Upstreaming libp2p broke NodeJS support of the web-client

[sisou]

General information

• Library version(s): >= 0.19.0
• Browser version(s): NodeJS
• Devices/Simulators/Machine affected: all
• Reproducible in the testnet? (Yes/No): yes
• Related issues: feat(websocket-websys): add support for different WASM environments libp2p/rust-libp2p#4889, network-libp2p: Go back to upstream libp2p #2047

Bug report

Expected behavior

NodeJS can run the client in a worker_thread, for which I created a special entry point.

Actual behavior

The client in a NodeJS worker thread panics with 'to have a window or worker context', which is the exact panic added in libp2p/rust-libp2p#4889.

Steps to reproduce

Run the NodeJS example in the web-client/example/node. Make sure to follow the instructions in web-client/example/README.md.

Crash log? Screenshots? Videos? Sample project?

2024-03-08T20:11:56.006000128Z ERROR panic                | thread '<unnamed>' panicked at 'to have a window or worker context': /home/soeren/.cargo/registry/src/index.crates.io-6f17d22bba15001f/libp2p-websocket-websys-0.3.1/src/lib.rs:308 

Question or Feature Request

Add support for NodeJS Worker Threads to libp2p's websocket-websys transport?

I need to find out how it worked before libp2p was upstreamed in #2047, as it did work before, when I added NodeJS support and the example.

[sisou]

Before the change to WindowOrWorkerContext, the websocket-websys transport was patched to bind to global setTimeout and clearTimeout directly, without having to have web-sys's Window or WorkerContext: jsdanielh/rust-libp2p@34f65f1

[sisou]

I'm able to pass the first initialization check of the client by mocking WorkerGlobalScope in the NodeJS worker thread:

global.WorkerGlobalScope = global;

However, the client still exits with an error later on, after requesting/receiving ZKPs. On my second run, it actually established consensus and then errored and exited. Unfortunately the error is not very descriptive, even when run in vscode's debugger:

 Uncaught Error Error
     at <anonymous> (node:internal/event_target:1062:28)
     at processTicksAndRejections (node:internal/process/task_queues:77:11)
     --- TickObject ---
     at init (node:internal/inspector_async_hook:25:19)
     at emitInitNative (node:internal/async_hooks:200:43)
     at emitInitScript (node:internal/async_hooks:503:3)
     at nextTick (node:internal/process/task_queues:132:5)
     at emitUncaughtException (node:internal/event_target:1062:11)
     at [nodejs.internal.kHybridDispatch] (node:internal/event_target:794:9)
     at exports.emitMessage (node:internal/per_context/messageport:23:28)
     at callbackTrampoline (node:internal/async_hooks:128:17)
     --- MESSAGEPORT ---
     at init (node:internal/inspector_async_hook:25:19)
     at emitInitNative (node:internal/async_hooks:200:43)
     at Worker (node:internal/worker:207:21)
     at <anonymous> (file:///home/soeren/nimiq/core-rs-albatross/web-client/example/node/node_modules/@nimiq/core-web/nodejs/index.js:15:11)
     at create (file:///home/soeren/nimiq/core-rs-albatross/web-client/example/node/node_modules/@nimiq/core-web/lib/node/client-proxy.js:28:22)
     at main (file:///home/soeren/nimiq/core-rs-albatross/web-client/example/node/index.js:7:39)
     at <anonymous> (file:///home/soeren/nimiq/core-rs-albatross/web-client/example/node/index.js:15:1)
     at Module._compile (node:internal/modules/cjs/loader:1356:14)
     at Module._extensions..js (node:internal/modules/cjs/loader:1414:10)
     at Module.load (node:internal/modules/cjs/loader:1197:32)
     at Module._load (node:internal/modules/cjs/loader:1013:12)
     at executeUserEntryPoint (node:internal/modules/run_main:128:12)
     at <anonymous> (node:internal/main/run_main_module:28:49)
 Process exited with code 1

[Eligioo]

It turns out the uncaught error that is thrown and mentioned by @sisou is closure invoked recursively or after being dropped meaning that on the 'JS-side' a closure (probably) is being called that already has been dropped. In NodeJS this is causes an uncaught exception, which makes sense. But nowadays NodeJS actually exits the process on uncaught exceptions.

So this is unrelated to requesting/receiving zkps but rather connections that aren't cleaned up properly.

[sisou]

I have created a PR in libp2p to fix part of the closure invoked after being dropped problem: libp2p/rust-libp2p#5229. The main problem remains, though.

[sisou]

Workaround was merged in #2302.