1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
//! What:
//! The progress of a push: whether the push and everything it started are at work, sampled at a small interval, and the
//! clock that stops it on a stall (no sign of work for the stall window) or at a ceiling (running that long in all).
//! Why:
//! A fixed deadline stops a slow push that is still working and lets a stuck push take its whole deadline. A process shows
//! that it works in three ways, each read from `/proc`, and only a tree that shows none of them is stalled:
//! bytes moved: `rchar` plus `wchar` of `/proc/<pid>/io`, summed over the tree;
//! processor time: `utime` plus `stime` of `/proc/<pid>/stat` growing;
//! a task able to run or waiting for the disk: state `R` or `D` in `/proc/<pid>/task/<tid>/stat`.
//! Measured on 2026-11-03, which decided the three:
//! a 30 MB push to a local bare remote moved 94 MB of `rchar` and 156 MB of `wchar` over its four processes while
//! `read_bytes` was 64 KB and `write_bytes` came late and in bursts, so the block-layer counters see neither pipes nor
//! sockets and lag behind the transfer;
//! `rchar` and `wchar` count `read` and `write` calls, not `send` and `recv` (102 bytes for 100000 bytes through
//! `send`/`recv`, 100102 through `read`), so a stretch that only curl performs shows through the pipe that feeds it;
//! a `dd` of 64 MB with `conv=fsync` was in state `D` in six of seven looks and used 0 processor ticks (a tick is 10 ms),
//! so a wait for the disk shows in neither bytes nor processor time, only in the state;
//! a parent in `vfork` is in state `D` with 0 ticks, the way the controls hold a tree in disk wait exactly;
//! a process reading a silent socket and a `sleep` are in state `S` with a constant tick count, so a wait on a server
//! that says nothing is a stall.
//! What each costs: bytes miss work that moves none; processor time makes a process that spins or polls look alive, and its
//! resolution is a tick; state `D` makes a process wedged for ever in the kernel (a dead network file system) look alive, and
//! state `R` one that is only waiting for a processor; the ceiling bounds all three. None of them sees a server that checks the
//! push before it answers: that tree is in state `S`, indistinguishable from a dead peer, so a check longer than the stall
//! window is stopped, and raising the window is the remedy, which the stop message names.
//! Where `/proc/<pid>/io` does not exist (no `/proc`, or a kernel without task I/O accounting) nothing is measured and the
//! stall window is a fixed deadline, as before.
//!
//! In TS you'd write (pseudocode):
//! ```ts
//! const clock = new Clock(child.pid, { stallMs, ceilingMs }); while (!exited) { const stop = clock.check(); if (stop) return stop; }
//! ```
/// The patience a clock keeps, and what a stop reports.
use super::bounded_push::{Patience, StopCause, StopReport};
/// The processes below the push, and the identity of each.
use super::process_tree::{TreeMember, descendants_of, member_of};
/// What each process of the tree was last seen to have moved.
use std::collections::HashMap;
/// Time spans and the monotonic clock.
use std::time::{Duration, Instant};
/// How often the counters of the tree are read.
pub const SAMPLE_INTERVAL: Duration = Duration::from_millis(500);
/// What:
/// The bytes a process has read and written, from its `/proc/<pid>/io`, or nothing where they cannot be read.
/// Why:
/// `rchar` and `wchar` are the counters that see pipes and sockets read and written with `read` and `write`.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// const io = Object.fromEntries(readFileSync(`/proc/${pid}/io`, 'utf8').split('\n').map((line) => line.split(': ')));
/// return Number(io.rchar) + Number(io.wchar);
/// ```
#[cfg(target_os = "linux")]
fn bytes_moved_by(pid: i32) -> Option<u64> {
let text: String = std::fs::read_to_string(format!("/proc/{pid}/io")).ok()?;
let mut read: Option<u64> = None;
let mut written: Option<u64> = None;
for line in text.lines() {
if let Some(value) = line.strip_prefix("rchar: ") {
read = value.trim().parse::<u64>().ok();
} else if let Some(value) = line.strip_prefix("wchar: ") {
written = value.trim().parse::<u64>().ok();
}
}
return Some(read?.saturating_add(written?));
}
/// Only Linux counts a process's bytes; elsewhere nothing is measured.
// UNTESTED ON WINDOWS: no gate compiles or runs this branch (the crate does not build for Windows; README, Requirements).
#[cfg(not(target_os = "linux"))]
fn bytes_moved_by(_pid: i32) -> Option<u64> {
return None;
}
/// What:
/// The state letter and the processor ticks (`utime` plus `stime`) of a `/proc/<pid>/stat` line.
/// Why:
/// The command name sits in parentheses and may hold spaces and parentheses, so the fields are read after the last
/// closing parenthesis; the state is the first of them and the two times are the twelfth and thirteenth.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// const [state, , , , , , , , , , , utime, stime] = text.slice(text.lastIndexOf(')') + 2).split(' ');
/// ```
fn parse_activity(stat: &str) -> Option<(char, u64)> {
let after_name: &str = stat.get(stat.rfind(')')? + 2..)?;
let fields: Vec<&str> = after_name.split(' ').collect();
let state: char = fields.first()?.chars().next()?;
let user: u64 = fields.get(11)?.parse().ok()?;
let system: u64 = fields.get(12)?.parse().ok()?;
return Some((state, user.saturating_add(system)));
}
/// Whether a state letter is one of a task that is at work: able to run (`R`) or waiting for the disk (`D`).
fn is_working(state: char) -> bool {
return state == 'R' || state == 'D';
}
/// What:
/// A process's processor ticks, and whether any of its tasks is able to run or waiting for the disk.
/// Why:
/// A thread that waits for the disk (an `fsync`) is in state `D` while the process's own state may be `S`, so every task
/// is read.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// { cpu: parse(`/proc/${pid}/stat`).utime + stime, working: tasks(pid).some((task) => 'RD'.includes(task.state)) }
/// ```
#[cfg(target_os = "linux")]
fn activity_of(pid: i32) -> Option<(u64, bool)> {
let text: String = std::fs::read_to_string(format!("/proc/{pid}/stat")).ok()?;
let (state, ticks): (char, u64) = parse_activity(text.as_str())?;
let mut working: bool = is_working(state);
if !working && let Ok(tasks) = std::fs::read_dir(format!("/proc/{pid}/task")) {
for task in tasks.flatten() {
if let Ok(line) = std::fs::read_to_string(task.path().join("stat"))
&& let Some((task_state, _)) = parse_activity(line.as_str())
&& is_working(task_state)
{
working = true;
break;
}
}
}
return Some((ticks, working));
}
/// Only Linux reads what a process is doing; elsewhere nothing is.
// UNTESTED ON WINDOWS: no gate compiles or runs this branch (the crate does not build for Windows; README, Requirements).
#[cfg(not(target_os = "linux"))]
fn activity_of(_pid: i32) -> Option<(u64, bool)> {
return None;
}
/// Whether this system counts the bytes of a process at all.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// existsSync('/proc/self/io')
/// ```
fn system_measures() -> bool {
return bytes_moved_by(i32::try_from(std::process::id()).unwrap_or(1)).is_some();
}
/// What:
/// The work of the tree under a root so far, and whether a sample saw any sign of work since the one before.
/// Why:
/// A process that ends takes its counters with it, so each process is remembered by its number and start time, and only
/// what it moved or used since it was last seen is added; a task able to run or waiting for the disk is a sign of work at
/// the sample that finds it.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// class Meter { moved = 0; sample(rootPid): boolean }
/// ```
#[derive(Debug, Default)]
pub struct Meter {
/// What each process of the tree had moved and used when it was last seen: bytes, then processor ticks.
seen: HashMap<(i32, u64), (u64, u64)>,
/// Every byte the tree has moved since the first sample.
moved: u64,
}
impl Meter {
/// A meter that has seen nothing.
pub fn new() -> Self {
return Self::default();
}
/// Every byte the tree has moved so far.
pub fn moved(&self) -> u64 {
return self.moved;
}
/// Take one look at one process: true when it moved bytes or used processor time since the last look, or is at work now.
fn observe(&mut self, key: (i32, u64), bytes: u64, ticks: u64, working: bool) -> bool {
let (bytes_before, ticks_before): (u64, u64) = self.seen.insert(key, (bytes, ticks)).unwrap_or((0, 0));
let moved_now: u64 = bytes.saturating_sub(bytes_before);
self.moved = self.moved.saturating_add(moved_now);
return moved_now > 0 || ticks > ticks_before || working;
}
/// Read the counters and the states of the root and everything below it; true when any of them shows work since the last
/// sample. A root that cannot be read (it ended) shows nothing.
pub fn sample(&mut self, root: i32) -> bool {
let mut members: Vec<TreeMember> = member_of(root).into_iter().collect();
members.extend(descendants_of(root));
let mut progressed: bool = false;
for member in members {
let bytes: Option<u64> = bytes_moved_by(member.pid);
let activity: Option<(u64, bool)> = activity_of(member.pid);
if bytes.is_none() && activity.is_none() {
continue;
}
let (ticks, working): (u64, bool) = activity.unwrap_or((0, false));
if self.observe((member.pid, member.started), bytes.unwrap_or(0), ticks, working) {
progressed = true;
}
}
return progressed;
}
}
/// What:
/// The clock of one bounded child: it says when the child has stalled, run past its ceiling or passed its fixed deadline.
/// Why:
/// The waiting loop asks it every few milliseconds; it reads the counters only every `SAMPLE_INTERVAL`.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// class Clock { check(): StopCause | undefined }
/// ```
pub struct Clock {
/// The process whose tree is measured.
root: i32,
/// What the child may do: a fixed time, or a stall window under a ceiling.
patience: Patience,
/// When the child started.
started: Instant,
/// When the tree last moved bytes (or the start, before it first does).
last_progress: Instant,
/// When the counters are next read.
next_sample: Instant,
/// The counters, or nothing where the system does not count bytes and the stall window is a fixed deadline.
meter: Option<Meter>,
}
impl Clock {
/// Start the clock of a child that has just been started.
pub fn start(pid: u32, patience: Patience) -> Self {
let now: Instant = Instant::now();
let measured: bool = matches!(patience, Patience::Progress { .. }) && system_measures();
return Self {
root: i32::try_from(pid).unwrap_or(i32::MAX),
patience,
started: now,
last_progress: now,
next_sample: now,
meter: if measured { Some(Meter::new()) } else { None },
};
}
/// The same clock as on a system that counts no bytes, for the control of that fallback.
#[cfg(test)]
pub fn unmeasured(mut self) -> Self {
self.meter = None;
return self;
}
/// The report of a stop now, for the limit that was hit.
fn report(&self, limit: Duration, now: Instant) -> StopReport {
return StopReport { limit, moved: self.meter.as_ref().map_or(0, Meter::moved), elapsed: now.duration_since(self.started) };
}
/// Whether the child must be stopped now, and why.
pub fn check(&mut self) -> Option<StopCause> {
let now: Instant = Instant::now();
let elapsed: Duration = now.duration_since(self.started);
let (stall, ceiling): (Duration, Duration) = match self.patience {
Patience::Fixed(limit) => return (elapsed >= limit).then_some(StopCause::Deadline),
Patience::Progress { stall, ceiling } => (stall, ceiling),
};
let Some(meter) = self.meter.as_mut() else {
// Nothing is measured here, so the stall window is the whole deadline.
return (elapsed >= stall).then_some(StopCause::Deadline);
};
if now >= self.next_sample {
if meter.sample(self.root) {
self.last_progress = now;
}
self.next_sample = now + SAMPLE_INTERVAL;
}
if elapsed >= ceiling {
return Some(StopCause::Ceiling(self.report(ceiling, now)));
}
if now.duration_since(self.last_progress) >= stall {
return Some(StopCause::Stalled(self.report(stall, now)));
}
return None;
}
}
/// Progress behavior is controlled in the test module.
#[cfg(test)]
#[path = "push_progress_tests.rs"]
pub(crate) mod tests;