subc-daemon 0.44.12

Embeddable subc daemon: bootstrap, module supervision, and opaque-byte splice routing.
Documentation
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
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
use super::*;
use crate::resource_totals::{CommitFence, Declaration, DeclarationArrival};
use std::sync::{
    atomic::{AtomicBool, Ordering},
    Weak,
};
use tokio::sync::{mpsc, OwnedSemaphorePermit, Semaphore};
mod refusals;

const IN_FLIGHT: usize = 4;
// Native macOS process-table walks measured 104–131 ms, so even one revision
// conflict can exhaust this window. Retain starts on expiry; other entries must stay retryable.
const VALIDATION_WINDOW: Duration = Duration::from_millis(200);
const REPLY_WINDOW: Duration = Duration::from_millis(200);

#[derive(Clone, Default)]
pub(crate) struct Queues(Arc<Mutex<HashMap<ConnectionId, Arc<Queue>>>>);

struct Queue {
    sender: mpsc::Sender<Work>,
    capacity: Arc<Semaphore>,
    fences: Mutex<BTreeMap<u64, Arc<CommitFence>>>,
    closed: AtomicBool,
    refusals: refusals::Refusals,
    losses: Arc<crate::resource_totals::declaration_loss::Losses>,
}

struct Work {
    losses: Arc<crate::resource_totals::declaration_loss::Losses>,
    handler: ControlHandler,
    ctx: RouteCtx,
    frame: Frame,
    batch: Declaration,
    fence: Arc<CommitFence>,
    arrival: DeclarationArrival,
    _permit: OwnedSemaphorePermit,
}

impl Queues {
    #[cfg(test)]
    pub(crate) fn fault_for_test(
        &self,
        handler: &ControlHandler,
        ctx: &RouteCtx,
        closed: bool,
        corr: u64,
    ) {
        let queue = self.queue(handler, ctx);
        queue.closed.store(closed, Ordering::Release);
        if !closed {
            queue
                .fences
                .lock()
                .unwrap()
                .insert(corr, Arc::new(CommitFence::default()));
        }
    }
    fn queue(&self, handler: &ControlHandler, ctx: &RouteCtx) -> Arc<Queue> {
        let module_id = handler
            .registry
            .get_module_by_connection(ctx.connection_id)
            .ok()
            .flatten()
            .map(|registration| registration.manifest.module_id);
        {
            let mut queues = self.0.lock().unwrap_or_else(|p| p.into_inner());
            Arc::clone(queues.entry(ctx.connection_id).or_insert_with(|| {
                let (sender, receiver) = mpsc::channel(IN_FLIGHT);
                let queue = Arc::new(Queue {
                    sender,
                    capacity: Arc::new(Semaphore::new(IN_FLIGHT)),
                    fences: Mutex::default(),
                    closed: AtomicBool::new(false),
                    losses: module_id
                        .as_deref()
                        .map(|id| handler.resource_tracker.declaration_loss(id))
                        .unwrap_or_default(),
                    refusals: refusals::Refusals::new(
                        ctx.egress.clone(),
                        handler.counters.clone(),
                        module_id,
                        Arc::clone(&handler.registry),
                    ),
                });
                tokio::spawn(run(Arc::downgrade(&queue), receiver));
                queue
            }))
        }
    }

    pub(crate) fn refuse(&self, handler: &ControlHandler, ctx: &RouteCtx, frame: Frame) {
        self.queue(handler, ctx).refusals.send(frame);
    }

    pub(crate) async fn enqueue(
        &self,
        handler: &ControlHandler,
        ctx: &RouteCtx,
        frame: Frame,
        batch: Declaration,
    ) -> Result<Vec<Frame>, RouterError> {
        let arrival = handler.resource_tracker.declaration_arrival();
        if let Some(error) = super::check_batch(
            &frame,
            batch.started.len(),
            batch.finished.len(),
            batch.known.as_ref().map_or(0, Vec::len),
        )? {
            self.refuse(handler, ctx, error);
            return Ok(Vec::new());
        }
        let queue = self.queue(handler, ctx);
        // This supervised reader also carries tool traffic and cancellation.
        // Refuse excess work immediately instead of parking the reader.
        let Ok(permit) = Arc::clone(&queue.capacity).try_acquire_owned() else {
            queue.losses.mark(&batch.started);
            queue.refusals.send(control_error_frame(
                &frame,
                "busy",
                "four agent command declarations are in flight; retry",
            )?);
            return Ok(Vec::new());
        };
        let fence = Arc::new(CommitFence::default());
        {
            let mut fences = queue.fences.lock().unwrap_or_else(|p| p.into_inner());
            if queue.closed.load(Ordering::Acquire) {
                drop(fences);
                queue.losses.mark(&batch.started);
                queue.refusals.send(control_error_frame(
                    &frame,
                    subc_protocol::error_codes::DECLARATION_CLOSED,
                    "declaration queue is closed; retry on a current supervised connection",
                )?);
                return Ok(Vec::new());
            }
            if fences.contains_key(&frame.header.corr) {
                drop(fences);
                queue.losses.mark(&batch.started);
                queue.refusals.send(control_error_frame(
                    &frame,
                    subc_protocol::error_codes::DUPLICATE_CORRELATION,
                    "declaration correlation is already in flight; retry with a fresh correlation",
                )?);
                return Ok(Vec::new());
            }
            fences.insert(frame.header.corr, Arc::clone(&fence));
        }
        let corr = frame.header.corr;
        let refused = frame.clone();
        let sent = queue.sender.try_send(Work {
            losses: Arc::clone(&queue.losses),
            handler: handler.clone(),
            ctx: ctx.clone(),
            frame,
            batch,
            fence,
            arrival,
            _permit: permit,
        });
        if let Err(error) = sent {
            queue.losses.mark(&error.into_inner().batch.started);
            queue
                .fences
                .lock()
                .unwrap_or_else(|p| p.into_inner())
                .remove(&corr);
            queue.refusals.send(control_error_frame(
                &refused,
                subc_protocol::error_codes::DECLARATION_CLOSED,
                "declaration queue is closed; retry on a current supervised connection",
            )?);
            return Ok(Vec::new());
        }
        Ok(Vec::new())
    }

    pub(crate) fn close(&self, connection: ConnectionId) {
        let Some(queue) = self
            .0
            .lock()
            .unwrap_or_else(|p| p.into_inner())
            .remove(&connection)
        else {
            return;
        };
        queue.closed.store(true, Ordering::Release);
        queue.capacity.close();
        for fence in queue
            .fences
            .lock()
            .unwrap_or_else(|p| p.into_inner())
            .values()
        {
            fence.cancel();
        }
    }

    pub(crate) fn cancel(&self, connection: ConnectionId, corr: u64) -> bool {
        if let Some(queue) = self
            .0
            .lock()
            .unwrap_or_else(|p| p.into_inner())
            .get(&connection)
        {
            if let Some(fence) = queue
                .fences
                .lock()
                .unwrap_or_else(|p| p.into_inner())
                .get(&corr)
            {
                fence.cancel();
                return true;
            }
        }
        false
    }
}

async fn run(queue: Weak<Queue>, mut receiver: mpsc::Receiver<Work>) {
    while let Some(work) = receiver.recv().await {
        let corr = work.frame.header.corr;
        let started = work.batch.started.clone();
        let fence = Arc::clone(&work.fence);
        let retention = Arc::new(CommitFence::default());
        let reply = validate(
            work.handler,
            work.ctx.connection_id,
            work.frame.clone(),
            work.batch,
            work.fence,
            Arc::clone(&retention),
            work.arrival,
        )
        .await;
        if !fence.committed()
            && !retention.committed()
            && reply.as_ref().map_or(true, |frames| {
                frames
                    .iter()
                    .any(|frame| frame.header.ty == FrameType::Error)
            })
        {
            work.losses.mark(&started);
        }
        let frames = reply.unwrap_or_else(|error| {
            control_error_frame(&work.frame, "backend_error", error.to_string())
                .into_iter()
                .collect()
        });
        for frame in frames {
            if !matches!(
                tokio::time::timeout(REPLY_WINDOW, work.ctx.egress.send(frame)).await,
                Ok(Ok(()))
            ) {
                break;
            }
        }
        if let Some(queue) = queue.upgrade() {
            queue
                .fences
                .lock()
                .unwrap_or_else(|p| p.into_inner())
                .remove(&corr);
        }
    }
}

async fn validate(
    handler: ControlHandler,
    connection: ConnectionId,
    frame: Frame,
    batch: Declaration,
    fence: Arc<CommitFence>,
    retention: Arc<CommitFence>,
    arrival: DeclarationArrival,
) -> Result<Vec<Frame>, RouterError> {
    static SLOTS: std::sync::OnceLock<Arc<Semaphore>> = std::sync::OnceLock::new();
    // Declarations have their own bounded readers, not the list/sampler pool.
    let slots = SLOTS.get_or_init(|| Arc::new(Semaphore::new(4)));
    let deadline = tokio::time::Instant::now() + VALIDATION_WINDOW;
    let corr = frame.header.corr;
    let timeout_frame = frame.clone();
    let pending_handler = handler.clone();
    let pending_batch = batch.clone();
    let timed_out = || {
        control_error_frame(
            &timeout_frame,
            "busy",
            "agent command validation timed out; retry",
        )
        .map(|reply| vec![reply])
    };
    let retain = || {
        pending_handler.handle_agent_commands_fenced(
            connection,
            timeout_frame.clone(),
            pending_batch.clone(),
            DeclarationValidation {
                fence: &retention,
                arrival,
                defer: Some("agent command validation timed out; retry uncompleted finished and known entries"),
            },
        )
    };
    let permit = match tokio::time::timeout_at(deadline, Arc::clone(slots).acquire_owned()).await {
        Ok(Ok(permit)) => permit,
        _ if fence.cancel_for_retention() => return retain(),
        _ => return timed_out(),
    };
    let worker_fence = Arc::clone(&fence);
    let mut worker = tokio::task::spawn_blocking(move || {
        let _permit = permit;
        handler.handle_agent_commands_fenced(
            connection,
            frame,
            batch,
            DeclarationValidation {
                fence: &worker_fence,
                arrival,
                defer: None,
            },
        )
    });
    match tokio::time::timeout_at(deadline, &mut worker).await {
        Ok(result) => result.map_err(|error| RouterError::backend(0, corr, error.to_string()))?,
        Err(_) if fence.cancel_for_retention() => retain(),
        Err(_) if !fence.committed() => timed_out(),
        // The atomic commit decision preceded the timeout. Its short mutation
        // must supply the actual reply, never an error claiming it was cancelled.
        Err(_) => worker
            .await
            .map_err(|error| RouterError::backend(0, corr, error.to_string()))?,
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[tokio::test]
    async fn oversized_batches_are_refused_without_waiting_for_the_loss_metadata_lock() {
        let handler = ControlHandler::new(Arc::new(Registry::default()));
        let (sender, _receiver) = mpsc::channel(1);
        let ctx = RouteCtx {
            connection_id: ConnectionId::new(905),
            egress: FrameSink::new(sender),
        };
        let queue = handler.agent_declarations.queue(&handler, &ctx);
        let losses = Arc::clone(&queue.losses);
        let (ready, wait) = std::sync::mpsc::channel();
        let holder =
            std::thread::spawn(move || losses.hold_for_test(ready, Duration::from_millis(500)));
        wait.recv().unwrap();
        let began = std::time::Instant::now();
        handler
            .agent_declarations
            .enqueue(
                &handler,
                &ctx,
                frame(1),
                Declaration {
                    started: vec![
                        AgentCommandRoot {
                            pid: 42,
                            start_time_ms: 1
                        };
                        257
                    ],
                    ..Declaration::default()
                },
            )
            .await
            .unwrap();
        let elapsed = began.elapsed();
        holder.join().unwrap();
        assert!(
            elapsed < Duration::from_millis(200),
            "an oversized request scanned or waited under the loss lock: {elapsed:?}"
        );
    }

    #[tokio::test]
    async fn a_busy_started_batch_records_missing_coverage_in_module_diagnostics() {
        let registry = Arc::new(Registry::default());
        registry
            .register_with_control_ops(
                subc_protocol::manifest::ModuleManifest::builder("lost-module", "0.1.0")
                    .protocol_ver(1)
                    .build(),
                1,
                ConnectionId::new(901),
                Vec::new(),
            )
            .unwrap();
        let handler = ControlHandler::new(registry);
        let (sender, _receiver) = mpsc::channel(1);
        let ctx = RouteCtx {
            connection_id: ConnectionId::new(901),
            egress: FrameSink::new(sender),
        };
        let queue = handler.agent_declarations.queue(&handler, &ctx);
        let _held = Arc::clone(&queue.capacity)
            .acquire_many_owned(4)
            .await
            .unwrap();
        handler
            .agent_declarations
            .enqueue(
                &handler,
                &ctx,
                frame(1),
                Declaration {
                    started: vec![AgentCommandRoot {
                        pid: 42,
                        start_time_ms: 1,
                    }],
                    ..Declaration::default()
                },
            )
            .await
            .unwrap();
        let loss = handler.resource_tracker.declaration_loss("lost-module");
        assert!(loss.lower());
        let reply = handler.handle_server_describe(frame(2)).unwrap();
        let body: serde_json::Value = serde_json::from_slice(&reply[0].body).unwrap();
        assert_eq!(
            body["counters"]["agent_command_declaration_loss_by_module"]["lost-module"]
                ["lost_identities"],
            1
        );
        assert_eq!(
            body["counters"]["agent_command_declaration_loss_by_module"]["lost-module"]["overflow"],
            false
        );
    }
    use crate::router::FrameSink;

    fn frame(corr: u64) -> Frame {
        Frame::build(
            FrameType::Request,
            control_flags(),
            0,
            0,
            corr,
            b"{}".to_vec(),
        )
        .unwrap()
    }

    #[tokio::test]
    async fn fifth_declaration_is_refused_without_parking_cancel_or_goodbye() {
        let handler = ControlHandler::new(Arc::new(Registry::default()));
        let queues = handler.agent_declarations.clone();
        let (sender, mut receiver) = mpsc::channel(1);
        let ctx = RouteCtx {
            connection_id: ConnectionId::new(900),
            egress: FrameSink::new(sender),
        };
        ctx.egress.send(frame(0)).await.unwrap();
        for corr in 1..=4 {
            queues
                .enqueue(&handler, &ctx, frame(corr), Declaration::default())
                .await
                .unwrap();
        }
        let fifth = tokio::time::timeout(
            Duration::from_millis(20),
            queues.enqueue(&handler, &ctx, frame(5), Declaration::default()),
        )
        .await
        .expect("the shared reader must never await a declaration permit")
        .unwrap();
        assert!(
            fifth.is_empty(),
            "busy replies use nonblocking egress, not the reader's ordinary reply path"
        );
        let cancel = Frame::build(FrameType::Cancel, control_flags(), 0, 0, 4, Vec::new()).unwrap();
        assert!(
            handler
                .handle_control_frame(&ctx, cancel)
                .await
                .unwrap()
                .is_empty(),
            "the next Cancel is dispatched on the same connection while replies are held"
        );
        assert_eq!(receiver.recv().await.unwrap().header.corr, 0);
        let mut accepted = Vec::new();
        let mut busy = false;
        for _ in 0..5 {
            let reply = tokio::time::timeout(Duration::from_secs(2), receiver.recv())
                .await
                .unwrap()
                .unwrap();
            if reply.header.corr == 5 {
                assert_eq!(
                    serde_json::from_slice::<ErrorBody>(&reply.body)
                        .unwrap()
                        .code,
                    "busy"
                );
                busy = true;
            } else {
                accepted.push(reply.header.corr);
            }
        }
        assert_eq!(
            accepted,
            vec![1, 2, 3, 4],
            "admitted replies preserve arrival order"
        );
        assert!(busy);
        let goodbye =
            Frame::build(FrameType::Goodbye, control_flags(), 0, 0, 6, Vec::new()).unwrap();
        handler.handle_control_frame(&ctx, goodbye).await.unwrap();
        assert!(
            !queues.cancel(ctx.connection_id, 4),
            "Goodbye cleanup removes queued cancellation state"
        );
    }

    #[tokio::test]
    async fn blocked_reply_expires_and_releases_its_declaration_permit() {
        let handler = ControlHandler::new(Arc::new(Registry::default()));
        let queues = Queues::default();
        let (sender, _receiver) = mpsc::channel(1);
        let ctx = RouteCtx {
            connection_id: ConnectionId::new(901),
            egress: FrameSink::new(sender),
        };
        ctx.egress.send(frame(0)).await.unwrap();
        queues
            .enqueue(&handler, &ctx, frame(1), Declaration::default())
            .await
            .unwrap();
        let queue = Arc::clone(queues.0.lock().unwrap().get(&ctx.connection_id).unwrap());
        tokio::time::timeout(Duration::from_secs(2), async {
            while queue.capacity.available_permits() != IN_FLIGHT {
                tokio::task::yield_now().await;
            }
        })
        .await
        .expect("a stalled writer must not retain admission forever");
        assert!(queue.fences.lock().unwrap().is_empty());
    }
}