feat(iroh-willow): Event subscriptions

[Frando]

Description

• Moves event subscription from a wrapper of the store into the EntryStorage trait
• Adds persistent subscriptions, with interface - mostly - from Store trait earthstar-project/willow-rs#21
• Exposes subscriptions over RPC
• Adds client methods for subscriptions
• Adds a smoke test

Notes & open questions

Should look into remaining differences between our trait and willow-rs with regard to event subscriptions, and see if anything from here should be proposed for change in the willow-rs trait. Notably:

• subscribe_area takes SubscribeOpts not QueryOpts, with option to ignore ingested event for some session id, and to receive only Ingested event. This lets me replace the previous impl easily in the WGPS impl, and it makes sense IMO.

Has a few TODOs on optimizing things.

Will need much more tests.

Replaces #2681 (had wrong branch)

Change checklist

• Self-review.
• Documentation updates following the style guide, if relevant.
• Tests if relevant.
• All breaking changes documented.

[github-actions]

Documentation for this PR has been generated and is available at: https://n0-computer.github.io/iroh/pr/2682/docs/iroh/

Last updated: 2024-09-09T17:31:26Z

[matheus23]

I think I've played with this enough in tauri-todos to confirm the subscription bits are working. Still need to take a look at the actual code, but honestly, that'll take a bit until I warm up to the codebase generally.

[matheus23]

This kind of forwarding of events seems slightly weird to me. Although I guess it's either this or more invasive changes to the trait Storage/trait EntryStorage (allowing you to just give it a sender instead of it returning a stream).

[Frando]

I agree that this is not ideal. We do similar things already currently in the RPC boundary for old iroh-docs.
Maybe it makes sense for the subscribe methods on the store trait to take in a sender instead of returning a stream.
This brings up the question though which trait it should be generic over: Sink? ufotofu::Consumer? a custom trait that has only async fn send(item: Event)?

[Frando]

Added a comment for now.

[matheus23]

Yes these are really good TODOs. Also need this on PathExt and EntryExt in the data_model.rs file.

[matheus23]

This test fails on this branch, but works in the willow branch.
It's waiting on syncs.complete_all().await, which never completes (neither with SessionMode::ReconcileOnce nor SessionMode::Continuous).
The test succeeds if we share the ticket using RestrictArea::None instead of RestrictArea::Restrict(...).

Suggested change

	#[tokio::test]
	async fn test_regression_restricted_area_sync() -> testresult::TestResult {
	iroh_test::logging::setup_multithreaded();
	const TIMEOUT: std::time::Duration = std::time::Duration::from_secs(2);
	let (alfie_addr, alfie) = spawn_node().await;
	let (betty_addr, betty) = spawn_node().await;
	info!("alfie is {}", alfie_addr.node_id.fmt_short());
	info!("betty is {}", betty_addr.node_id.fmt_short());
	let alfie_user = alfie.spaces().create_user().await?;
	let betty_user = betty.spaces().create_user().await?;
	let alfie_space = alfie
	.spaces()
	.create(NamespaceKind::Owned, alfie_user)
	.await?;
	let space_ticket = alfie_space
	.share(
	betty_user,
	AccessMode::Write,
	// RestrictArea::None, // succeeds with this
	RestrictArea::Restrict(Area::new_subspace(betty_user)),
	)
	.await?;
	let (betty_space, syncs) = betty
	.spaces()
	.import_and_sync(space_ticket, SessionMode::ReconcileOnce)
	.await?;
	let completion = tokio::time::timeout(TIMEOUT, syncs.complete_all()).await?;
	println!("Completed syncs: {completion:#?}");
	let stream = betty_space.get_many(Range3d::new_full()).await?;
	let entries: Vec<_> = stream.try_collect().await?;
	println!("{entries:#?}");
	Ok(())
	}

[matheus23]

I've also seen Space::sync_once just hang when two nodes try to sync that have never exchanged capabilities.

[Frando]

I think the test failed to pass because this branch was missing a commit from main: #2683
Or at least the test passes just fine for me after rebasing this branch on willow!

[Frando]

I've also seen Space::sync_once just hang when two nodes try to sync that have never exchanged capabilities.

I think this is currently expected behavior: PAI waits forever for matching capabilities to be submitted. We need to close our side after we have submitted everything I think. Aljoscha made some recent modifications to WGPS to support this properly (I still have to check them out)

[Frando]

The test passes reliably for me.