it. 👍

[oswaldoacauan]

@okbel thanks for the update 🙏🏻 . Is there any estimated date for the logging, telemetry and instrumentation examples?

[leerob]

#156

[sgharms]

I think @leerob has offered something formidable, and it is definitely an improvement that should be included; however, for simplicity’s sake and to establish a general pattern, I’d like to see something where the log file is written on the same machine as the next instance.

I’m wondering if we can squint and see the Vercel business model: if calls are processed on lambdas, there’s no local persistence and an approach like #156 makes sense :). I’d argue that for developing middleware and dev mode in general that models for local persistence be demonstrated (or, I suppose, clearly rejected in the documentation so that those considering next middleware see this limitation up front).

[leerob]

I’d like to see something where the log file is written on the same machine as the next instance.

This isn't a pattern we want to advocate for, whether you're running serverfull or serverless.

I’m wondering if we can squint and see the Vercel business model: if calls are processed on lambdas, there’s no local persistence and an approach like #156 makes sense.

Next.js works open-source, on whatever infra you provide, so it's not really about Vercel's serverless infra - more so that when we talk to companies building with Next.js, they aren't persisting logs to a local file, but rather ingesting them into some logging system 👍

[ValentinH]

Talking about ingesting them, we are facing issues with winston not being able to run in the Edge runtime. Is it a known issue? What logger do you recommend for logging in middlewares?

[sgharms]

@ValentinH

We were unable to log (locally) in the middleware as fs is not available.

The PR suggests that fetch() is available as an alternative.

For us, this meant using a custom server (in docs) to get this functionality. It’s discouraged by vercel as one loses some optimizations, but I didn’t see any other way to get local logging.

good news: the server code is fairly easy and the docs are solid.

[ValentinH]

I've gone through the custom server path in the past but this time we are hosted on Vercel so we want to stay on the default setup. 😞

[sgharms]

Seems like there is community desire for access to some/all of the node runtime in the middleware.

Having to fetch() POST just to get basic telemetry data seems unnecessarily complex.

[lfades]

Closing as per #164 (comment)

To provider more context, Middleware can be executed multiple times for the same path, and we don't have a reliable way to send logs at 1. the right time and 2. while avoiding sending it twice.

The logging and analytics example have been removed until we have a more appropriate solution for it.