Run reconstruction inside a scoped rayon pool (#8075)

Co-Authored-By: Jimmy Chen <jchen.tc@gmail.com>

Co-Authored-By: Eitan Seri- Levi <eserilev@gmail.com>

Co-Authored-By: Eitan Seri-Levi <eserilev@ucsc.edu>
This commit is contained in:
Eitan Seri-Levi
2025-09-23 23:37:34 -07:00
committed by GitHub
parent d80c0ff5b5
commit af274029e8
11 changed files with 123 additions and 60 deletions

View File

@@ -12,7 +12,6 @@ logging = { workspace = true }
metrics = { workspace = true }
num_cpus = { workspace = true }
parking_lot = { workspace = true }
rayon = { workspace = true }
serde = { workspace = true }
slot_clock = { workspace = true }
strum = { workspace = true }

View File

@@ -38,7 +38,6 @@
//! checks the queues to see if there are more parcels of work that can be spawned in a new worker
//! task.
use crate::rayon_manager::RayonManager;
use crate::work_reprocessing_queue::{
QueuedBackfillBatch, QueuedColumnReconstruction, QueuedGossipBlock, ReprocessQueueMessage,
};
@@ -48,7 +47,6 @@ use lighthouse_network::{MessageId, NetworkGlobals, PeerId};
use logging::TimeLatch;
use logging::crit;
use parking_lot::Mutex;
use rayon::ThreadPool;
pub use scheduler::work_reprocessing_queue;
use serde::{Deserialize, Serialize};
use slot_clock::SlotClock;
@@ -61,7 +59,7 @@ use std::sync::Arc;
use std::task::Context;
use std::time::{Duration, Instant};
use strum::IntoStaticStr;
use task_executor::TaskExecutor;
use task_executor::{RayonPoolType, TaskExecutor};
use tokio::sync::mpsc;
use tokio::sync::mpsc::error::TrySendError;
use tracing::{debug, error, trace, warn};
@@ -76,7 +74,6 @@ use work_reprocessing_queue::{
};
mod metrics;
pub mod rayon_manager;
pub mod scheduler;
/// The maximum size of the channel for work events to the `BeaconProcessor`.
@@ -810,7 +807,6 @@ pub struct BeaconProcessor<E: EthSpec> {
pub network_globals: Arc<NetworkGlobals<E>>,
pub executor: TaskExecutor,
pub current_workers: usize,
pub rayon_manager: RayonManager,
pub config: BeaconProcessorConfig,
}
@@ -1609,10 +1605,7 @@ impl<E: EthSpec> BeaconProcessor<E> {
}
Work::ChainSegmentBackfill(process_fn) => {
if self.config.enable_backfill_rate_limiting {
task_spawner.spawn_blocking_with_rayon(
self.rayon_manager.low_priority_threadpool.clone(),
process_fn,
)
task_spawner.spawn_blocking_with_rayon(RayonPoolType::LowPriority, process_fn)
} else {
// use the global rayon thread pool if backfill rate limiting is disabled.
task_spawner.spawn_blocking(process_fn)
@@ -1681,17 +1674,16 @@ impl TaskSpawner {
}
/// Spawns a blocking task on a rayon thread pool, dropping the `SendOnDrop` after task completion.
fn spawn_blocking_with_rayon<F>(self, thread_pool: Arc<ThreadPool>, task: F)
fn spawn_blocking_with_rayon<F>(self, rayon_pool_type: RayonPoolType, task: F)
where
F: FnOnce() + Send + 'static,
{
self.executor.spawn_blocking(
self.executor.spawn_blocking_with_rayon(
move || {
thread_pool.install(|| {
task();
});
task();
drop(self.send_idle_on_drop)
},
rayon_pool_type,
WORKER_TASK_NAME,
)
}

View File

@@ -1,27 +0,0 @@
use rayon::{ThreadPool, ThreadPoolBuilder};
use std::sync::Arc;
const DEFAULT_LOW_PRIORITY_DIVISOR: usize = 4;
const MINIMUM_LOW_PRIORITY_THREAD_COUNT: usize = 1;
pub struct RayonManager {
/// Smaller rayon thread pool for lower-priority, compute-intensive tasks.
/// By default ~25% of CPUs or a minimum of 1 thread.
pub low_priority_threadpool: Arc<ThreadPool>,
}
impl Default for RayonManager {
fn default() -> Self {
let low_prio_threads =
(num_cpus::get() / DEFAULT_LOW_PRIORITY_DIVISOR).max(MINIMUM_LOW_PRIORITY_THREAD_COUNT);
let low_priority_threadpool = Arc::new(
ThreadPoolBuilder::new()
.num_threads(low_prio_threads)
.build()
.expect("failed to build low-priority rayon pool"),
);
Self {
low_priority_threadpool,
}
}
}