Files
tokio/tokio/src/runtime/metrics/batch.rs
T
Carl Lerche 79a7e78c0d rt(threaded): basic self-tuning of injection queue (#5720)
Each multi-threaded runtime worker prioritizes pulling tasks off of its
local queue. Every so often, it checks the injection (global) queue for
work submitted there. Previously, "every so often," was a constant
"number of tasks polled" value. Tokio sets a default of 61, but allows
users to configure this value.

If workers are under load with tasks that are slow to poll, the
injection queue can be starved. To prevent starvation in this case, this
commit implements some basic self-tuning. The multi-threaded scheduler
tracks the mean task poll time using an exponentially-weighted moving
average. It then uses this value to pick an interval at which to check
the injection queue.

This commit is a first pass at adding self-tuning to the scheduler.
There are other values in the scheduler that could benefit from
self-tuning (e.g. the maintenance interval). Additionally, the
current-thread scheduler could also benfit from self-tuning. However, we
have reached the point where we should start investigating ways to unify
logic in both schedulers. Adding self-tuning to the current-thread
scheduler will be punted until after this unification.
2023-06-01 08:13:24 -07:00

163 lines
4.8 KiB
Rust

use crate::runtime::metrics::{HistogramBatch, WorkerMetrics};
use std::sync::atomic::Ordering::Relaxed;
use std::time::{Duration, Instant};
pub(crate) struct MetricsBatch {
/// Number of times the worker parked.
park_count: u64,
/// Number of times the worker woke w/o doing work.
noop_count: u64,
/// Number of tasks stolen.
steal_count: u64,
/// Number of times tasks where stolen.
steal_operations: u64,
/// Number of tasks that were polled by the worker.
poll_count: u64,
/// Number of tasks polled when the worker entered park. This is used to
/// track the noop count.
poll_count_on_last_park: u64,
/// Number of tasks that were scheduled locally on this worker.
local_schedule_count: u64,
/// Number of tasks moved to the global queue to make space in the local
/// queue
overflow_count: u64,
/// The total busy duration in nanoseconds.
busy_duration_total: u64,
/// Instant at which work last resumed (continued after park).
processing_scheduled_tasks_started_at: Instant,
/// If `Some`, tracks poll times in nanoseconds
poll_timer: Option<PollTimer>,
}
struct PollTimer {
/// Histogram of poll counts within each band.
poll_counts: HistogramBatch,
/// Instant when the most recent task started polling.
poll_started_at: Instant,
}
impl MetricsBatch {
pub(crate) fn new(worker_metrics: &WorkerMetrics) -> MetricsBatch {
let now = Instant::now();
MetricsBatch {
park_count: 0,
noop_count: 0,
steal_count: 0,
steal_operations: 0,
poll_count: 0,
poll_count_on_last_park: 0,
local_schedule_count: 0,
overflow_count: 0,
busy_duration_total: 0,
processing_scheduled_tasks_started_at: now,
poll_timer: worker_metrics
.poll_count_histogram
.as_ref()
.map(|worker_poll_counts| PollTimer {
poll_counts: HistogramBatch::from_histogram(worker_poll_counts),
poll_started_at: now,
}),
}
}
pub(crate) fn submit(&mut self, worker: &WorkerMetrics) {
worker.park_count.store(self.park_count, Relaxed);
worker.noop_count.store(self.noop_count, Relaxed);
worker.steal_count.store(self.steal_count, Relaxed);
worker
.steal_operations
.store(self.steal_operations, Relaxed);
worker.poll_count.store(self.poll_count, Relaxed);
worker
.busy_duration_total
.store(self.busy_duration_total, Relaxed);
worker
.local_schedule_count
.store(self.local_schedule_count, Relaxed);
worker.overflow_count.store(self.overflow_count, Relaxed);
if let Some(poll_timer) = &self.poll_timer {
let dst = worker.poll_count_histogram.as_ref().unwrap();
poll_timer.poll_counts.submit(dst);
}
}
/// The worker is about to park.
pub(crate) fn about_to_park(&mut self) {
self.park_count += 1;
if self.poll_count_on_last_park == self.poll_count {
self.noop_count += 1;
} else {
self.poll_count_on_last_park = self.poll_count;
}
}
/// Start processing a batch of tasks
pub(crate) fn start_processing_scheduled_tasks(&mut self) {
self.processing_scheduled_tasks_started_at = Instant::now();
}
/// Stop processing a batch of tasks
pub(crate) fn end_processing_scheduled_tasks(&mut self) {
let busy_duration = self.processing_scheduled_tasks_started_at.elapsed();
self.busy_duration_total += duration_as_u64(busy_duration);
}
/// Start polling an individual task
pub(crate) fn start_poll(&mut self) {
self.poll_count += 1;
if let Some(poll_timer) = &mut self.poll_timer {
poll_timer.poll_started_at = Instant::now();
}
}
/// Stop polling an individual task
pub(crate) fn end_poll(&mut self) {
if let Some(poll_timer) = &mut self.poll_timer {
let elapsed = duration_as_u64(poll_timer.poll_started_at.elapsed());
poll_timer.poll_counts.measure(elapsed, 1);
}
}
pub(crate) fn inc_local_schedule_count(&mut self) {
self.local_schedule_count += 1;
}
}
cfg_rt_multi_thread! {
impl MetricsBatch {
pub(crate) fn incr_steal_count(&mut self, by: u16) {
self.steal_count += by as u64;
}
pub(crate) fn incr_steal_operations(&mut self) {
self.steal_operations += 1;
}
pub(crate) fn incr_overflow_count(&mut self) {
self.overflow_count += 1;
}
}
}
fn duration_as_u64(dur: Duration) -> u64 {
u64::try_from(dur.as_nanos()).unwrap_or(u64::MAX)
}