🐛 BUG: Uncaught exception

[eloff]

Which Cloudflare product(s) does this pertain to?

Wrangler or miniflare

What version(s) of the tool(s) are you using?

3.5.0

What version of Node are you using?

18.16.1

What operating system are you using?

Linux (Ubuntu 20.04)

Describe the Bug

In one of my handlers, after executing a select query with batch, I get the following exception:

workerd/server/server.c++:2533: error: Uncaught exception: kj/async-io-unix.c++:186: disconnected: remote.worker_do_not_log; Request failed due to internal error

It seems it's in some kind of destructor or finalizer or other thread / cleanup task, since execution continues after the query, because I can see console.log output from subsequent code. However, the server crashes before it is able to return the response (client doesn't get a response back.) I can reproduce this every single run. Commenting out that query makes the exception go away. No other queries are executed in that request.

I deleted, the database, deleted .wrangler dir, and recreated the database. Same issue.

I created a new hello world project (following the getting started guide), containing only the "hello world" response and got the same exception on the first run only. Subsequent runs were fine. That's interesting, what state could wrangler be keeping that's the same for the machine, but not in the directory/project/code which all changed?

I switched back to the original project, reproduced the exception again. Went back to the new hello world, but it's fine now, I can't reproduce the exception in the hello world project again. Odd.

I then added the same database, same migrations, same query to the hello world project. No issues. I don't know how to make a minimal repro for it.

edit: I tried again this morning, after rebooting overnight. The error is gone, the reboot fixed it. It wasn't the query at fault here, probably not d1 (directly anyway). There's something else going on, and somehow a reboot fixed it.

It's tough to debug a problem with so little to go on. Is there anything I can do to get more info? Is there a way to build a debug version of wrangler or miniflare? Enable logging?

Please provide a link to a minimal reproduction

No response

Please provide any relevant error logs

No response

[aroman]

i get the identical error in my workers project, and i don't use d3. running on macOS 13.5 with wrangler 3.5.0. also filed here: #1010

[eloff]

I tried again this morning, after rebooting overnight. The error is gone, the reboot fixed it. It wasn't the query at fault here, probably not d1 (directly anyway). There's something else going on, and somehow a reboot fixed it.

[jorgemartins-uon]

I'm having the same issue:

[chitalian]

I am getting this issue when quickly closing a connection to my worker

[penalosa]

I've transferred this to workerd since it looks like it's an error log coming from workerd rather than Wrangler directly

[jasnell]

We'll need a simple reproduction example in order to diagnose this.

[SyntaxColoring]

I'm not sure if this is exactly the same thing:

workerd/server/server.c++:3069: error: Uncaught exception: kj/compat/http.c++:3939: disconnected: worker_do_not_log; Request failed due to internal error

(Note kj/compat/http.c++:3939 instead of kj/async-io-unix.c++:186, which the OP reported.)

@jasnell I can reproduce this, but sometimes it requires patience.

1. Clone https://github.com/SyntaxColoring/cloudflare-crash-repro.
2. npm install && npm run dev.
3. In something like Postman, open a bunch of concurrent WebSocket connections to localhost:8787. I opened a few to localhost:8787 and a few to localhost:8787/some/other/path.
4. Send a few messages on each connection.
5. Wait a few minutes. The error will probably happen on its own. Some (but not necessarily all?) of the WebSocket connections will disconnect. If it doesn't happen after a few minutes, send some more WebSocket messages and wait some more.

Here are the associated log files from one reproduction.

wrangler-2024-01-14_18-07-47_964.log

[gaberogan]

I was able to reproduce this using a bare minimum worker template with fetch('https://example.com').

Strangely it seems to work fine with fetch('http://ip-api.com/json'), so I think it's an issue with SSL.

Here is the reproduction: https://github.com/cloudflare/workers-sdk/files/14214588/test-worker.zip

I am using wrangler 3.28.0 and macOS 13.5 and node.js 18, but I don't think this matters because I've experienced this reliably on both my windows and mac computers, with different sdk versions and different node.js versions

Last week I was not experiencing this and I have no idea why..

cc @jasnell

[gaberogan]

Can we please rename this issue to something more useful? Like "internal error" on fetch with HTTPS

cc @jasnell

[akainocode]

I would like to confirm this issue, but it seemed to resolve with a restart of my machine.

I am currently using WSL2 though, that might have an affect on it but above people have mentioned that it happens regardless of OS.

[robmv]

We'll need a simple reproduction example in order to diagnose this.

I am new to Cloudfare Workers, so I am starting the Rust example project, so I think I have the simplest reproduction:

1. Install the Rust example npx wrangler generate hello-world-rust https://github.com/cloudflare/workers-sdk/templates/experimental/worker-rust
2. Run npx wrangler dev
3. Open the example on the browser with [b] option.
4. When you get the Hello, World! output, just press Control+R to reload the browser without releasing the keys in order to reload indefinitely.

The same error presented here, with different line number, maybe it isn ´t the same case but I find it weird that a fast and simple REST endpoint like this could fail so easily.

 [ERROR] workerd/server/server.c++:3340: error: Uncaught exception: kj/async-io-unix.c++:778: disconnected: remote.worker_do_not_log; Request failed due to internal error`