Indirection/abstraction over `tokio::spawn` and friends

[matheus23]

Iroh's use of tokio is very particular/opinionated.

• No use of tokio::spawn without wrapping the JoinHandle with AbortOnDropHandle
• No use of JoinSet::spawn without some loop somewhere calling tasks.join_next().await, if !tasks.is_empty() (to avoid leaking memory from task results accumulating) (EDIT: this might only be necessary if used with else inside tokio::select!, see refactor(iroh, iroh-relay): JoinSet disabling in tokio::select! #3052 )
• Every spawn must use an .instrument() for tracing.

Furthermore, for iroh in the browser (#2799) we will need to abstract away all tokio::spawn usage because it's not available in the browser! (I've already implemented a JoinSet alternative on top of wasm_bindgen_futures for exactly that reason)

Given that such an indirection would have to be created for the browser work anyways, this might be a good opportunity to introduce an API at that indirection that avoids all of the above footguns (forgetting .instrument(), leaking tasks, leaking memory).

[rklaehn]

Iroh's use of tokio is very particular/opinionated

sane is the word you are looking for...

[matheus23]

Perhaps worth keeping in mind that AbortOnDropHandle isn't the be-all end-all of task leaking: n0-computer/iroh-gossip#20

I highly doubt it but perhaps we should spend a couple of minutes thinking about if there's a good solution for this problem once we get to this issue.

[TotalKrill]

From what I can see this is now the last issue to get webassembly support for iroh-net as outlined in #2799

Is there a plan to do this? :)

[matheus23]

@TotalKrill happy to see people so interested in this :)

As I wrote in the issue:

This is a non-exhaustive list of things that need to land for Wasm support:

So even if this lands, there will be more work required. It's just us keeping track of things to make sure we have a rough overview.

But to answer the actual question, yes, you can see the module that implements this indirection in the matheus23/iroh-browser branch here:

• https://github.com/n0-computer/iroh/blob/matheus23/iroh-browser/iroh-net-report/src/task.rs

If you want to, you're free to depend on this branch yourself and start to play with iroh in browsers - it works with the default public relays we have deployed as well. You can see me playing and testing that branch over here as well: https://github.com/matheus23/browser-iroh-net-demo/blob/update/Cargo.toml

Possibly very soon we'll have a demo branch that's easier to set up for all folks interested to play with.

[matheus23]

This was partially addressed in #3156

[dignifiedquire]

closing this, as the critical parts are done, if we have concrete ideas how to improve this we can open a new issue