Flaky optimism p2p::can_sync test

[Rjected]

A recent merge queue run failed in a way that indicates we have something flaky in p2p::can_sync:
https://github.com/paradigmxyz/reth/actions/runs/9007839847/job/24748566483

The important logs:

2024-05-08T20:09:21.442257Z ERROR reth_node_builder::launch::common: Failed to build global thread pool: ThreadPoolBuildError { kind: GlobalPoolAlreadyInitialized }
2024-05-08T20:09:26.441518Z ERROR reth_node_builder::launch::common: Failed to build global thread pool: ThreadPoolBuildError { kind: GlobalPoolAlreadyInitialized }
2024-05-08T20:09:44.902706Z ERROR blockchain_tree: Reverting canonical chain failed with error: OptimisticTargetRevert(88)

We might need something lazy for the thread pool, or it might have to do with launching multiple nodes? Have not debugged this in depth

The error is thrown here:

reth/crates/node/builder/src/launch/common.rs

Lines 110 to 126 in d467744

	/// Configure global settings this includes:
	///
	/// - Raising the file descriptor limit
	/// - Configuring the global rayon thread pool
	pub fn configure_globals(&self) {
	// Raise the fd limit of the process.
	// Does not do anything on windows.
	let _ = fdlimit::raise_fd_limit();
	// Limit the global rayon thread pool, reserving 2 cores for the rest of the system
	let _ = ThreadPoolBuilder::new()
	.num_threads(
	available_parallelism().map_or(25, |cpus| max(cpus.get().saturating_sub(2), 2)),
	)
	.build_global()
	.map_err(|e| error!("Failed to build global thread pool: {:?}", e));
	}

[mattsse]

we can simply remove the error log here

[Rjected]

we can simply remove the error log here

The test is timing out though, so we need to do something else to fix the test probably

[onbjerg]

This is because you can't have more than one threadpool per process, I ran into this with one of the benchmarks we hadn't kept up to date, too. Not entirely sure what the best way to circumvent this is

Edit: Actually this is a bit different, but might be a similar error. I ran into this with Tokio (the benchmark tool spawned a tokio runtime, after which the function we called did the same)

[Rjected]

Actually the flake might be more related to

2024-05-08T20:09:44.902706Z ERROR blockchain_tree: Reverting canonical chain failed with error: OptimisticTargetRevert(88)

But I still need to repro

[onbjerg]

Yeah reading the threadpoolbuilder docs, it returns an error but does not mention anything about it being fatal or stalling, so should be ok: https://docs.rs/rayon/latest/rayon/struct.ThreadPoolBuilder.html#method.build_global

[joshieDo]

That specific error log is expected and part of the e2e test @Rjected
so I'm thinking it's really about the long duration of the test (20s if i recall) unsure now on why its timing out

[Rjected]

@joshieDo do you mind taking a look at this? I haven't run into this in a while though, so it may just be difficult to repro

[github-actions]

This issue is stale because it has been open for 21 days with no activity.

[shekhirin]

I believe it's not an issue anymore, please re-open if occurs again.

[Rjected]

Still an issue https://github.com/paradigmxyz/reth/actions/runs/9506077502/job/26202446729?pr=8808