magi/queue.rs
1//! The task queue: what magi should do next, and who asked for it.
2//!
3//! The queue is what lets magi run unattended. `magi serve` takes the next
4//! task, runs the graph on it, records the outcome, and takes the next one.
5//!
6//! It is also the reason an agent can ask for work. `magi task add` is the
7//! whole interface, and it is the same command whether a human types it at a
8//! prompt, a phone posts it through the web UI, or an implementer inside a run
9//! shells out to it because it noticed something worth doing but out of scope.
10//! magi's CLI is the operating surface for both kinds of user; the queue is
11//! where their intentions meet.
12//!
13//! One task is one JSON file under [`Queue`]'s root. Files rather than a
14//! database because the operator has to be able to read, edit, and delete the
15//! backlog with the tools already on the machine, and because a crashed daemon
16//! must leave a queue the next one can pick up without recovery ceremony.
17//!
18//! # Shape
19//!
20//! [`Task`] is data plus *pure* state transitions - [`Task::fail`] decides
21//! whether an attempt was the last one, and touches no disk. [`Queue`] owns all
22//! I/O and is constructed with its root, so a test drives a real queue in a
23//! temp directory without setting a process-global home. Splitting them this
24//! way is why the retry policy below can be asserted directly.
25//!
26//! # Bounded by construction
27//!
28//! An autonomous loop that retries forever is a way to spend money on a task
29//! that cannot succeed. Every claim increments [`Task::attempts`]; a task that
30//! has burned its attempts becomes [`TaskStatus::Held`] and waits for a human
31//! rather than for another agent.
32
33use std::path::{Path, PathBuf};
34
35use anyhow::{Context, Result, bail};
36use jiff::Timestamp;
37use serde::{Deserialize, Serialize};
38
39use crate::ask::Questions;
40
41/// On-disk format for a queued task. Bumped when a field's meaning changes.
42///
43/// 6: added [`Task::resume_override`] (a field only; `#[serde(default)]`, so
44/// an older record reads as `None` and [`read_path`] still accepts it).
45///
46/// 5: added [`Task::triage_applied`], the ids of triage questions whose
47/// answer has already been applied to this task. A "resume" answer used to
48/// leave no trace ([`Task::release`] clears [`Task::hold_reason`], which was
49/// the only place the applied marker lived), so when the released task failed
50/// its attempts and went back to `held`, the next idle pass found the same
51/// answered question "not applied" and released it again with `attempts` reset
52/// to 0 - the `max_attempts` bound never held. `#[serde(default)]` so an older
53/// record reads as empty. A task already looping when this build arrives has
54/// no record, so it is released once more, recorded, and then stays held.
55///
56/// 4: added [`Task::blocked_from`], the status a task had the moment it
57/// became [`TaskStatus::Blocked`], so [`Task::unblock`] restores it instead
58/// of always landing on [`TaskStatus::Queued`]. Without it, a task a human
59/// or `crate::triage` had deliberately left [`TaskStatus::Held`] — machine
60/// or manual — would lose that the instant `crate::conduct` blocked it on a
61/// follow-up question, and come back `Queued` the moment the question was
62/// answered, regardless of what the answer said: exactly the loop where a
63/// task the operator told to stay held instead re-enters the competition
64/// queue every time someone answers a question about it. `#[serde(default)]`
65/// so an older record reads as `None`; [`Task::unblock`] then falls back to
66/// inferring `Held` from surviving hold evidence ([`Task::hold_reason`] /
67/// [`Task::hold_source`], never cleared by [`Task::block`]) rather than
68/// guessing `Queued` outright — see [`Task::unblock`]'s own doc.
69///
70/// 3: added [`HoldSource`] so conductor recovery cannot release a hold an
71/// operator deliberately placed. Old records default to `None` and are
72/// protected as operator-held until an explicit release; the safe direction
73/// when their author was never recorded.
74///
75/// 2: added [`TaskStatus::Blocked`], [`Task::blocked_by`] and
76/// [`Task::block_reason`] (`crate::conduct`'s decisions) and
77/// [`Task::answers`] (operator answers carried forward to the next
78/// conductor prompt and the next run's instruction). All three are
79/// `#[serde(default)]`, so [`read_path`] accepts anything up to and
80/// including this schema rather than only an exact match — a task written
81/// by a build that only knew about schema 1 has nothing to say about
82/// blocking or answers, and defaulting those fields is exactly as good a
83/// reading as a value that build never had a chance to write.
84pub const SCHEMA: u32 = 6;
85
86/// Who placed the current hold.
87#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
88#[serde(rename_all = "lowercase")]
89pub enum HoldSource {
90 /// An operator used the CLI or web UI.
91 Manual,
92 /// The daemon or conductor placed the hold as part of its own recovery.
93 Machine,
94}
95
96impl HoldSource {
97 /// Short human-facing label for reports and the CLI.
98 pub fn label(self) -> &'static str {
99 match self {
100 Self::Manual => "manual",
101 Self::Machine => "machine",
102 }
103 }
104}
105
106/// Where a task came from. Recorded because "who asked for this" is the first
107/// question about an autonomous run, and the answer is not recoverable later.
108#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
109#[serde(tag = "kind", rename_all = "lowercase")]
110pub enum Source {
111 /// A person, at a terminal or through the web UI.
112 Human,
113 /// An agent inside a run, via `magi task add`. Both ids are recorded so a
114 /// task can be traced back to the exact seat that asked for it.
115 Agent {
116 /// Run the asking agent belonged to.
117 run: String,
118 /// Node it was working in, e.g. `implement` or `review`.
119 node: String,
120 },
121 /// A GitHub issue, imported by number.
122 Issue {
123 /// Issue number.
124 number: u64,
125 /// `owner/repo`, as `gh` reports it.
126 repo: String,
127 },
128}
129
130impl Source {
131 /// Short human-facing label, for lists and the web UI.
132 pub fn label(&self) -> String {
133 match self {
134 Self::Human => "human".to_owned(),
135 Self::Agent { run, node } => format!("{node}@{}", short(run)),
136 Self::Issue { number, .. } => format!("issue #{number}"),
137 }
138 }
139}
140
141/// Where a task is in its life.
142#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
143#[serde(rename_all = "lowercase")]
144pub enum TaskStatus {
145 /// Waiting to be claimed.
146 Queued,
147 /// Claimed by a daemon; a run is in flight.
148 Running,
149 /// A run finished and its gate passed.
150 Done,
151 /// A run finished without passing, and attempts remain.
152 Failed,
153 /// Out of attempts, or held by hand. The loop will not pick it up.
154 Held,
155 /// Waiting on another task or an unanswered question. See
156 /// [`Task::blocked_by`]. Set and cleared by `crate::conduct` and
157 /// `crate::daemon`'s deterministic resolver, never by hand.
158 Blocked,
159}
160
161impl TaskStatus {
162 /// Is this task eligible for a daemon to claim?
163 pub fn runnable(self) -> bool {
164 matches!(self, Self::Queued | Self::Failed)
165 }
166
167 /// Lowercase name, as it appears on disk and in the API.
168 pub fn as_str(self) -> &'static str {
169 match self {
170 Self::Queued => "queued",
171 Self::Running => "running",
172 Self::Done => "done",
173 Self::Failed => "failed",
174 Self::Held => "held",
175 Self::Blocked => "blocked",
176 }
177 }
178}
179
180/// One unit of work.
181#[derive(Debug, Clone, Serialize, Deserialize)]
182#[serde(deny_unknown_fields)]
183pub struct Task {
184 /// On-disk format version.
185 pub schema: u32,
186 /// Task id, e.g. `20260902-140501-a1b2`.
187 pub id: String,
188 /// One line, for lists and notifications.
189 pub title: String,
190 /// The task itself, handed to the graph verbatim.
191 pub instruction: String,
192 /// Repository to work in.
193 pub repo: PathBuf,
194 /// Who asked.
195 pub source: Source,
196 /// Higher runs first; ties break oldest-first so nothing starves.
197 #[serde(default)]
198 pub priority: i32,
199 /// Run this task alone: one implementer, no panel of judges to convince.
200 ///
201 /// `#[serde(default)]` so a queue file written before this field existed
202 /// still reads, as `false` - the ordinary multi-candidate competition,
203 /// unchanged. A task set to `solo` still runs the whole graph; only the
204 /// candidate count the daemon builds it with changes, and
205 /// [`crate::graph::Runner`] already collapses a single-candidate run to
206 /// implement → review → gate → merge on its own (see
207 /// [`crate::graph::Runner::review`]'s doc), so nothing about judging,
208 /// deliberation or voting had to change to support this.
209 #[serde(default)]
210 pub solo: bool,
211 /// Current state.
212 pub status: TaskStatus,
213 /// How many times this task has been claimed.
214 #[serde(default)]
215 pub attempts: usize,
216 /// Runs this task has produced, oldest first.
217 #[serde(default)]
218 pub runs: Vec<String>,
219 /// Why the last attempt did not land.
220 #[serde(default)]
221 pub last_error: Option<String>,
222 /// What a human hold is waiting on.
223 ///
224 /// `None` covers both the ordinary cases: a hold the loop makes itself
225 /// (out of attempts, or the disk gate closed) explains itself through
226 /// [`Task::last_error`] instead, and a human hold nobody bothered to
227 /// explain is still a valid hold. The queue has no way to express a
228 /// dependency between two tasks, so on the occasions a hold really is
229 /// "wait for that other task first", this is the only place that reason
230 /// survives - see [`Task::hold_manual`] and [`Task::release`].
231 ///
232 /// `#[serde(default)]` so a queue file written before this field existed
233 /// still reads, with no reason recorded rather than a parse error.
234 #[serde(default)]
235 pub hold_reason: Option<String>,
236 /// Who placed [`Task::hold_reason`]. `None` is a compatible old record;
237 /// see [`Task::operator_held`] for its deliberately conservative meaning.
238 #[serde(default)]
239 pub hold_source: Option<HoldSource>,
240 /// Diagnostic detail excerpted from the run that led to a hold - what a
241 /// human would have found opening `artifacts/` by hand, not the one-line
242 /// reason in [`Task::last_error`]. Set only when a run's own attempts are
243 /// exhausted and the task becomes [`TaskStatus::Held`]; `daemon` computes
244 /// it from the run's own record, since this module has no notion of a
245 /// run's internals. Bounded in length by the writer - see
246 /// `daemon::diagnostic` - so a verbose run cannot make this file grow
247 /// without limit.
248 ///
249 /// `#[serde(default)]` so a queue file written before this field existed
250 /// still reads, with no diagnostic recorded rather than a parse error.
251 #[serde(default)]
252 pub diagnostic: Option<String>,
253 /// What this task is waiting on: other task ids, unanswered
254 /// `crate::ask::Question` ids, or both. Non-empty exactly when
255 /// [`TaskStatus::Blocked`]; emptying it — see [`Task::unblock`] — is what
256 /// puts the task back at [`TaskStatus::Queued`].
257 ///
258 /// Set by `crate::conduct`'s decisions and cleared deterministically by
259 /// `crate::daemon` as each dependency resolves, never by a person. Never
260 /// `#[serde(default)]` is skipped: a queue file from before this field
261 /// existed has nothing to report here, and an empty list is exactly that.
262 #[serde(default)]
263 pub blocked_by: Vec<String>,
264 /// One line explaining the current [`Task::blocked_by`], written by
265 /// `crate::conduct`. Cleared whenever `blocked_by` empties.
266 #[serde(default)]
267 pub block_reason: Option<String>,
268 /// The status this task had the moment [`Task::block`] most recently
269 /// moved it to [`TaskStatus::Blocked`] — what [`Task::unblock`] restores
270 /// once nothing is left in `blocked_by`, instead of always landing on
271 /// [`TaskStatus::Queued`]. See [`SCHEMA`]'s doc for schema 4 on why this
272 /// exists: an answer to a question `crate::conduct` filed about a
273 /// [`TaskStatus::Held`] task must not itself be what puts the task back
274 /// in the competition queue.
275 ///
276 /// `#[serde(default)]` so a queue file written before this field existed
277 /// reads as `None`; [`Task::unblock`] treats that the same as a task
278 /// blocked straight from `Queued`, unless surviving hold evidence says
279 /// otherwise.
280 #[serde(default)]
281 pub blocked_from: Option<TaskStatus>,
282 /// Questions `crate::conduct` asked about this task that the operator has
283 /// since answered, oldest first — what was asked, and what they said.
284 ///
285 /// A blocking question's id leaves [`Task::blocked_by`] the moment
286 /// [`crate::ask::QuestionStatus::Answered`] is observed, but the id alone
287 /// tells nobody what was decided. This is what carries the answer's
288 /// *content* forward: into the next conductor prompt for this task, and
289 /// into the instruction handed to the next run — see `crate::daemon`'s
290 /// deterministic blocker resolution. Kept for the task's whole life, the
291 /// same as [`Task::runs`]: a release resets attempts, not evidence.
292 #[serde(default)]
293 pub answers: Vec<AnsweredQuestion>,
294 /// Ids of the `crate::triage` questions whose answer has been applied to
295 /// this task. Unlike [`Task::hold_reason`], [`Task::release`] and every
296 /// hold transition leave it alone, so an answer is applied at most once
297 /// however many times the task is held again. See [`SCHEMA`]'s doc for
298 /// schema 5. `#[serde(default)]` so an older record reads as empty.
299 #[serde(default)]
300 pub triage_applied: Vec<String>,
301 /// The operator's "resume" answer to a triage question, kept until the
302 /// task actually runs (or is done) so `crate::conduct` cannot silently
303 /// undo it and `crate::triage` can tell that a hold it sees now came
304 /// *after* the answer. See [`OperatorResume`]. `#[serde(default)]`.
305 #[serde(default)]
306 pub resume_override: Option<OperatorResume>,
307 /// Set by `crate::conduct` when it chooses `Review` recovery for a task
308 /// whose branch survived a blocked run: the branch to reopen with
309 /// `crate::graph::Runner::review` instead of competing from scratch.
310 ///
311 /// Requeues the task the same way [`Task::release`] does, so it is
312 /// picked up by the ordinary loop; `crate::daemon` reads this field once,
313 /// when it actually starts the run, and clears it either way — consumed
314 /// on success, dropped if the branch no longer exists by then. Never set
315 /// from the conductor's own words: `crate::daemon` derives the branch
316 /// name itself from the task's last run, so a hallucinated branch can
317 /// never reach here.
318 #[serde(default)]
319 pub review_branch: Option<String>,
320 /// A release deliberately starts a new competition instead of resuming
321 /// the prior run. History remains as evidence in `runs`.
322 #[serde(default)]
323 pub fresh_start: bool,
324 /// Marked by an operator (`magi task interrupt`) to ask `magi serve` to
325 /// run this one ahead of whatever it already has in flight, once
326 /// `[daemon] pause_for_interrupts` is on - see
327 /// `crate::daemon::advance_interrupt`. Never set by the loop itself, and
328 /// deliberately a different operation from [`Task::set_priority`]: a
329 /// priority only reorders the queue a claim has not reached yet, while
330 /// this asks a run already in flight to park at its next safe boundary
331 /// and step aside. `#[serde(default)]` so a queue file written before
332 /// this field existed still reads, as `false` - no task interrupts
333 /// anything unless asked to, exactly as before.
334 #[serde(default)]
335 pub interrupt: bool,
336 /// Marked by `magi task add --urgent`: `crate::daemon::poll` dispatches
337 /// this task through its own one-slot `urgent_sem` the moment it is
338 /// runnable, in addition to whatever is already running under the
339 /// ordinary `[daemon] max_concurrent_runs` pool - never instead of it,
340 /// and never by pausing or otherwise touching that run. This is the
341 /// opposite direction from [`Task::interrupt`]: that one asks a run
342 /// already in flight to step aside; this one never asks anything to
343 /// step aside, it only spends one additional, temporary concurrency
344 /// slot. The two are independent and may both be set on the same task,
345 /// but this exemption stops at `[daemon] pause_for_interrupts`'s own
346 /// park/resume handoff (75dd): while an interrupt sequence is actively
347 /// parking, running, or resuming - its own, or an unrelated task's -
348 /// `crate::daemon::interrupt_gate` withholds an urgent candidate exactly
349 /// like an ordinary one, never exempted. 75dd's "at most one run, ever,
350 /// at once" guarantee takes precedence, because the alternative is a run
351 /// still genuinely in flight (only *asked* to park, not yet gone) ending
352 /// up alongside a second one this feature let through - the very thing
353 /// that guarantee exists to rule out.
354 ///
355 /// `#[serde(default)]` so a queue file written before this field existed
356 /// still reads, as `false` - no task claims the urgent slot unless asked
357 /// to, exactly as before.
358 #[serde(default)]
359 pub urgent: bool,
360 /// When the task was filed.
361 pub created_at: Timestamp,
362 /// Last change to this file.
363 pub updated_at: Timestamp,
364}
365
366/// A triage "resume" answer and what became of it. See
367/// [`Task::resume_override`].
368#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
369pub struct OperatorResume {
370 /// The triage question the operator answered.
371 pub question_id: String,
372 /// When the answer was applied.
373 pub at: Timestamp,
374 /// The reason `crate::conduct` gave for holding the task again after the
375 /// answer, if it did. The conductor may do this once.
376 #[serde(default)]
377 pub conductor_rehold: Option<String>,
378 /// The operator answered "resume" a second time, to the question about
379 /// that contradiction: the conductor may no longer hold this task.
380 #[serde(default)]
381 pub forced: bool,
382}
383
384/// One question `crate::conduct` asked about a task, and what the operator
385/// said back. See [`Task::answers`].
386#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
387pub struct AnsweredQuestion {
388 /// The question as asked, e.g. [`crate::ask::Question::summary`].
389 pub question: String,
390 /// What the operator answered.
391 pub answer: String,
392}
393
394impl Task {
395 /// File a new task. Persist it with [`Queue::put`].
396 pub fn new(title: String, instruction: String, repo: PathBuf, source: Source) -> Self {
397 let now = Timestamp::now();
398 Self {
399 schema: SCHEMA,
400 id: new_id(),
401 title,
402 instruction,
403 repo,
404 source,
405 priority: 0,
406 solo: false,
407 status: TaskStatus::Queued,
408 attempts: 0,
409 runs: Vec::new(),
410 last_error: None,
411 hold_reason: None,
412 hold_source: None,
413 diagnostic: None,
414 blocked_by: Vec::new(),
415 block_reason: None,
416 blocked_from: None,
417 answers: Vec::new(),
418 triage_applied: Vec::new(),
419 resume_override: None,
420 review_branch: None,
421 fresh_start: false,
422 interrupt: false,
423 urgent: false,
424 created_at: now,
425 updated_at: now,
426 }
427 }
428
429 /// Short form used in reports, matching a run's short id.
430 pub fn short(&self) -> &str {
431 short(&self.id)
432 }
433
434 /// Record that triage question `question_id`'s answer has been applied.
435 pub fn mark_triage_applied(&mut self, question_id: &str) {
436 if !self.triage_applied(question_id) {
437 self.triage_applied.push(question_id.to_owned());
438 }
439 }
440
441 /// Has triage question `question_id`'s answer already been applied?
442 pub fn triage_applied(&self, question_id: &str) -> bool {
443 self.triage_applied.iter().any(|id| id == question_id)
444 }
445
446 /// Record that a run has started for this task.
447 ///
448 /// Clears [`Task::interrupt`]: a mark to run ahead of whatever else is
449 /// in flight is fulfilled the moment this task actually gets its turn,
450 /// dispatched same as any other. Without this, a task whose run fails
451 /// and requeues - still `runnable`, still carrying the mark from its
452 /// first attempt - would keep re-triggering `crate::daemon`'s interrupt
453 /// scheduler and re-parking whatever it interrupted on every later
454 /// boundary, for as long as its attempts hold out, instead of the
455 /// one-shot "let this go next" the mark is meant to be.
456 pub fn start(&mut self, run: String) {
457 self.status = TaskStatus::Running;
458 self.attempts += 1;
459 self.runs.push(run);
460 self.last_error = None;
461 self.fresh_start = false;
462 self.interrupt = false;
463 // The answer has been honoured: the task got its turn.
464 self.resume_override = None;
465 }
466
467 /// Record a successful run.
468 ///
469 /// Both `magi task done` and `POST /api/queue/{id}/done` can close a held
470 /// *or blocked* task directly, with no release in between, so this clears
471 /// `hold_reason` and `blocked_by`/`block_reason` the same way
472 /// [`Task::release`] does. Otherwise a task held for "waiting on 3ed9", or
473 /// blocked on a dependency that never actually finished, and then closed
474 /// as done without ever being released would still read as waiting on
475 /// something in `magi task show` and on its card, after it no longer is.
476 pub fn succeed(&mut self) {
477 self.status = TaskStatus::Done;
478 self.resume_override = None;
479 self.last_error = None;
480 self.hold_reason = None;
481 self.hold_source = None;
482 self.diagnostic = None;
483 self.blocked_by.clear();
484 self.block_reason = None;
485 self.blocked_from = None;
486 }
487
488 /// Record a failed attempt. Out of attempts means held for a human, rather
489 /// than retried until the money runs out.
490 ///
491 /// Clears [`Task::diagnostic`] unconditionally: it belongs to whatever run
492 /// produced it, and a caller that has one for *this* attempt sets it
493 /// itself right after calling this, once it knows the task actually ended
494 /// up [`TaskStatus::Held`] - see `daemon::diagnostic`. Without the clear, a
495 /// task released after a diagnosed hold and then failed again for an
496 /// unrelated, undiagnosed reason (a config error, say) would go on
497 /// showing the previous run's diagnostic as if it explained the new one.
498 pub fn fail(&mut self, why: impl Into<String>, max_attempts: usize) {
499 self.last_error = Some(why.into());
500 self.diagnostic = None;
501 self.status = if self.attempts >= max_attempts {
502 self.hold_source = Some(HoldSource::Machine);
503 TaskStatus::Held
504 } else {
505 TaskStatus::Failed
506 };
507 }
508
509 /// Record an attempt that failed for a reason the task is not responsible
510 /// for - the agent CLIs ran out of quota and the judging panel collapsed.
511 ///
512 /// This refunds the attempt on purpose. A quota window closing at 4am must
513 /// not spend the backlog's retry budget: the operator would come back to a
514 /// queue of held tasks that were never actually judged, and would have to
515 /// release every one by hand to find out which had a real problem. The task
516 /// goes back to `Failed`, which the loop retries, so a reset quota picks the
517 /// work up where it stopped.
518 pub fn stall(&mut self, why: impl Into<String>) {
519 self.last_error = Some(why.into());
520 self.diagnostic = None;
521 self.attempts = self.attempts.saturating_sub(1);
522 self.status = TaskStatus::Failed;
523 }
524
525 /// Whether this held task may only be released by an operator.
526 ///
527 /// Old files did not record a source. Preserve every such hold rather
528 /// than guessing that it was automatic and risking duplicate work. New
529 /// automatic holds record [`HoldSource::Machine`] and remain recoverable.
530 pub fn operator_held(&self) -> bool {
531 self.status == TaskStatus::Held && !matches!(self.hold_source, Some(HoldSource::Machine))
532 }
533
534 /// Take this task out of the loop's reach by an operator action.
535 ///
536 /// Clears `blocked_by`/`block_reason` unconditionally, the same as
537 /// [`Task::release`] and for the same reason its own comment already
538 /// gives: a human choosing to hold a *blocked* task overrides its wait
539 /// outright, the same as it overrides an ordinary hold. Without this, a
540 /// task held straight out of [`TaskStatus::Blocked`] - the web UI's "Hold"
541 /// button is reachable on a blocked task, same as "Mark done" - kept
542 /// reading as still waiting on a dependency it no longer had any claim on.
543 pub fn hold_manual(&mut self, reason: Option<String>) {
544 self.status = TaskStatus::Held;
545 if reason.is_some() {
546 self.hold_reason = reason;
547 }
548 self.hold_source = Some(HoldSource::Manual);
549 self.blocked_by.clear();
550 self.block_reason = None;
551 self.blocked_from = None;
552 }
553
554 /// Take this task out of the loop's reach during automatic recovery.
555 ///
556 /// Clears `blocked_by`/`block_reason` for the same reason
557 /// [`Task::hold_manual`] does.
558 pub fn hold_machine(&mut self, reason: Option<String>) {
559 self.status = TaskStatus::Held;
560 if reason.is_some() {
561 self.hold_reason = reason;
562 }
563 self.hold_source = Some(HoldSource::Machine);
564 self.blocked_by.clear();
565 self.block_reason = None;
566 self.blocked_from = None;
567 }
568
569 /// Block this task on other task ids and/or open question ids, chosen by
570 /// `crate::conduct`. Pure: the caller still owns writing it back with
571 /// [`Queue::put`].
572 ///
573 /// Records [`Task::blocked_from`] the first time this moves the task into
574 /// [`TaskStatus::Blocked`], and leaves it alone on a later call that adds
575 /// or replaces `blocked_by` while the task is already `Blocked` - a
576 /// second question about an already-blocked task must not overwrite the
577 /// status it should eventually return to with `Blocked` itself.
578 pub fn block(&mut self, blocked_by: Vec<String>, reason: Option<String>) {
579 if self.status != TaskStatus::Blocked {
580 self.blocked_from = Some(self.status);
581 }
582 self.status = TaskStatus::Blocked;
583 self.blocked_by = blocked_by;
584 self.block_reason = reason;
585 }
586
587 /// Remove one resolved dependency (a task id that became [`TaskStatus::Done`],
588 /// or a question id that became [`crate::ask::QuestionStatus::Answered`]).
589 /// Once nothing is left in [`Task::blocked_by`], the task returns to
590 /// whatever [`Task::blocked_from`] recorded - deciding *why* a task was
591 /// blocked was `crate::conduct`'s job, but noticing a dependency resolved
592 /// needs no model at all, and restoring the status it interrupted needs
593 /// nothing more than what `block` already wrote down.
594 ///
595 /// A task blocked while `Running` restores to [`TaskStatus::Queued`]
596 /// instead: whatever process was running it is gone by the time this
597 /// runs, so there is nothing left to resume. A task with no recorded
598 /// `blocked_from` - a pre-schema-4 record, or one blocked before this
599 /// field existed - falls back to [`TaskStatus::Held`] when it still
600 /// carries hold evidence ([`Task::hold_reason`] or [`Task::hold_source`],
601 /// neither ever cleared by `block`), and to `Queued` otherwise: the same
602 /// choice `block` itself would have recorded, reconstructed from what
603 /// survived.
604 ///
605 /// A no-op, on purpose, for a task that is not [`TaskStatus::Blocked`]:
606 /// `crate::daemon`'s deterministic resolver runs over every task on every
607 /// poll, and a task that moved on for some other reason must not be
608 /// dragged back by a stale id it still happens to carry.
609 pub fn unblock(&mut self, resolved_id: &str) {
610 if self.status != TaskStatus::Blocked {
611 return;
612 }
613 self.blocked_by.retain(|id| id != resolved_id);
614 if self.blocked_by.is_empty() {
615 self.status = match self.blocked_from {
616 Some(TaskStatus::Running) => TaskStatus::Queued,
617 Some(other) => other,
618 None if self.hold_reason.is_some() || self.hold_source.is_some() => {
619 TaskStatus::Held
620 }
621 None => TaskStatus::Queued,
622 };
623 self.block_reason = None;
624 self.blocked_from = None;
625 }
626 }
627
628 /// Record that a question `crate::conduct` asked about this task has been
629 /// answered, so the answer's content — not just the fact that the
630 /// question is gone — reaches the next conductor prompt and the next
631 /// run's instruction. See [`Task::answers`].
632 pub fn record_answer(&mut self, question: String, answer: String) {
633 self.answers.push(AnsweredQuestion { question, answer });
634 }
635
636 /// Requeue this task to reopen its last run as a review-only pass against
637 /// `branch` (`crate::graph::Runner::review`) rather than competing from
638 /// scratch. See [`Task::review_branch`].
639 pub fn request_review(&mut self, branch: String) {
640 self.release();
641 self.review_branch = Some(branch);
642 }
643
644 /// Requeue after a conductor chose a new competition. Unlike an ordinary
645 /// operator release, this deliberately does not resume the old run.
646 pub fn requeue(&mut self) {
647 self.release();
648 self.fresh_start = true;
649 }
650
651 /// Change how urgently this task should run next.
652 ///
653 /// Refused once the task is `running`: priority only feeds the sort
654 /// [`Queue::next_runnable`] does over tasks waiting to be claimed, and a
655 /// running task has already left that pool. Accepting the write anyway
656 /// would look like it worked while changing nothing until - and unless -
657 /// this attempt fails and the task becomes runnable again, which is a
658 /// surprise the phone should not hand back as a success.
659 pub fn set_priority(&mut self, priority: i32) -> Result<()> {
660 if self.status == TaskStatus::Running {
661 bail!(
662 "task {} is running; its priority cannot be changed until \
663 this attempt finishes",
664 self.short()
665 );
666 }
667 self.priority = priority;
668 Ok(())
669 }
670
671 /// Mark (or unmark) this task to interrupt whatever `magi serve` already
672 /// has in flight, once `[daemon] pause_for_interrupts` is on. See
673 /// [`Task::interrupt`].
674 ///
675 /// Setting it is restricted to a task the loop could pick up on its own
676 /// right now - [`TaskStatus::runnable`] - for the same reason as
677 /// [`Task::set_priority`]: a task already `running` has been claimed, and
678 /// a task that is `done`, `held`, or `blocked` is not going to compete
679 /// for the daemon's attention regardless of this flag. Unlike priority,
680 /// this is never silently inert while `running` - it is refused outright,
681 /// because the entire feature this flag drives (`crate::daemon`'s
682 /// interrupt scheduler) is scoped to tasks still waiting to be claimed.
683 /// Clearing it back to `false` carries no such risk and is always
684 /// allowed, including on a task that moved on since it was set.
685 pub fn set_interrupt(&mut self, interrupt: bool) -> Result<()> {
686 if interrupt && !self.status.runnable() {
687 bail!(
688 "task {} is {}; only a queued or failed task can be marked \
689 to interrupt",
690 self.short(),
691 self.status.as_str()
692 );
693 }
694 self.interrupt = interrupt;
695 Ok(())
696 }
697
698 /// Replace this task's title and instruction wholesale.
699 ///
700 /// Restricted to `queued` and `held`. A `running` task's instruction has
701 /// already been handed to the graph, so a run in flight and the file on
702 /// disk must not be allowed to disagree about what was asked; a `done` or
703 /// `failed` task is a record of what actually happened and editing it
704 /// after the fact would falsify that record. `id`, `created_at`,
705 /// `source`, and `runs` are left untouched on purpose - an edit stands in
706 /// for "delete and refile", and keeping the id, the timestamp, the
707 /// attribution, and the run history is the entire reason it exists
708 /// instead.
709 pub fn edit(&mut self, title: String, instruction: String) -> Result<()> {
710 if !matches!(self.status, TaskStatus::Queued | TaskStatus::Held) {
711 bail!(
712 "task {} is {}; only a queued or held task's instruction can \
713 be edited",
714 self.short(),
715 self.status.as_str()
716 );
717 }
718 self.title = title;
719 self.instruction = instruction;
720 Ok(())
721 }
722
723 /// Record a run that produced a pull request without merging it.
724 ///
725 /// The task is held rather than retried, and it costs no further attempt
726 /// either way. The work the task asked for exists: it is sitting on a
727 /// branch, in a pull request, waiting for CI or for a person. Retrying
728 /// would spend the whole competition budget a second time and then race a
729 /// second branch against the pull request the first one opened - which is
730 /// exactly what happened to run 01c2, whose finished and green pull request
731 /// was re-competed from scratch four seconds after it opened.
732 ///
733 /// A pull request nobody merged is a request for a person, not a failure.
734 pub fn handed_off(&mut self, why: impl Into<String>) {
735 self.last_error = Some(why.into());
736 self.diagnostic = None;
737 self.status = TaskStatus::Held;
738 self.hold_source = Some(HoldSource::Machine);
739 }
740
741 /// Put a held or finished task back in line, with its attempt count reset
742 /// so a release is a real second chance rather than an instant re-hold.
743 /// The run history is kept: attempts reset, evidence does not.
744 pub fn release(&mut self) {
745 self.status = TaskStatus::Queued;
746 self.attempts = 0;
747 self.last_error = None;
748 // Otherwise the next person who holds this task reads a reason that
749 // belonged to whatever it was waiting on last time.
750 self.hold_reason = None;
751 self.hold_source = None;
752 self.diagnostic = None;
753 // A release also un-blocks: the dependency or question `blocked_by`
754 // named may still be unresolved, but a human (or `crate::conduct`)
755 // choosing to release the task overrides that wait outright, the same
756 // as it overrides an ordinary hold.
757 self.blocked_by.clear();
758 self.block_reason = None;
759 self.blocked_from = None;
760 self.review_branch = None;
761 self.fresh_start = false;
762 }
763}
764
765/// A queue on disk.
766#[derive(Debug, Clone)]
767pub struct Queue {
768 root: PathBuf,
769}
770
771impl Queue {
772 /// The operator's queue, `<home>/queue`.
773 pub fn open() -> Self {
774 Self::at(crate::run::home().join("queue"))
775 }
776
777 /// A queue at an explicit root. Tests use this; so could an operator who
778 /// wants a queue per project.
779 pub fn at(root: PathBuf) -> Self {
780 Self { root }
781 }
782
783 /// Directory holding the task files.
784 pub fn root(&self) -> &Path {
785 &self.root
786 }
787
788 /// Path for one task id.
789 pub fn path_of(&self, id: &str) -> PathBuf {
790 self.root.join(format!("{id}.json"))
791 }
792
793 /// Write a task, atomically, so a daemon killed mid-write leaves the
794 /// previous state readable rather than a truncated file.
795 pub fn put(&self, task: &mut Task) -> Result<()> {
796 task.updated_at = Timestamp::now();
797 std::fs::create_dir_all(&self.root)
798 .with_context(|| format!("create {}", self.root.display()))?;
799 let body = serde_json::to_string_pretty(task).context("serialize task")?;
800 let path = self.path_of(&task.id);
801 let tmp = path.with_extension("json.tmp");
802 std::fs::write(&tmp, &body).with_context(|| format!("write {}", tmp.display()))?;
803 std::fs::rename(&tmp, &path).with_context(|| format!("replace {}", path.display()))?;
804 // A machine hold is news for the notification centre, filed beside
805 // this queue (`<home>/queue` -> `<home>/notifications`) rather than
806 // through a process-global, so a queue in a temp directory notifies
807 // into that directory.
808 if let (Some(notice), Some(home)) = (
809 crate::notices::task_held(task),
810 self.root.parent().filter(|p| !p.as_os_str().is_empty()),
811 ) {
812 crate::notices::raise_in(home, notice);
813 }
814 Ok(())
815 }
816
817 /// Load a task by id or unambiguous id prefix.
818 pub fn get(&self, id: &str) -> Result<Task> {
819 let resolved = self.resolve_id(id)?;
820 read_path(&self.path_of(&resolved))
821 }
822
823 /// Remove a task, and the claim lock that belongs to it.
824 ///
825 /// `in_flight` comes from the caller — a live daemon's heartbeat naming
826 /// this task — because the task's own `running` status cannot answer the
827 /// question. A daemon killed mid-competition leaves the status at
828 /// `running` and an orphaned `.lock` behind, and a guard that trusted
829 /// either would make the task undeletable for good: the phone showed
830 /// exactly that, refusing a task whose daemon had been gone for an hour.
831 ///
832 /// So the lock is removed with the task rather than respected. Any lock
833 /// still there once no live daemon claims the task is by definition stale,
834 /// and leaving it would make a deleted task look claimed to
835 /// [`Queue::claim`] and to whoever reads the directory.
836 ///
837 /// Anything still `blocked` on the id just deleted is quarantined to a
838 /// machine hold in the same call - see [`Removal::quarantined`] - rather
839 /// than left to wait on a dependency that no longer exists. Best-effort:
840 /// a dependent claimed by something else right now, or one whose write
841 /// fails, is simply left for `crate::daemon::resolve_blockers`'s own poll
842 /// (or `crate::triage::run_once`) to catch on its own next pass, and does
843 /// not fail this removal.
844 ///
845 /// `questions` is the store [`missing_blockers`] checks a `blocked_by` id
846 /// against before calling it gone - the same store the caller already
847 /// resolves `id`'s own home from, passed in rather than reopened here so
848 /// a test queue at an explicit root is never quarantined against the
849 /// operator's real questions directory.
850 pub fn remove(&self, id: &str, in_flight: bool, questions: &Questions) -> Result<Removal> {
851 let resolved = self.resolve_id(id)?;
852 if in_flight {
853 bail!("task {resolved} is being run by a live daemon right now");
854 }
855 let path = self.path_of(&resolved);
856 std::fs::remove_file(&path).with_context(|| format!("remove {}", path.display()))?;
857 let lock = self.lock_path(&resolved);
858 if let Err(e) = std::fs::remove_file(&lock) {
859 if e.kind() != std::io::ErrorKind::NotFound {
860 return Err(e).with_context(|| format!("remove {}", lock.display()));
861 }
862 }
863 let quarantined = self.quarantine_dependents_of(&resolved, questions);
864 Ok(Removal {
865 id: resolved,
866 quarantined,
867 })
868 }
869
870 /// Move every `blocked` task naming `dependency` in its own `blocked_by`
871 /// to a machine hold, now that `dependency`'s own file is gone. See
872 /// [`Queue::remove`]'s own doc for why this is best-effort.
873 fn quarantine_dependents_of(&self, dependency: &str, questions: &Questions) -> Vec<String> {
874 let mut quarantined = Vec::new();
875 for listed in self.list() {
876 if listed.status != TaskStatus::Blocked
877 || !listed.blocked_by.iter().any(|b| b == dependency)
878 {
879 continue;
880 }
881 let Ok(_claim) = self.claim(&listed.id) else {
882 continue;
883 };
884 let Ok(mut task) = self.get(&listed.id) else {
885 continue;
886 };
887 if task.status != TaskStatus::Blocked
888 || !task.blocked_by.iter().any(|b| b == dependency)
889 {
890 continue;
891 }
892 let missing = missing_blockers(self, questions, &task.blocked_by);
893 task.hold_machine(Some(missing_blocker_hold_reason(
894 &task.blocked_by,
895 &missing,
896 )));
897 if self.put(&mut task).is_ok() {
898 quarantined.push(task.id.clone());
899 }
900 }
901 quarantined
902 }
903
904 /// Path of the claim lock for a task. One definition, so `claim` and
905 /// `remove` cannot end up naming different files.
906 fn lock_path(&self, id: &str) -> PathBuf {
907 self.root.join(format!("{id}.lock"))
908 }
909
910 /// Every task on disk, highest priority first and newest first within a
911 /// priority. This is what `magi task list` and `GET /api/queue` print, so
912 /// a raised priority has to move a task here the moment it is saved, not
913 /// only in [`Queue::next_runnable`]'s own ordering - the operator reading
914 /// the backlog and the loop about to drain it must agree on what "first"
915 /// means. Every existing task defaults to priority 0, so this is a no-op
916 /// change from the old newest-first order for a queue nobody has
917 /// reprioritised.
918 ///
919 /// Unreadable files are skipped rather than fatal: one corrupt task must
920 /// not take the queue - or the web UI, or an unattended daemon - down
921 /// with it.
922 pub fn list(&self) -> Vec<Task> {
923 let mut tasks: Vec<Task> = std::fs::read_dir(&self.root)
924 .into_iter()
925 .flatten()
926 .flatten()
927 .map(|e| e.path())
928 .filter(|p| p.extension().is_some_and(|x| x == "json"))
929 .filter_map(|p| read_path(&p).ok())
930 .collect();
931 tasks.sort_unstable_by(|a, b| b.priority.cmp(&a.priority).then_with(|| b.id.cmp(&a.id)));
932 tasks
933 }
934
935 /// The task a daemon should run next, or `None` when the queue is idle.
936 ///
937 /// Highest priority first, oldest first within a priority, so a burst of
938 /// agent-filed work cannot starve the task a human filed this morning.
939 pub fn next_runnable(&self) -> Option<Task> {
940 let mut runnable: Vec<Task> = self
941 .list()
942 .into_iter()
943 .filter(|t| t.status.runnable())
944 .collect();
945 runnable.sort_unstable_by(|a, b| b.priority.cmp(&a.priority).then(a.id.cmp(&b.id)));
946 runnable.into_iter().next()
947 }
948
949 /// Take exclusive ownership of a task.
950 ///
951 /// The lock is a `create_new` file next to the task, which is atomic on
952 /// every platform magi targets. It exists so two daemons - or a daemon and
953 /// a human running `magi run` - cannot drive one task into two competing
954 /// runs. The returned guard releases on drop, including on panic.
955 pub fn claim(&self, id: &str) -> Result<Claim> {
956 std::fs::create_dir_all(&self.root)
957 .with_context(|| format!("create {}", self.root.display()))?;
958 let path = self.lock_path(id);
959 match std::fs::OpenOptions::new()
960 .write(true)
961 .create_new(true)
962 .open(&path)
963 {
964 Ok(mut f) => {
965 use std::io::Write as _;
966 // Best effort: the pid is for the human looking at a stale lock.
967 let _ = writeln!(f, "{}", std::process::id());
968 Ok(Claim { path })
969 }
970 Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
971 bail!("task {id} is already claimed ({} exists)", path.display())
972 }
973 Err(e) => Err(e).with_context(|| format!("lock {}", path.display())),
974 }
975 }
976
977 /// Expand an id prefix to exactly one task id.
978 pub fn resolve_id(&self, prefix: &str) -> Result<String> {
979 if self.path_of(prefix).is_file() {
980 return Ok(prefix.to_owned());
981 }
982 let hits: Vec<String> = self
983 .list()
984 .into_iter()
985 .map(|t| t.id)
986 .filter(|id| id.starts_with(prefix) || id.ends_with(prefix))
987 .collect();
988 match hits.len() {
989 1 => Ok(hits.into_iter().next().expect("exactly one hit")),
990 0 => bail!("no task matches `{prefix}`"),
991 _ => bail!(
992 "`{prefix}` matches {} tasks: {}",
993 hits.len(),
994 hits.join(", ")
995 ),
996 }
997 }
998
999 /// Change detection token for the queue.
1000 ///
1001 /// Combines file names and modification times of all task files in the
1002 /// queue, so adding, modifying, or deleting any task — even an older one —
1003 /// moves the revision and notifies connected clients via the change stream.
1004 /// Returns 0 when the queue is completely empty.
1005 pub fn revision(&self) -> u64 {
1006 use std::hash::{Hash as _, Hasher as _};
1007
1008 let mut entries: Vec<(String, u64)> = std::fs::read_dir(&self.root)
1009 .into_iter()
1010 .flatten()
1011 .flatten()
1012 .filter(|e| e.path().extension().is_some_and(|ext| ext == "json"))
1013 .filter_map(|e| {
1014 let name = e.file_name().to_string_lossy().into_owned();
1015 let mtime = e
1016 .metadata()
1017 .ok()?
1018 .modified()
1019 .ok()?
1020 .duration_since(std::time::UNIX_EPOCH)
1021 .ok()?
1022 .as_millis() as u64;
1023 Some((name, mtime))
1024 })
1025 .collect();
1026
1027 if entries.is_empty() {
1028 return 0;
1029 }
1030
1031 entries.sort_unstable();
1032 let mut hasher = std::hash::DefaultHasher::new();
1033 for (name, mtime) in &entries {
1034 name.hash(&mut hasher);
1035 mtime.hash(&mut hasher);
1036 }
1037 let h = hasher.finish();
1038 if h == 0 { 1 } else { h }
1039 }
1040}
1041
1042/// What [`Queue::remove`] did, beyond deleting the named task's own file.
1043#[derive(Debug, Clone)]
1044pub struct Removal {
1045 /// The id actually removed - `id` expanded from a prefix, if it was one.
1046 pub id: String,
1047 /// Every `blocked` task that named [`Removal::id`] in its own
1048 /// `blocked_by` and was moved to a machine hold as a result, rather than
1049 /// left waiting on a dependency this call just erased.
1050 pub quarantined: Vec<String>,
1051}
1052
1053/// Exclusive ownership of a task, released on drop.
1054#[derive(Debug)]
1055pub struct Claim {
1056 path: PathBuf,
1057}
1058
1059impl Drop for Claim {
1060 fn drop(&mut self) {
1061 let _ = std::fs::remove_file(&self.path);
1062 }
1063}
1064
1065/// The first line of a task, trimmed to a title. Used when the caller gives a
1066/// body but no title, which is the normal case for an agent piping a file in.
1067pub fn title_from(instruction: &str, max: usize) -> String {
1068 // The first non-blank line, whatever it is. A markdown heading is the
1069 // task's own summary - agents pipe in `# Rework the config loader` and mean
1070 // exactly that - so it is preferred over the prose beneath it rather than
1071 // skipped as decoration. Leading list and heading markers are stripped
1072 // because they are syntax, not words.
1073 let line = instruction
1074 .lines()
1075 .map(str::trim)
1076 .find(|l| !l.is_empty())
1077 .unwrap_or("(empty task)")
1078 .trim_start_matches(['#', '-', '*', '>', ' '])
1079 .trim();
1080 if line.is_empty() {
1081 return "(empty task)".to_owned();
1082 }
1083 if line.chars().count() <= max {
1084 return line.to_owned();
1085 }
1086 let head: String = line.chars().take(max.saturating_sub(1)).collect();
1087 format!("{head}…")
1088}
1089
1090fn read_path(path: &Path) -> Result<Task> {
1091 let body = std::fs::read_to_string(path).with_context(|| format!("read {}", path.display()))?;
1092 let task: Task =
1093 serde_json::from_str(&body).with_context(|| format!("parse {}", path.display()))?;
1094 // Greater-than, not not-equal: every field added since schema 1 carries
1095 // `#[serde(default)]`, so an older task has nothing to say about it and
1096 // defaulting is exactly as good a reading as a value that build never had
1097 // a chance to write. Only a schema *ahead* of this build - a meaning it
1098 // cannot possibly know - is refused rather than guessed at.
1099 if task.schema > SCHEMA {
1100 bail!(
1101 "task {} was written by a different magi (schema {}, this build \
1102 speaks {SCHEMA})",
1103 task.id,
1104 task.schema
1105 );
1106 }
1107 Ok(task)
1108}
1109
1110/// Ids inside a `blocked_by` list that name neither an existing task file nor
1111/// an existing question file - a dependency deleted (`magi task rm`, or by
1112/// hand) while something was still waiting on it.
1113///
1114/// Existence is decided by [`Queue::path_of`]/[`Questions::path_of`]
1115/// `is_file()` alone, never by [`Queue::get`]/[`Questions::get`] succeeding:
1116/// those also fail on a merely unreadable file - mid-write, corrupt, or from
1117/// a schema ahead of this build (see [`read_path`]) - and misreading "cannot
1118/// read it right now" as "it was deleted" would quarantine a task over a
1119/// transient failure. `blocked_by` always carries a full id, written by
1120/// `crate::conduct` or `crate::triage` from a real task's or question's own
1121/// `id`/`short`, never a prefix a caller typed - so the exact-path check is
1122/// complete on its own, with no [`Queue::resolve_id`] fallback needed.
1123pub fn missing_blockers(
1124 queue: &Queue,
1125 questions: &Questions,
1126 blocked_by: &[String],
1127) -> Vec<String> {
1128 blocked_by
1129 .iter()
1130 .filter(|id| !queue.path_of(id).is_file() && !questions.path_of(id).is_file())
1131 .cloned()
1132 .collect()
1133}
1134
1135/// The `hold_reason` text for a task quarantined because one or more of its
1136/// `blocked_by` ids no longer exist. Shared by `crate::daemon::resolve_blockers`,
1137/// `crate::triage::run_once`, and [`Queue::remove`]'s own dependent
1138/// quarantine, so the three call sites read as the same event to an operator
1139/// looking at `magi task show` rather than three different wordings for it.
1140///
1141/// Names the full original `blocked_by` list, not just `missing` - a task
1142/// quarantined here can also have named a dependency that was still
1143/// perfectly valid, and [`Task::hold_machine`] clears `blocked_by` on the way
1144/// in, so this text is the only place that information survives for an
1145/// operator deciding whether to release the task outright.
1146pub fn missing_blocker_hold_reason(blocked_by: &[String], missing: &[String]) -> String {
1147 format!(
1148 "blocked on {} but {} no longer exist(s) on disk - see `magi task triage`",
1149 blocked_by.join(", "),
1150 missing.join(", "),
1151 )
1152}
1153
1154fn short(id: &str) -> &str {
1155 id.split('-').next_back().unwrap_or(id)
1156}
1157
1158fn new_id() -> String {
1159 let stamp = jiff::Zoned::now().strftime("%Y%m%d-%H%M%S");
1160 let seed = crate::rng::entropy();
1161 format!("{stamp}-{:04x}", (seed ^ (seed >> 32)) & 0xffff)
1162}
1163
1164#[cfg(test)]
1165mod tests {
1166 use super::*;
1167
1168 #[test]
1169 fn triage_applied_survives_release_and_old_records_read_as_empty() {
1170 let mut t = Task::new(
1171 "t".to_owned(),
1172 "i".to_owned(),
1173 PathBuf::from("r"),
1174 Source::Human,
1175 );
1176 t.mark_triage_applied("q1");
1177 t.mark_triage_applied("q1");
1178 t.hold_machine(Some("x".to_owned()));
1179 t.release();
1180 assert_eq!(t.triage_applied, ["q1"]);
1181 assert!(t.triage_applied("q1") && !t.triage_applied("q2"));
1182
1183 let mut v = serde_json::to_value(&t).unwrap();
1184 v.as_object_mut().unwrap().remove("triage_applied");
1185 let old: Task = serde_json::from_value(v).unwrap();
1186 assert!(old.triage_applied.is_empty());
1187 }
1188
1189 /// A queue of its own, with no process-global state - which is the point of
1190 /// `Queue::at`, and why these can run in parallel.
1191 fn queue() -> (tempfile::TempDir, Queue) {
1192 let dir = tempfile::tempdir().unwrap();
1193 let q = Queue::at(dir.path().join("queue"));
1194 (dir, q)
1195 }
1196
1197 #[test]
1198 fn putting_a_machine_held_task_files_a_notification_beside_the_queue() {
1199 let dir = tempfile::tempdir().unwrap();
1200 let q = Queue::at(dir.path().join("queue"));
1201 let mut t = task("held");
1202 q.put(&mut t).unwrap();
1203 assert_eq!(
1204 crate::notices::Notices::at(dir.path().join("notifications"))
1205 .list()
1206 .len(),
1207 0
1208 );
1209 t.hold_machine(Some("out of attempts".to_owned()));
1210 q.put(&mut t).unwrap();
1211 let listed = crate::notices::Notices::at(dir.path().join("notifications")).list();
1212 assert_eq!(listed.len(), 1);
1213 assert!(listed[0].message.contains("out of attempts"));
1214 }
1215
1216 fn task(title: &str) -> Task {
1217 Task::new(
1218 title.to_owned(),
1219 format!("do {title}"),
1220 PathBuf::from("."),
1221 Source::Human,
1222 )
1223 }
1224
1225 #[test]
1226 fn a_markdown_heading_is_the_title_not_decoration() {
1227 // A task file's heading is the summary its author already wrote, so it
1228 // beats the prose underneath. Getting this backwards was visible in the
1229 // first smoke test: a task titled "# Rework the config loader" listed
1230 // as "It re-reads the file on every lookup".
1231 assert_eq!(
1232 title_from("# Rework the config loader\n\nIt re-reads it.\n", 40),
1233 "Rework the config loader"
1234 );
1235 assert_eq!(title_from("- fix the thing", 40), "fix the thing");
1236 assert_eq!(title_from("> quoted task", 40), "quoted task");
1237 // Nothing usable at all still has to produce something printable.
1238 assert_eq!(title_from(" \n\n", 40), "(empty task)");
1239 assert_eq!(title_from("###\n", 40), "(empty task)");
1240 }
1241
1242 #[test]
1243 fn a_long_title_is_elided_by_characters_not_bytes() {
1244 // Byte truncation would split a multi-byte character and panic.
1245 let long = "課題".repeat(30);
1246 let title = title_from(&long, 10);
1247 assert_eq!(title.chars().count(), 10);
1248 assert!(title.ends_with('…'));
1249 }
1250
1251 #[test]
1252 fn priority_wins_and_ties_break_oldest_first() {
1253 let (_dir, q) = queue();
1254 let mut a = task("first");
1255 let mut b = task("second");
1256 let mut c = task("urgent");
1257 // Ids carry a timestamp, so force a known order.
1258 a.id = "20260101-000001-aaaa".to_owned();
1259 b.id = "20260101-000002-bbbb".to_owned();
1260 c.id = "20260101-000003-cccc".to_owned();
1261 c.priority = 5;
1262 for t in [&mut a, &mut b, &mut c] {
1263 q.put(t).unwrap();
1264 }
1265
1266 // Priority first...
1267 assert_eq!(q.next_runnable().unwrap().id, c.id);
1268 c.hold_machine(None);
1269 q.put(&mut c).unwrap();
1270 // ...then oldest, so a burst of new work cannot starve older work.
1271 assert_eq!(q.next_runnable().unwrap().id, a.id);
1272 assert_eq!(q.list().len(), 3, "b is still waiting its turn");
1273 }
1274
1275 #[test]
1276 fn a_blocked_task_never_starves_another_runnable_one() {
1277 let (_dir, q) = queue();
1278 let mut blocked = task("blocked");
1279 blocked.block(vec!["something".to_owned()], None);
1280 q.put(&mut blocked).unwrap();
1281
1282 let mut runnable = task("free to go");
1283 q.put(&mut runnable).unwrap();
1284
1285 let next = q.next_runnable().expect("a runnable task is still offered");
1286 assert_eq!(next.id, runnable.id);
1287 }
1288
1289 #[test]
1290 fn a_held_task_is_never_offered_to_the_loop() {
1291 let (_dir, q) = queue();
1292 let mut t = task("held");
1293 q.put(&mut t).unwrap();
1294 assert!(q.next_runnable().is_some());
1295
1296 t.hold_machine(None);
1297 q.put(&mut t).unwrap();
1298 assert!(
1299 q.next_runnable().is_none(),
1300 "a held task must wait for a human"
1301 );
1302
1303 // A failed task, by contrast, is exactly what the loop should retry.
1304 t.status = TaskStatus::Failed;
1305 q.put(&mut t).unwrap();
1306 assert!(q.next_runnable().is_some());
1307 }
1308
1309 #[test]
1310 fn attempts_are_capped_and_then_the_task_is_held() {
1311 let mut t = task("doomed");
1312
1313 t.start("run-1".to_owned());
1314 t.fail("gate red", 2);
1315 assert_eq!(t.status, TaskStatus::Failed, "one attempt of two: retry");
1316
1317 t.start("run-2".to_owned());
1318 t.fail("gate red", 2);
1319 assert_eq!(
1320 t.status,
1321 TaskStatus::Held,
1322 "out of attempts: stop spending money on it"
1323 );
1324 assert_eq!(t.runs, ["run-1", "run-2"]);
1325 assert_eq!(t.last_error.as_deref(), Some("gate red"));
1326 }
1327
1328 #[test]
1329 fn a_quota_stall_is_refunded_so_the_backlog_survives_the_night() {
1330 let mut t = task("stalled by quota");
1331
1332 t.start("run-1".to_owned());
1333 assert_eq!(t.attempts, 1);
1334 t.stall("judge-1, judge-2 out of quota");
1335 assert_eq!(
1336 t.attempts, 0,
1337 "a closed quota window must not spend the task's retry budget"
1338 );
1339 assert_eq!(t.status, TaskStatus::Failed, "the loop should retry it");
1340 assert_eq!(
1341 t.last_error.as_deref(),
1342 Some("judge-1, judge-2 out of quota")
1343 );
1344
1345 // A task can therefore stall all night and still get its real attempts
1346 // once the quota resets - which is the whole point.
1347 for _ in 0..20 {
1348 t.start("run-n".to_owned());
1349 t.stall("still out of quota");
1350 }
1351 t.start("run-real".to_owned());
1352 t.fail("gate red", 2);
1353 assert_eq!(
1354 t.status,
1355 TaskStatus::Failed,
1356 "the first attempt that was really judged is attempt one"
1357 );
1358 }
1359
1360 #[test]
1361 fn releasing_a_held_task_gives_it_a_real_second_chance() {
1362 let mut t = task("retry me");
1363 t.start("run-1".to_owned());
1364 t.fail("gate red", 1);
1365 assert_eq!(t.status, TaskStatus::Held);
1366
1367 t.release();
1368 assert_eq!(t.status, TaskStatus::Queued);
1369 // Without resetting attempts the next failure would re-hold at once,
1370 // and a release would be a no-op the operator cannot see.
1371 assert_eq!(t.attempts, 0);
1372 assert!(t.last_error.is_none());
1373 assert_eq!(
1374 t.runs.len(),
1375 1,
1376 "history is kept: attempts reset, evidence does not"
1377 );
1378 }
1379
1380 #[test]
1381 fn a_hold_reason_survives_and_a_release_clears_it() {
1382 let mut t = task("waiting on something else");
1383 t.hold_manual(Some(
1384 "waiting for 20260101-000000-aaaa to land first".to_owned(),
1385 ));
1386 assert_eq!(t.status, TaskStatus::Held);
1387 assert_eq!(
1388 t.hold_reason.as_deref(),
1389 Some("waiting for 20260101-000000-aaaa to land first")
1390 );
1391
1392 // Holding again with no reason must not erase the one already there.
1393 t.hold_manual(None);
1394 assert_eq!(
1395 t.hold_reason.as_deref(),
1396 Some("waiting for 20260101-000000-aaaa to land first"),
1397 "a bare re-hold keeps whatever a human already wrote down"
1398 );
1399
1400 // A hold with no reason at all is still an ordinary, allowed hold.
1401 let mut plain = task("no reason given");
1402 plain.hold_manual(None);
1403 assert_eq!(plain.status, TaskStatus::Held);
1404 assert!(plain.hold_reason.is_none());
1405
1406 t.release();
1407 assert_eq!(t.status, TaskStatus::Queued);
1408 assert!(
1409 t.hold_reason.is_none(),
1410 "a stale reason must not greet the next person who holds this task"
1411 );
1412 }
1413
1414 #[test]
1415 fn closing_a_held_task_as_done_clears_its_hold_reason_too() {
1416 // `done` can close a held task directly - neither `magi task done`
1417 // nor `POST /api/queue/{id}/done` requires a release first - so a
1418 // task held for "waiting on 3ed9" and then closed without ever being
1419 // released must not still read as waiting on it afterwards.
1420 let mut t = task("landed by hand while held");
1421 t.hold_manual(Some("waiting on 3ed9".to_owned()));
1422 assert_eq!(t.hold_reason.as_deref(), Some("waiting on 3ed9"));
1423
1424 t.succeed();
1425 assert_eq!(t.status, TaskStatus::Done);
1426 assert!(
1427 t.hold_reason.is_none(),
1428 "a done task cannot still be waiting on something"
1429 );
1430 }
1431
1432 #[test]
1433 fn holding_or_closing_a_blocked_task_clears_its_dependency_too() {
1434 // The web UI's "Hold" and "Mark done" buttons are both reachable on a
1435 // `blocked` task, not just on `queued`/`held` ones - neither requires
1436 // a release first. A task moved off `Blocked` that way must not still
1437 // carry the dependency it was waiting on: a dependency graph built
1438 // from `blocked_by` would otherwise keep drawing an edge for a task
1439 // that is not blocked on anything any more.
1440 let mut held = task("held straight out of blocked");
1441 held.block(
1442 vec!["20260101-000000-dead".to_owned()],
1443 Some("waiting on the migration script".to_owned()),
1444 );
1445 assert_eq!(held.status, TaskStatus::Blocked);
1446
1447 held.hold_manual(None);
1448 assert_eq!(held.status, TaskStatus::Held);
1449 assert!(
1450 held.blocked_by.is_empty(),
1451 "hold overrides the wait, same as release"
1452 );
1453 assert!(held.block_reason.is_none());
1454
1455 let mut done = task("closed straight out of blocked");
1456 done.block(
1457 vec!["20260101-000000-dead".to_owned()],
1458 Some("waiting on the migration script".to_owned()),
1459 );
1460 done.succeed();
1461 assert_eq!(done.status, TaskStatus::Done);
1462 assert!(
1463 done.blocked_by.is_empty(),
1464 "a done task cannot still be waiting on a dependency"
1465 );
1466 assert!(done.block_reason.is_none());
1467 }
1468
1469 #[test]
1470 fn a_blocked_task_is_never_offered_to_the_loop() {
1471 let mut t = task("blocked");
1472 assert!(t.status.runnable());
1473 t.block(
1474 vec!["dep-id".to_owned()],
1475 Some("waits on dep-id".to_owned()),
1476 );
1477 assert_eq!(t.status, TaskStatus::Blocked);
1478 assert!(!t.status.runnable());
1479 assert_eq!(TaskStatus::Blocked.as_str(), "blocked");
1480 }
1481
1482 #[test]
1483 fn unblocking_the_last_dependency_returns_the_task_to_queued() {
1484 let mut t = task("blocked on two");
1485 t.block(
1486 vec!["a".to_owned(), "b".to_owned()],
1487 Some("waits on a and b".to_owned()),
1488 );
1489
1490 t.unblock("a");
1491 assert_eq!(t.status, TaskStatus::Blocked, "b is still outstanding");
1492 assert_eq!(t.blocked_by, ["b"]);
1493
1494 t.unblock("b");
1495 assert_eq!(t.status, TaskStatus::Queued);
1496 assert!(t.blocked_by.is_empty());
1497 assert!(t.block_reason.is_none());
1498 }
1499
1500 #[test]
1501 fn unblocking_an_id_on_a_task_that_is_not_blocked_is_a_no_op() {
1502 let mut t = task("never blocked");
1503 t.unblock("whatever");
1504 assert_eq!(t.status, TaskStatus::Queued);
1505 }
1506
1507 #[test]
1508 fn a_held_task_blocked_on_a_question_returns_to_held_not_queued() {
1509 // The bug this guards: a task an operator (or `crate::triage`) has
1510 // deliberately held, once `crate::conduct` blocks it on a follow-up
1511 // question, must not silently re-enter the competition queue the
1512 // moment that question is answered - whatever the answer said.
1513 let mut t = task("held, then asked about");
1514 t.hold_machine(Some("out of attempts".to_owned()));
1515 assert_eq!(t.status, TaskStatus::Held);
1516
1517 t.block(vec!["q1".to_owned()], Some("what now?".to_owned()));
1518 assert_eq!(t.status, TaskStatus::Blocked);
1519
1520 t.record_answer("what now?".to_owned(), "leave it held".to_owned());
1521 t.unblock("q1");
1522 assert_eq!(t.status, TaskStatus::Held, "must restore, not requeue");
1523 assert_eq!(t.hold_reason.as_deref(), Some("out of attempts"));
1524 assert_eq!(t.hold_source, Some(HoldSource::Machine));
1525 assert!(t.blocked_from.is_none(), "consumed once restored");
1526 }
1527
1528 #[test]
1529 fn a_manually_held_task_blocked_on_a_question_returns_to_held() {
1530 let mut t = task("manually held, then asked about");
1531 t.hold_manual(Some("waiting on a dependency".to_owned()));
1532
1533 t.block(vec!["q1".to_owned()], None);
1534 t.unblock("q1");
1535
1536 assert_eq!(t.status, TaskStatus::Held);
1537 assert_eq!(t.hold_source, Some(HoldSource::Manual));
1538 }
1539
1540 #[test]
1541 fn re_blocking_an_already_blocked_task_keeps_the_original_blocked_from() {
1542 // A second `Task::block` call - `crate::conduct` adding a question on
1543 // top of an existing block - must not overwrite `blocked_from` with
1544 // `Blocked` itself, or the task would restore into itself.
1545 let mut t = task("held, blocked twice");
1546 t.hold_machine(None);
1547 t.block(vec!["q1".to_owned()], Some("first".to_owned()));
1548 t.block(
1549 vec!["q1".to_owned(), "q2".to_owned()],
1550 Some("second".to_owned()),
1551 );
1552
1553 t.unblock("q1");
1554 assert_eq!(t.status, TaskStatus::Blocked, "q2 still outstanding");
1555 t.unblock("q2");
1556 assert_eq!(t.status, TaskStatus::Held);
1557 }
1558
1559 #[test]
1560 fn unblocking_a_task_blocked_while_running_lands_on_queued_not_running() {
1561 // Whatever process was driving the run is gone by the time a
1562 // conductor's question about it gets answered - there is nothing left
1563 // to resume into.
1564 let mut t = task("blocked mid-run");
1565 t.start("run-1".to_owned());
1566 assert_eq!(t.status, TaskStatus::Running);
1567
1568 t.block(vec!["q1".to_owned()], None);
1569 t.unblock("q1");
1570 assert_eq!(t.status, TaskStatus::Queued);
1571 }
1572
1573 #[test]
1574 fn a_pre_schema_4_blocked_record_with_hold_evidence_restores_to_held() {
1575 // `blocked_from` is `None` for a record written before schema 4 (or,
1576 // equivalently, deserialized straight from an on-disk file that never
1577 // had the field). Held evidence surviving on the task - never cleared
1578 // by `block` - is the only way left to tell such a record apart from
1579 // one blocked straight out of `Queued`.
1580 let mut t = task("legacy record, held before it was blocked");
1581 t.hold_source = Some(HoldSource::Machine);
1582 t.hold_reason = Some("legacy hold reason".to_owned());
1583 t.status = TaskStatus::Blocked;
1584 t.blocked_by = vec!["q1".to_owned()];
1585 t.blocked_from = None;
1586
1587 t.unblock("q1");
1588 assert_eq!(t.status, TaskStatus::Held);
1589 }
1590
1591 #[test]
1592 fn a_pre_schema_4_blocked_record_with_no_hold_evidence_restores_to_queued() {
1593 let mut t = task("legacy record, ordinary dependency block");
1594 t.status = TaskStatus::Blocked;
1595 t.blocked_by = vec!["dep".to_owned()];
1596 t.blocked_from = None;
1597
1598 t.unblock("dep");
1599 assert_eq!(t.status, TaskStatus::Queued);
1600 }
1601
1602 #[test]
1603 fn answering_a_question_is_recorded_and_survives_a_release() {
1604 let mut t = task("asked something");
1605 t.block(vec!["q1".to_owned()], Some("which backend?".to_owned()));
1606 t.record_answer("Which backend?".to_owned(), "SQLite".to_owned());
1607 t.unblock("q1");
1608 assert_eq!(t.status, TaskStatus::Queued);
1609 assert_eq!(t.answers.len(), 1);
1610 assert_eq!(t.answers[0].answer, "SQLite");
1611
1612 // A release resets attempts, not evidence - the same rule
1613 // `releasing_a_held_task_gives_it_a_real_second_chance` asserts for
1614 // `runs`.
1615 t.release();
1616 assert_eq!(t.answers.len(), 1, "the answer is not lost on release");
1617 }
1618
1619 #[test]
1620 fn requesting_review_requeues_the_task_and_remembers_the_branch() {
1621 let mut t = task("blocked run with a surviving branch");
1622 t.start("run-1".to_owned());
1623 t.fail("blocked with major findings", 5);
1624 assert_eq!(t.status, TaskStatus::Failed);
1625
1626 t.request_review("magi/eba2/A".to_owned());
1627 assert_eq!(t.status, TaskStatus::Queued);
1628 assert_eq!(t.attempts, 0);
1629 assert_eq!(t.review_branch.as_deref(), Some("magi/eba2/A"));
1630
1631 // An ordinary release (a human overriding the choice) drops it again.
1632 t.release();
1633 assert!(t.review_branch.is_none());
1634 }
1635
1636 #[test]
1637 fn conductor_requeue_but_not_an_ordinary_release_forces_a_fresh_start() {
1638 let mut t = task("retry");
1639 t.start("run-1".to_owned());
1640 t.requeue();
1641 assert!(t.fresh_start);
1642
1643 t.release();
1644 assert!(!t.fresh_start);
1645 }
1646
1647 #[test]
1648 fn priority_can_be_changed_while_queued_but_not_while_running() {
1649 let mut t = task("reprioritise me");
1650 t.set_priority(5).unwrap();
1651 assert_eq!(t.priority, 5);
1652
1653 t.start("run-1".to_owned());
1654 let err = t.set_priority(9).unwrap_err().to_string();
1655 assert!(err.contains("running"), "{err}");
1656 assert_eq!(t.priority, 5, "the rejected write must not partially apply");
1657 }
1658
1659 #[test]
1660 fn interrupt_can_be_marked_while_queued_but_not_while_running() {
1661 let mut t = task("interrupt me");
1662 assert!(!t.interrupt, "off unless asked, same as any other task");
1663
1664 t.set_interrupt(true).unwrap();
1665 assert!(t.interrupt);
1666
1667 t.start("run-1".to_owned());
1668 assert!(
1669 !t.interrupt,
1670 "the mark is one-shot: dispatching the task fulfils it, \
1671 whatever the run that follows ends up doing"
1672 );
1673 let err = t.set_interrupt(true).unwrap_err().to_string();
1674 assert!(err.contains("running"), "{err}");
1675 // Clearing is always allowed, even on a running task - there is
1676 // nothing left for it to interrupt once it has been claimed.
1677 t.set_interrupt(false).unwrap();
1678 assert!(!t.interrupt);
1679 }
1680
1681 /// R2-1-1: a task whose run fails and requeues must not go on
1682 /// re-triggering `crate::daemon`'s interrupt scheduler on every later
1683 /// boundary, attempt after attempt, until it exhausts its budget.
1684 #[test]
1685 fn a_failed_run_does_not_leave_the_task_still_marked_to_interrupt() {
1686 let mut t = task("interrupt me");
1687 t.set_interrupt(true).unwrap();
1688 t.start("run-1".to_owned());
1689 t.fail("mock failure", 5);
1690 assert_eq!(t.status, TaskStatus::Failed);
1691 assert!(
1692 !t.interrupt,
1693 "one attempt already spent the mark; a retry is an ordinary \
1694 requeue, not a fresh interrupt request"
1695 );
1696 }
1697
1698 #[test]
1699 fn changing_priority_moves_a_task_ahead_in_the_real_queue_order() {
1700 let (_dir, q) = queue();
1701 let mut a = task("first filed");
1702 let mut b = task("second filed");
1703 a.id = "20260101-000001-aaaa".to_owned();
1704 b.id = "20260101-000002-bbbb".to_owned();
1705 q.put(&mut a).unwrap();
1706 q.put(&mut b).unwrap();
1707
1708 assert_eq!(
1709 q.next_runnable().unwrap().id,
1710 a.id,
1711 "with equal priority the older task goes first, so a burst of \
1712 new work cannot starve it"
1713 );
1714 assert_eq!(
1715 q.list()[0].id,
1716 b.id,
1717 "but the list an operator reads is newest first, the same as \
1718 before priority existed - a's turn to run does not make it the \
1719 newest task"
1720 );
1721
1722 let mut a = q.get(&a.id).unwrap();
1723 a.set_priority(10).unwrap();
1724 q.put(&mut a).unwrap();
1725
1726 assert_eq!(
1727 q.next_runnable().unwrap().id,
1728 a.id,
1729 "a raised priority must be reflected the moment it is saved"
1730 );
1731 // `magi task list` and `GET /api/queue` both print `Queue::list()`
1732 // directly, so the raised task has to lead there too - not only in
1733 // what the loop would claim next.
1734 assert_eq!(
1735 q.list()[0].id,
1736 a.id,
1737 "the raised task must sort first in the list an operator reads, \
1738 not only in next_runnable's own ordering"
1739 );
1740 }
1741
1742 #[test]
1743 fn editing_replaces_title_and_instruction_but_keeps_identity_and_history() {
1744 let mut t = Task::new(
1745 "old title".to_owned(),
1746 "old instruction".to_owned(),
1747 PathBuf::from("/repo"),
1748 Source::Agent {
1749 run: "20260101-000000-beef".to_owned(),
1750 node: "implement".to_owned(),
1751 },
1752 );
1753 let id = t.id.clone();
1754 let created_at = t.created_at;
1755 t.runs.push("20260101-000000-beef".to_owned());
1756
1757 t.edit("new title".to_owned(), "new instruction".to_owned())
1758 .unwrap();
1759
1760 assert_eq!(t.title, "new title");
1761 assert_eq!(t.instruction, "new instruction");
1762 assert_eq!(t.id, id, "editing must not mint a new id");
1763 assert_eq!(t.created_at, created_at);
1764 assert_eq!(
1765 t.source,
1766 Source::Agent {
1767 run: "20260101-000000-beef".to_owned(),
1768 node: "implement".to_owned(),
1769 },
1770 "editing must not turn agent attribution into human"
1771 );
1772 assert_eq!(t.runs, ["20260101-000000-beef"]);
1773 }
1774
1775 #[test]
1776 fn editing_is_refused_once_a_task_is_running_or_finished() {
1777 let mut running = task("in flight");
1778 running.start("run-1".to_owned());
1779 let err = running
1780 .edit("x".to_owned(), "y".to_owned())
1781 .unwrap_err()
1782 .to_string();
1783 assert!(err.contains("running"), "{err}");
1784
1785 let mut done = task("finished");
1786 done.succeed();
1787 let err = done
1788 .edit("x".to_owned(), "y".to_owned())
1789 .unwrap_err()
1790 .to_string();
1791 assert!(err.contains("done"), "{err}");
1792
1793 // Both queued and held are the point of the feature and must work.
1794 let mut queued = task("waiting");
1795 queued.edit("x".to_owned(), "y".to_owned()).unwrap();
1796 let mut held = task("parked");
1797 held.hold_machine(None);
1798 held.edit("x".to_owned(), "y".to_owned()).unwrap();
1799 }
1800
1801 #[test]
1802 fn a_task_recorded_without_a_hold_reason_still_reads_as_none() {
1803 let (_dir, q) = queue();
1804 let path = q.path_of("20260101-000000-aaaa");
1805 std::fs::create_dir_all(q.root()).unwrap();
1806 std::fs::write(
1807 &path,
1808 serde_json::json!({
1809 "schema": SCHEMA,
1810 "id": "20260101-000000-aaaa",
1811 "title": "from before hold reasons existed",
1812 "instruction": "from before hold reasons existed",
1813 "repo": ".",
1814 "source": { "kind": "human" },
1815 "status": "held",
1816 "created_at": Timestamp::now().to_string(),
1817 "updated_at": Timestamp::now().to_string(),
1818 })
1819 .to_string(),
1820 )
1821 .unwrap();
1822
1823 let task = q.get("20260101-000000-aaaa").expect("must still read");
1824 assert!(task.hold_reason.is_none());
1825 assert!(task.operator_held());
1826 }
1827
1828 #[test]
1829 fn a_legacy_reasoned_hold_defaults_to_operator_protection() {
1830 let (_dir, q) = queue();
1831 let path = q.path_of("20260101-000000-bbbb");
1832 std::fs::create_dir_all(q.root()).unwrap();
1833 std::fs::write(
1834 &path,
1835 serde_json::json!({
1836 "schema": 2,
1837 "id": "20260101-000000-bbbb",
1838 "title": "old manual recovery",
1839 "instruction": "old manual recovery",
1840 "repo": ".",
1841 "source": { "kind": "human" },
1842 "status": "held",
1843 "hold_reason": "active manual recovery run20260912-224242-daf5",
1844 "created_at": Timestamp::now().to_string(),
1845 "updated_at": Timestamp::now().to_string(),
1846 })
1847 .to_string(),
1848 )
1849 .unwrap();
1850
1851 let task = q.get("20260101-000000-bbbb").expect("must still read");
1852 assert_eq!(task.hold_source, None);
1853 assert!(task.operator_held());
1854 }
1855
1856 #[test]
1857 fn a_task_recorded_without_a_diagnostic_still_reads_as_none() {
1858 let (_dir, q) = queue();
1859 let path = q.path_of("20260101-000000-aaaa");
1860 std::fs::create_dir_all(q.root()).unwrap();
1861 std::fs::write(
1862 &path,
1863 serde_json::json!({
1864 "schema": SCHEMA,
1865 "id": "20260101-000000-aaaa",
1866 "title": "from before diagnostics existed",
1867 "instruction": "from before diagnostics existed",
1868 "repo": ".",
1869 "source": { "kind": "human" },
1870 "status": "held",
1871 "created_at": Timestamp::now().to_string(),
1872 "updated_at": Timestamp::now().to_string(),
1873 })
1874 .to_string(),
1875 )
1876 .unwrap();
1877
1878 let task = q.get("20260101-000000-aaaa").expect("must still read");
1879 assert!(task.diagnostic.is_none());
1880 }
1881
1882 #[test]
1883 fn a_schema_1_task_with_no_blocking_fields_still_reads() {
1884 // Written by a build that predates `blocked_by`, `block_reason`,
1885 // `answers` and `review_branch` entirely - literal `"schema": 1`,
1886 // not `SCHEMA`, since the whole point is a build older than this one.
1887 let (_dir, q) = queue();
1888 let path = q.path_of("20260101-000000-aaaa");
1889 std::fs::create_dir_all(q.root()).unwrap();
1890 std::fs::write(
1891 &path,
1892 serde_json::json!({
1893 "schema": 1,
1894 "id": "20260101-000000-aaaa",
1895 "title": "from before blocking existed",
1896 "instruction": "from before blocking existed",
1897 "repo": ".",
1898 "source": { "kind": "human" },
1899 "status": "queued",
1900 "created_at": Timestamp::now().to_string(),
1901 "updated_at": Timestamp::now().to_string(),
1902 })
1903 .to_string(),
1904 )
1905 .unwrap();
1906
1907 let task = q.get("20260101-000000-aaaa").expect("must still read");
1908 assert!(task.blocked_by.is_empty());
1909 assert!(task.block_reason.is_none());
1910 assert!(task.answers.is_empty());
1911 assert!(task.review_branch.is_none());
1912 }
1913
1914 #[test]
1915 fn releasing_or_finishing_a_task_clears_its_stale_diagnostic() {
1916 // A diagnostic belongs to the run that produced it. Left in place
1917 // across a release, an unrelated later failure - a config error, say -
1918 // would go on showing evidence for a problem that is no longer why the
1919 // task is stuck.
1920 let mut held = task("diagnosed");
1921 held.start("run-1".to_owned());
1922 held.fail("gate red", 1);
1923 held.diagnostic = Some("cargo test failed: ...".to_owned());
1924 assert_eq!(held.status, TaskStatus::Held);
1925
1926 held.release();
1927 assert!(held.diagnostic.is_none());
1928
1929 held.diagnostic = Some("cargo test failed: ...".to_owned());
1930 held.succeed();
1931 assert!(held.diagnostic.is_none());
1932 }
1933
1934 #[test]
1935 fn failing_a_task_always_clears_whatever_diagnostic_it_carried() {
1936 let mut t = task("retried");
1937 t.start("run-1".to_owned());
1938 t.diagnostic = Some("stale evidence from a previous hold".to_owned());
1939 t.fail("unrelated config error", 5);
1940 assert_eq!(t.status, TaskStatus::Failed);
1941 assert!(
1942 t.diagnostic.is_none(),
1943 "fail() must not let an old diagnostic outlive the run that produced it"
1944 );
1945 }
1946
1947 #[test]
1948 fn a_claim_is_exclusive_and_releases_on_drop() {
1949 let (_dir, q) = queue();
1950 let mut t = task("contended");
1951 q.put(&mut t).unwrap();
1952
1953 let held = q.claim(&t.id).unwrap();
1954 assert!(
1955 q.claim(&t.id).is_err(),
1956 "two daemons must not drive one task into two runs"
1957 );
1958 drop(held);
1959 assert!(q.claim(&t.id).is_ok(), "a released claim is reclaimable");
1960 }
1961
1962 #[test]
1963 fn a_round_trip_survives_disk() {
1964 let (_dir, q) = queue();
1965 let mut t = Task::new(
1966 "titled".to_owned(),
1967 "body".to_owned(),
1968 PathBuf::from("/repo"),
1969 Source::Agent {
1970 run: "20260101-000000-beef".to_owned(),
1971 node: "implement".to_owned(),
1972 },
1973 );
1974 t.priority = 3;
1975 q.put(&mut t).unwrap();
1976
1977 let back = q.get(&t.id).unwrap();
1978 assert_eq!(back.id, t.id);
1979 assert_eq!(back.priority, 3);
1980 assert_eq!(back.source.label(), "implement@beef");
1981 // A prefix is enough, the way run ids work everywhere else.
1982 assert_eq!(q.get(t.short()).unwrap().id, t.id);
1983 }
1984
1985 #[test]
1986 fn an_unreadable_task_does_not_take_the_queue_down() {
1987 let (_dir, q) = queue();
1988 let mut t = task("fine");
1989 q.put(&mut t).unwrap();
1990 std::fs::write(q.root().join("broken.json"), "{ not json").unwrap();
1991
1992 let listed = q.list();
1993 assert_eq!(listed.len(), 1, "the readable task still lists");
1994 assert_eq!(listed[0].id, t.id);
1995 }
1996
1997 #[test]
1998 fn a_task_recorded_without_a_solo_field_still_reads_as_not_solo() {
1999 let (_dir, q) = queue();
2000 let path = q.path_of("20260101-000000-aaaa");
2001 std::fs::create_dir_all(q.root()).unwrap();
2002 std::fs::write(
2003 &path,
2004 serde_json::json!({
2005 "schema": SCHEMA,
2006 "id": "20260101-000000-aaaa",
2007 "title": "from before solo existed",
2008 "instruction": "from before solo existed",
2009 "repo": ".",
2010 "source": { "kind": "human" },
2011 "status": "queued",
2012 "created_at": Timestamp::now().to_string(),
2013 "updated_at": Timestamp::now().to_string(),
2014 })
2015 .to_string(),
2016 )
2017 .unwrap();
2018
2019 let task = q.get("20260101-000000-aaaa").expect("must still read");
2020 assert!(!task.solo, "a queue file with no `solo` field means false");
2021 }
2022
2023 #[test]
2024 fn a_task_recorded_without_an_urgent_field_still_reads_as_not_urgent() {
2025 let (_dir, q) = queue();
2026 let path = q.path_of("20260101-000000-bbbb");
2027 std::fs::create_dir_all(q.root()).unwrap();
2028 std::fs::write(
2029 &path,
2030 serde_json::json!({
2031 "schema": SCHEMA,
2032 "id": "20260101-000000-bbbb",
2033 "title": "from before urgent existed",
2034 "instruction": "from before urgent existed",
2035 "repo": ".",
2036 "source": { "kind": "human" },
2037 "status": "queued",
2038 "created_at": Timestamp::now().to_string(),
2039 "updated_at": Timestamp::now().to_string(),
2040 })
2041 .to_string(),
2042 )
2043 .unwrap();
2044
2045 let task = q.get("20260101-000000-bbbb").expect("must still read");
2046 assert!(
2047 !task.urgent,
2048 "a queue file with no `urgent` field means false, same as `solo`"
2049 );
2050 }
2051
2052 #[test]
2053 fn a_task_from_a_future_schema_is_refused_rather_than_guessed_at() {
2054 let (_dir, q) = queue();
2055 let mut t = task("from the future");
2056 q.put(&mut t).unwrap();
2057 let path = q.path_of(&t.id);
2058 let body = std::fs::read_to_string(&path)
2059 .unwrap()
2060 .replace(&format!("\"schema\": {SCHEMA}"), "\"schema\": 99");
2061 std::fs::write(&path, body).unwrap();
2062
2063 let err = q.get(&t.id).unwrap_err().to_string();
2064 assert!(err.contains("schema 99"), "{err}");
2065 }
2066
2067 #[test]
2068 fn revision_moves_when_the_queue_changes() {
2069 let (_dir, q) = queue();
2070 assert_eq!(q.revision(), 0, "an empty queue has no revision");
2071 let mut t = task("first");
2072 q.put(&mut t).unwrap();
2073 assert!(q.revision() > 0, "a written task moves the revision");
2074 }
2075
2076 #[test]
2077 fn revision_moves_when_deleting_an_older_task() {
2078 let (dir, q) = queue();
2079 let questions = Questions::at(dir.path().join("questions"));
2080 let mut t1 = task("older");
2081 q.put(&mut t1).unwrap();
2082 // Ensure mtime ticks forward.
2083 std::thread::sleep(std::time::Duration::from_millis(10));
2084 let mut t2 = task("newer");
2085 q.put(&mut t2).unwrap();
2086
2087 let rev_before = q.revision();
2088 q.remove(&t1.id, false, &questions).unwrap();
2089 let rev_after = q.revision();
2090
2091 assert_ne!(
2092 rev_before, rev_after,
2093 "deleting an older task must change the revision so other clients see the deletion"
2094 );
2095 }
2096
2097 #[test]
2098 fn removing_a_task_takes_it_out_of_the_listing() {
2099 let (dir, q) = queue();
2100 let questions = Questions::at(dir.path().join("questions"));
2101 let mut t = task("delete me");
2102 q.put(&mut t).unwrap();
2103 let removed = q.remove(t.short(), false, &questions).unwrap();
2104 assert_eq!(removed.id, t.id, "a prefix resolves before deleting");
2105 assert!(removed.quarantined.is_empty(), "nothing was blocked on it");
2106 assert!(q.list().is_empty());
2107 assert!(
2108 q.remove(&t.id, false, &questions).is_err(),
2109 "removing twice is an error"
2110 );
2111 }
2112
2113 #[test]
2114 fn removing_a_task_takes_its_stale_lock_with_it() {
2115 let (dir, q) = queue();
2116 let questions = Questions::at(dir.path().join("questions"));
2117 let mut t = task("interrupted");
2118 q.put(&mut t).unwrap();
2119
2120 // A daemon killed mid-run leaves this behind. Nothing holds it: the
2121 // process that would have dropped the guard is gone.
2122 let claim = q.claim(&t.id).unwrap();
2123 std::mem::forget(claim);
2124 assert!(
2125 q.claim(&t.id).is_err(),
2126 "the orphaned lock is what makes the task look claimed"
2127 );
2128
2129 // A live daemon on this task is refused, whatever the lock says.
2130 let err = q.remove(&t.id, true, &questions).unwrap_err().to_string();
2131 assert!(err.contains("live daemon"), "{err}");
2132 assert!(q.get(&t.id).is_ok(), "a refused delete keeps the task");
2133
2134 // With no daemon behind it, the lock is stale and goes with the task.
2135 q.remove(&t.id, false, &questions).unwrap();
2136 assert!(q.list().is_empty());
2137 let mut again = task("interrupted");
2138 again.id = t.id.clone();
2139 q.put(&mut again).unwrap();
2140 assert!(
2141 q.claim(&t.id).is_ok(),
2142 "a task that comes back must be claimable, which a left-behind lock would prevent"
2143 );
2144 }
2145
2146 #[test]
2147 fn removing_a_task_quarantines_what_was_blocked_on_it() {
2148 let (dir, q) = queue();
2149 let questions = Questions::at(dir.path().join("questions"));
2150
2151 let mut dep = task("dependency");
2152 q.put(&mut dep).unwrap();
2153
2154 let mut still_valid = task("still valid");
2155 q.put(&mut still_valid).unwrap();
2156
2157 let mut blocked = task("waiting");
2158 blocked.block(
2159 vec![dep.id.clone(), still_valid.id.clone()],
2160 Some("waits on both".to_owned()),
2161 );
2162 q.put(&mut blocked).unwrap();
2163
2164 let removed = q.remove(&dep.id, false, &questions).unwrap();
2165 assert_eq!(removed.quarantined, [blocked.id.clone()]);
2166
2167 let after = q.get(&blocked.id).unwrap();
2168 assert_eq!(after.status, TaskStatus::Held);
2169 assert_eq!(after.hold_source, Some(HoldSource::Machine));
2170 assert!(after.blocked_by.is_empty());
2171 let reason = after.hold_reason.as_deref().unwrap_or_default();
2172 assert!(reason.contains(&dep.id), "{reason}");
2173 assert!(
2174 reason.contains(&still_valid.id),
2175 "the still-valid dependency must survive in the reason text: {reason}"
2176 );
2177 }
2178}