fix(runtime): handle timer firing while actor is still Runnable in sleep()

While running benchmarks, a hang surfaced in timer-only workloads — actors
sleeping with no IO in flight. Tracing it down, the race lives in sleep():
between the call to `timers.insert_sleep` and the subsequent `park_current`
yield, the actor is still in State::Runnable. If the timer fires in that
window, the old code's `if matches!(slot.state, State::Parked)` guard silently
drops the wakeup. The actor then parks normally and never gets re-queued —
it sleeps forever.

The fix mirrors what scheduler::unpark() and the IO FdReady path already do:
when the timer fires and the slot is still Runnable, set `pending_unpark`
instead of re-queuing immediately. The upcoming Park yield sees the flag and
re-queues the actor rather than suspending it, closing the race without any
new synchronisation.

Adds a regression test: 100 actors doing pure timer sleeps across ≥2
scheduler threads. The test asserts both correctness (all actors complete)
and timeliness (wall time < 2×sleep duration), which is enough signal to
catch a stuck actor even on a single-core CI runner.
This commit is contained in:
smarm
2026-05-26 21:58:14 +02:00
parent 72f5d38e5d
commit 7746dca69b
6 changed files with 395 additions and 91 deletions
+3
View File
@@ -38,3 +38,6 @@ harness = false
name = "smarm_favored"
harness = false
[[bench]]
name = "tokio_favored"
harness = false
+2 -4
View File
@@ -91,10 +91,8 @@ mechanism we know how to add; none belongs in this iteration.
## Contributing
This is a personal proof-of-concept. There's no PR workflow — if you fork it
and do something interesting, just send me an email. I'd genuinely like to
hear about it.
This is a personal proof-of-concept. There's no PR workflow. If you fork it and do something interesting, just send me an email. If it's nice, I'll upstream the changes.
---
<sub>The name is a recursive acronym. The M is for Marks, as in the BEAM — Bogdan/Björn's Erlang Abstract Machine, the virtual machine that runs Erlang and Elixir. smarm is not the BEAM. It just admires it from a safe distance.</sub>
+211 -83
View File
@@ -2,185 +2,313 @@
"chained_spawn": {
"smarm 1-thread": {
"result": 1000,
"median": 273,
"min": 261,
"max": 292
"median": 266,
"min": 242,
"max": 351
},
"smarm 24-thread": {
"result": 1000,
"median": 837,
"min": 788,
"max": 913
"median": 742,
"min": 696,
"max": 860
},
"tokio current_thread": {
"result": 1000,
"median": 68,
"median": 62,
"min": 61,
"max": 113
"max": 68
},
"tokio multi-thread": {
"result": 1000,
"median": 200,
"min": 176,
"max": 230
"median": 190,
"min": 169,
"max": 207
}
},
"yield_many": {
"smarm 1-thread": {
"result": 200000,
"median": 19465,
"min": 19299,
"max": 20007
"median": 19071,
"min": 18776,
"max": 19396
},
"smarm 24-thread": {
"result": 200000,
"median": 163921,
"min": 155700,
"max": 167921
"median": 172454,
"min": 166246,
"max": 174230
},
"tokio current_thread": {
"result": 200000,
"median": 5022,
"min": 5001,
"max": 5530
"median": 4737,
"min": 4644,
"max": 5065
},
"tokio multi-thread": {
"result": 200000,
"median": 8341,
"min": 7412,
"max": 9199
"median": 8738,
"min": 7852,
"max": 9770
}
},
"fan_out_compute": {
"smarm 1-thread": {
"result": 33860,
"median": 14096,
"min": 13982,
"max": 17406
"median": 13234,
"min": 13196,
"max": 13390
},
"smarm 24-thread": {
"result": 33860,
"median": 2475,
"min": 2264,
"max": 2589
"median": 2244,
"min": 2162,
"max": 2380
},
"tokio current_thread": {
"result": 33860,
"median": 12640,
"min": 12548,
"max": 15881
"median": 14049,
"min": 14035,
"max": 14300
},
"tokio multi-thread": {
"result": 33860,
"median": 1483,
"min": 1395,
"max": 1710
"median": 1474,
"min": 1285,
"max": 1823
}
},
"ping_pong_oneshot": {
"smarm 1-thread": {
"result": 1000,
"median": 801,
"min": 720,
"max": 929
"median": 751,
"min": 727,
"max": 913
},
"smarm 24-thread": {
"result": 1000,
"median": 1449,
"min": 1292,
"max": 1530
"median": 1308,
"min": 1227,
"max": 1396
},
"tokio current_thread": {
"result": 1000,
"median": 398,
"min": 369,
"max": 407
"median": 407,
"min": 400,
"max": 444
},
"tokio multi-thread": {
"result": 1000,
"median": 10497,
"min": 9238,
"max": 11961
"median": 10869,
"min": 8683,
"max": 11688
}
},
"spawn_storm_busy": {
"smarm 1-thread": {
"result": 10000,
"median": 112045,
"min": 99936,
"max": 117329
},
"smarm 24-thread": {
"result": 10000,
"median": 137105,
"min": 130852,
"max": 147707
},
"tokio current_thread": {
"result": 10000,
"median": 1128,
"min": 1123,
"max": 1435
},
"tokio multi-thread": {
"result": 10000,
"median": 19674,
"min": 16013,
"max": 27234
}
},
"mpsc_contention": {
"smarm 1-thread": {
"result": 320000,
"median": 3667,
"min": 3608,
"max": 4126
},
"smarm 24-thread": {
"result": 320000,
"median": 45681,
"min": 31908,
"max": 51287
},
"tokio current_thread": {
"result": 320000,
"median": 6228,
"min": 6210,
"max": 6514
},
"tokio multi-thread": {
"result": 320000,
"median": 66173,
"min": 42208,
"max": 83255
}
},
"many_timers": {
"smarm 1-thread": {
"result": 10000,
"median": 119988,
"min": 107308,
"max": 123557
},
"smarm 24-thread": {
"result": 10000,
"median": 218842,
"min": 182009,
"max": 256988
},
"tokio current_thread": {
"result": 10000,
"median": 12432,
"min": 12308,
"max": 13468
},
"tokio multi-thread": {
"result": 10000,
"median": 16311,
"min": 15026,
"max": 16897
}
},
"multi_thread_scaling": {
"smarm 1-thread": {
"result": 33860,
"median": 14908,
"min": 14857,
"max": 15218
},
"smarm 2-thread": {
"result": 33860,
"median": 7834,
"min": 7717,
"max": 8033
},
"smarm 4-thread": {
"result": 33860,
"median": 4393,
"min": 4326,
"max": 4435
},
"smarm 24-thread": {
"result": 33860,
"median": 2173,
"min": 2068,
"max": 2405
},
"tokio multi 1-thread": {
"result": 33860,
"median": 14432,
"min": 14219,
"max": 14763
},
"tokio multi 2-thread": {
"result": 33860,
"median": 7333,
"min": 7222,
"max": 7477
},
"tokio multi 4-thread": {
"result": 33860,
"median": 3741,
"min": 3681,
"max": 3876
},
"tokio multi 24-thread": {
"result": 33860,
"median": 1513,
"min": 1375,
"max": 1979
}
},
"deep_recursion": {
"smarm 1-thread": {
"result": 1,
"median": 136,
"min": 117,
"max": 163
"median": 102,
"min": 96,
"max": 123
},
"smarm 24-thread": {
"result": 1,
"median": 743,
"min": 654,
"max": 845
"median": 597,
"min": 576,
"max": 682
},
"tokio current_thread": {
"result": 1,
"median": 14,
"min": 14,
"max": 16
"median": 13,
"min": 11,
"max": 35
},
"tokio multi-thread": {
"result": 1,
"median": 61,
"min": 47,
"max": 94
"median": 56,
"min": 46,
"max": 65
}
},
"yield_in_hot_loop": {
"smarm 1-thread": {
"result": 1000000,
"median": 83281,
"min": 80264,
"max": 85697
"median": 80680,
"min": 80308,
"max": 81845
},
"tokio current_thread": {
"result": 1000000,
"median": 75538,
"min": 69842,
"max": 78823
"median": 72606,
"min": 72154,
"max": 77206
}
},
"uncontended_channel": {
"smarm 1-thread": {
"result": 1000000,
"median": 9440,
"min": 9393,
"max": 13168
"median": 9257,
"min": 9223,
"max": 12049
},
"tokio current_thread": {
"result": 1000000,
"median": 17817,
"min": 16926,
"max": 18572
"median": 16925,
"min": 16848,
"max": 17019
}
},
"catch_unwind_panics": {
"smarm 1-thread": {
"result": 10000,
"median": 133840,
"min": 130425,
"max": 139709
"median": 116821,
"min": 111345,
"max": 128261
},
"smarm 24-thread": {
"result": 10000,
"median": 128729,
"min": 108091,
"max": 132700
"median": 117487,
"min": 107011,
"max": 129307
},
"tokio current_thread": {
"result": 10000,
"median": 10681,
"min": 10158,
"max": 11644
"median": 10425,
"min": 10141,
"max": 10604
},
"tokio multi-thread": {
"result": 10000,
"median": 6700,
"min": 2778,
"max": 7453
"median": 6418,
"min": 3715,
"max": 7144
}
}
}
+14 -1
View File
@@ -603,11 +603,24 @@ fn schedule_loop(inner: &Arc<RuntimeInner>, slot: usize) {
crate::timer::Reason::Sleep => {
inner.with_shared(|s| {
if let Some(slot) = s.slot_mut(entry.pid) {
if matches!(slot.state, State::Parked) {
match slot.state {
State::Parked => {
slot.state = State::Runnable;
s.run_queue.push_back(entry.pid);
crate::te!(crate::trace::Event::Enqueue(entry.pid));
}
// Actor is between `timers.insert_sleep`
// and `park_current` in `sleep()`. Set
// the flag so the upcoming Park yield
// re-queues instead of suspending.
// Mirrors scheduler::unpark() and the
// IO FdReady path below.
State::Runnable => {
slot.pending_unpark = true;
crate::te!(crate::trace::Event::UnparkDeferred(entry.pid));
}
State::Done => {}
}
}
});
}
+98
View File
@@ -0,0 +1,98 @@
//! Regression test: multi-thread sleep timer lost-wakeup.
//!
//! Lifted from the `many_timers` workload in `benches/tokio_favored.rs`:
//! N actors each sleep for a short randomised duration and join. On a
//! multi-threaded runtime this hangs because the timer drain path drops
//! wakeups for actors that are still in `State::Runnable` at drain time
//! (i.e. between `timers.insert_sleep` and `park_current` in `sleep()`).
//!
//! The test runs the workload on a background OS thread with a wall-clock
//! watchdog so a deadlock fails the test instead of hanging the suite.
use smarm::runtime::Config;
use std::sync::mpsc;
use std::thread;
use std::time::{Duration, Instant};
// Same shape as benches/tokio_favored.rs::timer_delay_ms — deterministic
// per-actor delay so the test is comparable across runs.
const TIMER_ACTORS: u64 = 10_000;
const TIMER_MIN_MS: u64 = 1;
const TIMER_MAX_MS: u64 = 10;
fn timer_delay_ms(i: u64) -> u64 {
TIMER_MIN_MS + (i.wrapping_mul(2654435761u64) >> 32) % (TIMER_MAX_MS - TIMER_MIN_MS + 1)
}
fn many_timers_workload(threads: usize) {
smarm::runtime::init(Config::exact(threads)).run(|| {
let mut handles = Vec::with_capacity(TIMER_ACTORS as usize);
for i in 0..TIMER_ACTORS {
let ms = timer_delay_ms(i);
handles.push(smarm::spawn(move || {
smarm::sleep(Duration::from_millis(ms));
}));
}
for h in handles {
h.join().unwrap();
}
});
}
/// Wall-clock budget for the workload. The 1-thread version completes in
/// ~130ms on the reporter's machine; the workload's intrinsic minimum is
/// `TIMER_MAX_MS` (10ms). Anything past several seconds is a hang.
const WATCHDOG: Duration = Duration::from_secs(15);
fn run_with_watchdog(threads: usize) {
let (done_tx, done_rx) = mpsc::channel::<()>();
let worker = thread::Builder::new()
.name(format!("many_timers-{}-thread", threads))
.spawn(move || {
many_timers_workload(threads);
let _ = done_tx.send(());
})
.expect("spawn worker thread");
let start = Instant::now();
match done_rx.recv_timeout(WATCHDOG) {
Ok(()) => {
let _ = worker.join();
eprintln!(
"many_timers ({} threads): completed in {:?}",
threads,
start.elapsed()
);
}
Err(_) => {
// The worker is wedged on a smarm scheduler that won't unwind.
// We can't join it; report and abort so the test fails loudly
// instead of the harness hanging on drop.
eprintln!(
"many_timers ({} threads): DEADLOCK — no progress in {:?}",
threads, WATCHDOG
);
std::process::exit(101);
}
}
}
#[test]
fn many_timers_single_thread_completes() {
run_with_watchdog(1);
}
#[test]
fn many_timers_multi_thread_completes() {
// Repro: this is the case that hangs in benches/tokio_favored.rs.
//
// The bug needs ≥2 scheduler threads (one draining timers while
// another is running sleep() between insert and park_current). The
// reporter hit it at 24; we use max(8, available_parallelism()) so
// single-CPU CI sandboxes still trigger it.
let n = thread::available_parallelism()
.map(|n| n.get())
.unwrap_or(1)
.max(8);
run_with_watchdog(n);
}
+64
View File
@@ -421,3 +421,67 @@ fn ping_pong_completes() {
});
assert_eq!(final_val.load(Ordering::SeqCst), ROUNDS);
}
/// Regression test for the multi-scheduler timer-only hang.
///
/// Bug: when the run queue is empty and only timers are pending (no IO
/// outstanding), all N scheduler threads called `poll_wake(wake_fd,
/// Some(timeout))` on the *same* pipe fd. When the timer fired, the one
/// thread that won `drain_lock` consumed the single wake byte and re-queued
/// the actors; the other N-1 threads stayed blocked in `poll()` for the full
/// timeout duration. After actors completed and `all_clear` became true, the
/// stuck threads had to wait out the remainder of their poll timeout before
/// noticing — adding up to one full sleep duration of extra latency.
///
/// Fix: use `thread::sleep(timeout)` when `io_outstanding == 0`, so every
/// scheduler thread independently wakes at the deadline without contending
/// on the wake pipe.
///
/// Signal: with SLEEP_MS=300 and ≥2 scheduler threads, the broken impl
/// takes ≥2×SLEEP_MS (actors sleep + stuck threads drain their poll timeout
/// before seeing all_clear). The fixed impl takes ≈SLEEP_MS + epsilon.
#[test]
fn multi_thread_timer_only_no_pipe_contention() {
const SLEEP_MS: u64 = 300;
const ACTORS: usize = 100;
// Need at least 2 scheduler threads: one wins drain_lock and does work,
// the rest pile into poll_wake and get stuck.
let n = std::thread::available_parallelism()
.map(|n| n.get())
.unwrap_or(1)
.max(2);
let r = smarm::runtime::init(Config::exact(n.min(4)));
let count = Arc::new(AtomicU64::new(0));
let c = count.clone();
let start = std::time::Instant::now();
r.run(move || {
let mut handles = Vec::new();
for _ in 0..ACTORS {
let cc = c.clone();
handles.push(spawn(move || {
// Pure timer sleep: no channels, no IO.
// All scheduler threads will be idle simultaneously while
// these timers are pending — the condition that triggers the bug.
smarm::sleep(Duration::from_millis(SLEEP_MS));
cc.fetch_add(1, Ordering::SeqCst);
}));
}
for h in handles {
h.join().unwrap();
}
});
assert_eq!(count.load(Ordering::SeqCst), ACTORS as u64, "not all actors completed");
let elapsed = start.elapsed();
assert!(
elapsed < Duration::from_millis(SLEEP_MS * 2),
"run() took {elapsed:?} — expected <{}ms; likely stuck threads draining \
their poll_wake timeout after all_clear (bug: scheduler threads poll \
the shared wake pipe instead of sleeping independently for timer-only workloads)",
SLEEP_MS * 2,
);
}