a3s_flow/engine/
mod.rs

1use chrono::{DateTime, Utc};
2use std::sync::Arc;
3use uuid::Uuid;
4
5use crate::error::{FlowError, Result};
6use crate::model::{
7    project_run, ActiveHookSnapshot, FlowEvent, FlowEventEnvelope, HookStatus, RuntimeCommand,
8    StepStatus, WaitStatus, WorkflowRunSnapshot, WorkflowRunStatus, WorkflowRunSummary,
9    WorkflowRunSuspension, WorkflowSpec,
10};
11use crate::observe::{FlowEventObserver, NoopFlowEventObserver};
12use crate::runtime::{FlowRuntime, WorkflowInvocation};
13use crate::store::{FlowEventStore, InMemoryEventStore};
14
15mod operations;
16mod steps;
17mod validation;
18use steps::StepExecutionContext;
19use validation::{
20    ensure_child_operation_matches, ensure_hook_command_matches, ensure_progress_matches,
21    ensure_same_start, ensure_step_batch_valid, ensure_step_command_matches,
22    ensure_wait_command_matches, is_event_conflict, validate_run_id,
23};
24
25/// Builder for a [`FlowEngine`].
26pub struct FlowEngineBuilder {
27    store: Arc<dyn FlowEventStore>,
28    runtime: Arc<dyn FlowRuntime>,
29    observer: Arc<dyn FlowEventObserver>,
30    max_replay_iterations: usize,
31}
32
33impl FlowEngineBuilder {
34    pub fn new(runtime: Arc<dyn FlowRuntime>) -> Self {
35        Self {
36            store: Arc::new(InMemoryEventStore::new()),
37            runtime,
38            observer: Arc::new(NoopFlowEventObserver),
39            max_replay_iterations: 1024,
40        }
41    }
42
43    pub fn with_store(mut self, store: Arc<dyn FlowEventStore>) -> Self {
44        self.store = store;
45        self
46    }
47
48    pub fn with_observer(mut self, observer: Arc<dyn FlowEventObserver>) -> Self {
49        self.observer = observer;
50        self
51    }
52
53    pub fn with_max_replay_iterations(mut self, max_replay_iterations: usize) -> Self {
54        self.max_replay_iterations = max_replay_iterations.max(1);
55        self
56    }
57
58    pub fn build(self) -> FlowEngine {
59        FlowEngine {
60            store: self.store,
61            runtime: self.runtime,
62            observer: self.observer,
63            max_replay_iterations: self.max_replay_iterations,
64        }
65    }
66}
67
68/// Event-sourced workflow engine.
69#[derive(Clone)]
70pub struct FlowEngine {
71    store: Arc<dyn FlowEventStore>,
72    runtime: Arc<dyn FlowRuntime>,
73    observer: Arc<dyn FlowEventObserver>,
74    max_replay_iterations: usize,
75}
76
77impl FlowEngine {
78    pub fn builder(runtime: Arc<dyn FlowRuntime>) -> FlowEngineBuilder {
79        FlowEngineBuilder::new(runtime)
80    }
81
82    pub fn new(store: Arc<dyn FlowEventStore>, runtime: Arc<dyn FlowRuntime>) -> Self {
83        Self {
84            store,
85            runtime,
86            observer: Arc::new(NoopFlowEventObserver),
87            max_replay_iterations: 1024,
88        }
89    }
90
91    pub fn in_memory(runtime: Arc<dyn FlowRuntime>) -> Self {
92        Self::new(Arc::new(InMemoryEventStore::new()), runtime)
93    }
94
95    pub fn store(&self) -> Arc<dyn FlowEventStore> {
96        Arc::clone(&self.store)
97    }
98
99    pub fn observer(&self) -> Arc<dyn FlowEventObserver> {
100        Arc::clone(&self.observer)
101    }
102
103    /// Start a workflow run and drive it until completion or suspension.
104    pub async fn start(&self, spec: WorkflowSpec, input: serde_json::Value) -> Result<String> {
105        let run_id = Uuid::new_v4().to_string();
106        self.start_with_id(run_id, spec, input).await
107    }
108
109    /// Start a workflow run using a caller-provided durable run id.
110    ///
111    /// Reusing the same `run_id` with the same workflow spec and input is
112    /// idempotent. Reusing it with different spec or input returns a conflict.
113    pub async fn start_with_id(
114        &self,
115        run_id: impl Into<String>,
116        spec: WorkflowSpec,
117        input: serde_json::Value,
118    ) -> Result<String> {
119        spec.validate()?;
120        let run_id = run_id.into();
121        validate_run_id(&run_id)?;
122
123        for _ in 0..self.max_replay_iterations {
124            match self.store.list(&run_id).await {
125                Ok(history) => {
126                    let snapshot = project_run(&run_id, &history)?;
127                    ensure_same_start(&run_id, &snapshot, &spec, &input)?;
128                    if !history
129                        .iter()
130                        .any(|event| matches!(event.event, FlowEvent::RunStarted))
131                    {
132                        match self
133                            .record_event_at(&run_id, snapshot.last_sequence, FlowEvent::RunStarted)
134                            .await
135                        {
136                            Ok(_) => {}
137                            Err(err) if is_event_conflict(&err) => continue,
138                            Err(err) => return Err(err),
139                        }
140                    }
141                    match self.drive(&run_id).await {
142                        Ok(_) => return Ok(run_id),
143                        Err(err) if is_event_conflict(&err) => continue,
144                        Err(err) => return Err(err),
145                    }
146                }
147                Err(FlowError::RunNotFound(_)) => {
148                    let created = match self
149                        .record_event_at(
150                            &run_id,
151                            0,
152                            FlowEvent::RunCreated {
153                                spec: spec.clone(),
154                                input: input.clone(),
155                            },
156                        )
157                        .await
158                    {
159                        Ok(created) => created,
160                        Err(err) if is_event_conflict(&err) => continue,
161                        Err(err) => return Err(err),
162                    };
163                    match self
164                        .record_event_at(&run_id, created.sequence, FlowEvent::RunStarted)
165                        .await
166                    {
167                        Ok(_) => {}
168                        Err(err) if is_event_conflict(&err) => continue,
169                        Err(err) => return Err(err),
170                    }
171                    match self.drive(&run_id).await {
172                        Ok(_) => return Ok(run_id),
173                        Err(err) if is_event_conflict(&err) => continue,
174                        Err(err) => return Err(err),
175                    }
176                }
177                Err(err) => return Err(err),
178            }
179        }
180
181        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
182    }
183
184    /// Resume a wait once its timer has fired.
185    pub async fn resume_wait(&self, run_id: &str, wait_id: &str) -> Result<()> {
186        for _ in 0..self.max_replay_iterations {
187            let snapshot = self.snapshot(run_id).await?;
188            if snapshot.status.is_terminal() {
189                return Err(FlowError::RunTerminal(run_id.to_string()));
190            }
191            match snapshot.waits.get(wait_id) {
192                Some(wait) if wait.status == WaitStatus::Waiting => {
193                    match self
194                        .record_event_at(
195                            run_id,
196                            snapshot.last_sequence,
197                            FlowEvent::WaitCompleted {
198                                wait_id: wait_id.to_string(),
199                            },
200                        )
201                        .await
202                    {
203                        Ok(_) => {}
204                        Err(err) if is_event_conflict(&err) => continue,
205                        Err(err) => return Err(err),
206                    }
207                    match self.drive(run_id).await {
208                        Ok(_) => return Ok(()),
209                        Err(err) if is_event_conflict(&err) => continue,
210                        Err(err) => return Err(err),
211                    }
212                }
213                Some(_) => match self.drive(run_id).await {
214                    Ok(_) => return Ok(()),
215                    Err(err) if is_event_conflict(&err) => continue,
216                    Err(err) => return Err(err),
217                },
218                None => {
219                    return Err(FlowError::InvalidTransition(format!(
220                        "wait {wait_id} does not exist for run {run_id}"
221                    )))
222                }
223            }
224        }
225
226        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
227    }
228
229    /// Resume an active hook with external payload.
230    pub async fn resume_hook(
231        &self,
232        run_id: &str,
233        hook_id: &str,
234        payload: serde_json::Value,
235    ) -> Result<()> {
236        for _ in 0..self.max_replay_iterations {
237            let snapshot = self.snapshot(run_id).await?;
238            if snapshot.status.is_terminal() {
239                return Err(FlowError::RunTerminal(run_id.to_string()));
240            }
241            match snapshot.hooks.get(hook_id) {
242                Some(hook) if hook.status == HookStatus::Active => {
243                    match self
244                        .record_event_at(
245                            run_id,
246                            snapshot.last_sequence,
247                            FlowEvent::HookReceived {
248                                hook_id: hook_id.to_string(),
249                                payload: payload.clone(),
250                            },
251                        )
252                        .await
253                    {
254                        Ok(_) => {}
255                        Err(err) if is_event_conflict(&err) => continue,
256                        Err(err) => return Err(err),
257                    }
258                    match self.drive(run_id).await {
259                        Ok(_) => return Ok(()),
260                        Err(err) if is_event_conflict(&err) => continue,
261                        Err(err) => return Err(err),
262                    }
263                }
264                Some(_) => match self.drive(run_id).await {
265                    Ok(_) => return Ok(()),
266                    Err(err) if is_event_conflict(&err) => continue,
267                    Err(err) => return Err(err),
268                },
269                None => {
270                    return Err(FlowError::InvalidTransition(format!(
271                        "hook {hook_id} does not exist for run {run_id}"
272                    )))
273                }
274            }
275        }
276
277        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
278    }
279
280    /// Dispose an active hook without accepting a callback payload.
281    ///
282    /// This is useful when a host withdraws an approval request, expires a
283    /// webhook token, or closes an external callback route. The workflow is
284    /// driven after the disposal event so replay code can observe
285    /// [`WorkflowContext::hook_disposed`](crate::context::WorkflowContext::hook_disposed)
286    /// and complete, fail, or schedule an alternate path.
287    pub async fn dispose_hook(&self, run_id: &str, hook_id: &str) -> Result<()> {
288        for _ in 0..self.max_replay_iterations {
289            let snapshot = self.snapshot(run_id).await?;
290            if snapshot.status.is_terminal() {
291                return Err(FlowError::RunTerminal(run_id.to_string()));
292            }
293            match snapshot.hooks.get(hook_id) {
294                Some(hook) if hook.status == HookStatus::Active => {
295                    match self
296                        .record_event_at(
297                            run_id,
298                            snapshot.last_sequence,
299                            FlowEvent::HookDisposed {
300                                hook_id: hook_id.to_string(),
301                            },
302                        )
303                        .await
304                    {
305                        Ok(_) => {}
306                        Err(err) if is_event_conflict(&err) => continue,
307                        Err(err) => return Err(err),
308                    }
309                    match self.drive(run_id).await {
310                        Ok(_) => return Ok(()),
311                        Err(err) if is_event_conflict(&err) => continue,
312                        Err(err) => return Err(err),
313                    }
314                }
315                Some(_) => match self.drive(run_id).await {
316                    Ok(_) => return Ok(()),
317                    Err(err) if is_event_conflict(&err) => continue,
318                    Err(err) => return Err(err),
319                },
320                None => {
321                    return Err(FlowError::InvalidTransition(format!(
322                        "hook {hook_id} does not exist for run {run_id}"
323                    )))
324                }
325            }
326        }
327
328        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
329    }
330
331    /// Resume an active hook by its external token.
332    ///
333    /// This is the API webhook handlers normally want: the callback receives a
334    /// token, while `run_id` and `hook_id` remain engine internals.
335    pub async fn resume_hook_by_token(
336        &self,
337        token: &str,
338        payload: serde_json::Value,
339    ) -> Result<(String, String)> {
340        let mut matches = self
341            .store
342            .find_active_hooks_by_token(token)
343            .await?
344            .into_iter()
345            .map(|active| (active.run_id, active.hook.hook_id))
346            .collect::<Vec<_>>();
347
348        match matches.len() {
349            0 => Err(FlowError::HookTokenNotFound(token.to_string())),
350            1 => {
351                let (run_id, hook_id) = matches.remove(0);
352                self.resume_hook(&run_id, &hook_id, payload).await?;
353                Ok((run_id, hook_id))
354            }
355            _ => Err(FlowError::InvalidTransition(
356                "hook token is active in multiple runs (value redacted)".to_string(),
357            )),
358        }
359    }
360
361    /// Dispose an active hook by its external token.
362    ///
363    /// This mirrors [`resume_hook_by_token`](Self::resume_hook_by_token) for
364    /// callback routers that only know the public token.
365    pub async fn dispose_hook_by_token(&self, token: &str) -> Result<(String, String)> {
366        let mut matches = self
367            .store
368            .find_active_hooks_by_token(token)
369            .await?
370            .into_iter()
371            .map(|active| (active.run_id, active.hook.hook_id))
372            .collect::<Vec<_>>();
373
374        match matches.len() {
375            0 => Err(FlowError::HookTokenNotFound(token.to_string())),
376            1 => {
377                let (run_id, hook_id) = matches.remove(0);
378                self.dispose_hook(&run_id, &hook_id).await?;
379                Ok((run_id, hook_id))
380            }
381            _ => Err(FlowError::InvalidTransition(
382                "hook token is active in multiple runs (value redacted)".to_string(),
383            )),
384        }
385    }
386
387    /// List active waits whose `resume_at` is at or before `now`.
388    ///
389    /// Scheduler integrations can use this to inspect due timers before
390    /// deciding how aggressively to drive them.
391    pub async fn list_due_waits(&self, now: DateTime<Utc>) -> Result<Vec<(String, String)>> {
392        let mut due = Vec::new();
393        for run_id in self.store.list_run_ids().await? {
394            let snapshot = self.snapshot(&run_id).await?;
395            if snapshot.status.is_terminal() {
396                continue;
397            }
398            for wait in snapshot.waits.values() {
399                if wait.status == WaitStatus::Waiting && wait.resume_at <= now {
400                    due.push((run_id.clone(), wait.wait_id.clone()));
401                }
402            }
403        }
404        due.sort();
405        Ok(due)
406    }
407
408    /// Complete every due wait and drive the affected workflows.
409    ///
410    /// Returns the `(run_id, wait_id)` pairs that were resumed. A wait already
411    /// completed by another caller is skipped by [`Self::resume_wait`].
412    pub async fn resume_due_waits(&self, now: DateTime<Utc>) -> Result<Vec<(String, String)>> {
413        let due = self.list_due_waits(now).await?;
414        let mut resumed = Vec::with_capacity(due.len());
415        for (run_id, wait_id) in due {
416            self.resume_wait(&run_id, &wait_id).await?;
417            resumed.push((run_id, wait_id));
418        }
419        Ok(resumed)
420    }
421
422    /// List pending step retries whose `retry_after` is at or before `now`.
423    pub async fn list_due_retries(&self, now: DateTime<Utc>) -> Result<Vec<(String, String)>> {
424        let mut due = Vec::new();
425        for run_id in self.store.list_run_ids().await? {
426            let snapshot = self.snapshot(&run_id).await?;
427            if snapshot.status.is_terminal() {
428                continue;
429            }
430            for (step_id, _) in snapshot.due_retries(now) {
431                due.push((run_id.clone(), step_id));
432            }
433        }
434        due.sort();
435        Ok(due)
436    }
437
438    /// Drive every run with a due step retry.
439    pub async fn resume_due_retries(&self, now: DateTime<Utc>) -> Result<Vec<(String, String)>> {
440        let due = self.list_due_retries(now).await?;
441        let mut run_ids = Vec::new();
442        for (run_id, _) in &due {
443            if !run_ids.contains(run_id) {
444                run_ids.push(run_id.clone());
445            }
446        }
447        for run_id in run_ids {
448            self.drive_at(&run_id, now).await?;
449        }
450        Ok(due)
451    }
452
453    pub async fn snapshot(&self, run_id: &str) -> Result<WorkflowRunSnapshot> {
454        let history = self.store.list(run_id).await?;
455        project_run(run_id, &history)
456    }
457
458    pub async fn history(&self, run_id: &str) -> Result<Vec<FlowEventEnvelope>> {
459        self.store.list(run_id).await
460    }
461
462    pub async fn list_run_ids(&self) -> Result<Vec<String>> {
463        self.store.list_run_ids().await
464    }
465
466    pub async fn list_snapshots(&self) -> Result<Vec<WorkflowRunSnapshot>> {
467        let mut snapshots = Vec::new();
468        for run_id in self.store.list_run_ids().await? {
469            snapshots.push(self.snapshot(&run_id).await?);
470        }
471        Ok(snapshots)
472    }
473
474    /// Summarize run state across the active store.
475    ///
476    /// Suspension counters include only non-terminal runs, so a cancelled run
477    /// that still has an old wait or hook in history is not reported as
478    /// actionable work.
479    pub async fn run_summary(&self) -> Result<WorkflowRunSummary> {
480        let snapshots = self.list_snapshots().await?;
481        Ok(WorkflowRunSummary::from_snapshots(&snapshots))
482    }
483
484    /// List open waits, active hooks, and pending delayed retries.
485    ///
486    /// The `due` flag on wait and retry suspensions is computed against `now`.
487    /// Terminal runs are skipped so cancelled histories do not produce
488    /// actionable operator work.
489    pub async fn list_open_suspensions(
490        &self,
491        now: DateTime<Utc>,
492    ) -> Result<Vec<WorkflowRunSuspension>> {
493        let mut suspensions = Vec::new();
494        for run_id in self.store.list_run_ids().await? {
495            let snapshot = self.snapshot(&run_id).await?;
496            if snapshot.status.is_terminal() {
497                continue;
498            }
499            for wait in snapshot.waits.values() {
500                if wait.status == WaitStatus::Waiting {
501                    suspensions.push(WorkflowRunSuspension::Wait {
502                        run_id: run_id.clone(),
503                        wait: wait.clone(),
504                        due: wait.resume_at <= now,
505                    });
506                }
507            }
508            for hook in snapshot.hooks.values() {
509                if hook.status == HookStatus::Active {
510                    suspensions.push(WorkflowRunSuspension::Hook {
511                        run_id: run_id.clone(),
512                        hook: hook.clone(),
513                    });
514                }
515            }
516            for step in snapshot.steps.values() {
517                if step.status == StepStatus::Pending {
518                    if let Some(retry_after) = step.retry_after {
519                        suspensions.push(WorkflowRunSuspension::Retry {
520                            run_id: run_id.clone(),
521                            step: step.clone(),
522                            due: retry_after <= now,
523                        });
524                    }
525                }
526            }
527        }
528        suspensions.sort_by(|left, right| {
529            (left.run_id(), left.kind_order(), left.subject_id()).cmp(&(
530                right.run_id(),
531                right.kind_order(),
532                right.subject_id(),
533            ))
534        });
535        Ok(suspensions)
536    }
537
538    /// Return the earliest open wait or delayed retry across non-terminal runs.
539    ///
540    /// This is useful for hosts that want to sleep until the next scheduler tick
541    /// instead of polling at a fixed interval. Active hooks are intentionally
542    /// ignored because they do not have a scheduled wake-up time.
543    pub async fn next_wakeup(&self, now: DateTime<Utc>) -> Result<Option<WorkflowRunSuspension>> {
544        let mut wakeups = self.list_open_suspensions(now).await?;
545        wakeups.retain(|suspension| suspension.scheduled_at().is_some());
546        wakeups.sort_by(|left, right| {
547            (
548                left.scheduled_at(),
549                left.run_id(),
550                left.kind_order(),
551                left.subject_id(),
552            )
553                .cmp(&(
554                    right.scheduled_at(),
555                    right.run_id(),
556                    right.kind_order(),
557                    right.subject_id(),
558                ))
559        });
560        Ok(wakeups.into_iter().next())
561    }
562
563    /// List active external callback hooks across non-terminal runs.
564    ///
565    /// Callback routers and dashboards can use this to discover public hook
566    /// tokens and their audit metadata without projecting every run manually.
567    /// The result is sorted by run ID and hook ID for stable polling output.
568    pub async fn list_active_hooks(&self) -> Result<Vec<ActiveHookSnapshot>> {
569        self.store.list_active_hooks().await
570    }
571
572    /// Replay and dispatch until the run reaches a terminal state or an open
573    /// wait/hook suspension.
574    pub async fn drive(&self, run_id: &str) -> Result<WorkflowRunSnapshot> {
575        self.drive_at(run_id, Utc::now()).await
576    }
577
578    async fn drive_at(&self, run_id: &str, now: DateTime<Utc>) -> Result<WorkflowRunSnapshot> {
579        'replay: for _ in 0..self.max_replay_iterations {
580            let history = self.store.list(run_id).await?;
581            let snapshot = project_run(run_id, &history)?;
582            if snapshot.status.is_terminal()
583                || snapshot
584                    .waits
585                    .values()
586                    .any(|wait| wait.status == WaitStatus::Waiting)
587                || snapshot
588                    .hooks
589                    .values()
590                    .any(|hook| hook.status == HookStatus::Active)
591                || (snapshot.has_future_retry(now) && snapshot.due_retries(now).is_empty())
592            {
593                return Ok(snapshot);
594            }
595
596            let command = self
597                .runtime
598                .run_workflow(WorkflowInvocation {
599                    run_id: run_id.to_string(),
600                    spec: snapshot.spec.clone(),
601                    input: snapshot.input.clone(),
602                    history,
603                })
604                .await?;
605
606            match command {
607                RuntimeCommand::Complete { output } => {
608                    if snapshot.status == WorkflowRunStatus::Cancelling {
609                        return Err(FlowError::InvalidTransition(format!(
610                            "workflow run {run_id} completed after cancellation was requested; cleanup-aware cancellation must return cancel or fail"
611                        )));
612                    }
613                    match self
614                        .record_event_at(
615                            run_id,
616                            snapshot.last_sequence,
617                            FlowEvent::RunCompleted { output },
618                        )
619                        .await
620                    {
621                        Ok(_) => {}
622                        Err(err) if is_event_conflict(&err) => continue,
623                        Err(err) => return Err(err),
624                    }
625                    return self.snapshot(run_id).await;
626                }
627                RuntimeCommand::Fail { error } => {
628                    match self
629                        .record_event_at(
630                            run_id,
631                            snapshot.last_sequence,
632                            FlowEvent::RunFailed { error },
633                        )
634                        .await
635                    {
636                        Ok(_) => {}
637                        Err(err) if is_event_conflict(&err) => continue,
638                        Err(err) => return Err(err),
639                    }
640                    return self.snapshot(run_id).await;
641                }
642                RuntimeCommand::Cancel => {
643                    let cancellation = snapshot.cancellation.as_ref().ok_or_else(|| {
644                        FlowError::InvalidTransition(format!(
645                            "workflow run {run_id} returned cancel without a durable cancellation request"
646                        ))
647                    })?;
648                    match self
649                        .record_event_at(
650                            run_id,
651                            snapshot.last_sequence,
652                            FlowEvent::RunCancelled {
653                                reason: cancellation.request.reason.clone(),
654                            },
655                        )
656                        .await
657                    {
658                        Ok(_) => {}
659                        Err(err) if is_event_conflict(&err) => continue,
660                        Err(err) => return Err(err),
661                    }
662                    return self.snapshot(run_id).await;
663                }
664                RuntimeCommand::Timeout { deadline, reason } => {
665                    match self
666                        .record_event_at(
667                            run_id,
668                            snapshot.last_sequence,
669                            FlowEvent::RunTimedOut { deadline, reason },
670                        )
671                        .await
672                    {
673                        Ok(_) => {}
674                        Err(err) if is_event_conflict(&err) => continue,
675                        Err(err) => return Err(err),
676                    }
677                    return self.snapshot(run_id).await;
678                }
679                RuntimeCommand::RecordProgress { progress } => {
680                    progress.validate()?;
681                    if let Some(existing) = snapshot.progress(&progress.progress_id) {
682                        ensure_progress_matches(run_id, existing, &progress)?;
683                        return Err(FlowError::InvalidTransition(format!(
684                            "workflow rescheduled progress {} without progress",
685                            progress.progress_id
686                        )));
687                    }
688                    match self
689                        .record_event_at(
690                            run_id,
691                            snapshot.last_sequence,
692                            FlowEvent::RunProgressRecorded { progress },
693                        )
694                        .await
695                    {
696                        Ok(_) => {}
697                        Err(err) if is_event_conflict(&err) => continue,
698                        Err(err) => return Err(err),
699                    }
700                }
701                RuntimeCommand::LinkChildOperation { child } => {
702                    child.validate()?;
703                    if let Some(existing) = snapshot.child_operation(&child.reference_id) {
704                        ensure_child_operation_matches(run_id, existing, &child)?;
705                        return Err(FlowError::InvalidTransition(format!(
706                            "workflow rescheduled child operation {} without progress",
707                            child.reference_id
708                        )));
709                    }
710                    match self
711                        .record_event_at(
712                            run_id,
713                            snapshot.last_sequence,
714                            FlowEvent::ChildOperationLinked { child },
715                        )
716                        .await
717                    {
718                        Ok(_) => {}
719                        Err(err) if is_event_conflict(&err) => continue,
720                        Err(err) => return Err(err),
721                    }
722                }
723                RuntimeCommand::ScheduleStep {
724                    step_id,
725                    step_name,
726                    input,
727                    retry,
728                } => {
729                    if let Some(step) = snapshot.steps.get(&step_id) {
730                        ensure_step_command_matches(run_id, step, &step_name, &input, retry)?;
731                        if matches!(
732                            step.status,
733                            StepStatus::Completed | StepStatus::Failed | StepStatus::Cancelled
734                        ) {
735                            return Err(FlowError::InvalidTransition(format!(
736                                "workflow rescheduled terminal step {step_id} without progress"
737                            )));
738                        }
739                    }
740                    match self
741                        .execute_step(
742                            run_id,
743                            &snapshot,
744                            StepExecutionContext {
745                                step_id,
746                                step_name,
747                                input,
748                                retry,
749                                now,
750                            },
751                        )
752                        .await
753                    {
754                        Ok(()) => {}
755                        Err(err) if is_event_conflict(&err) => continue,
756                        Err(err) => return Err(err),
757                    }
758                }
759                RuntimeCommand::ScheduleSteps { steps } => {
760                    ensure_step_batch_valid(&steps)?;
761                    for step in &steps {
762                        if let Some(existing) = snapshot.steps.get(&step.step_id) {
763                            ensure_step_command_matches(
764                                run_id,
765                                existing,
766                                &step.step_name,
767                                &step.input,
768                                step.retry,
769                            )?;
770                        }
771                    }
772                    if steps.iter().all(|step| {
773                        snapshot.steps.get(&step.step_id).is_some_and(|existing| {
774                            matches!(
775                                existing.status,
776                                StepStatus::Completed | StepStatus::Failed | StepStatus::Cancelled
777                            )
778                        })
779                    }) {
780                        let step_ids = steps
781                            .iter()
782                            .map(|step| step.step_id.as_str())
783                            .collect::<Vec<_>>()
784                            .join(", ");
785                        return Err(FlowError::InvalidTransition(format!(
786                            "workflow rescheduled only terminal steps without progress: {step_ids}"
787                        )));
788                    }
789                    match self.execute_step_batch(run_id, &snapshot, steps, now).await {
790                        Ok(()) => {}
791                        Err(err) if is_event_conflict(&err) => continue 'replay,
792                        Err(err) => return Err(err),
793                    }
794                }
795                RuntimeCommand::WaitUntil { wait_id, resume_at } => {
796                    match snapshot.waits.get(&wait_id) {
797                        Some(wait) => {
798                            ensure_wait_command_matches(run_id, wait, resume_at)?;
799                            match wait.status {
800                                WaitStatus::Completed => continue,
801                                WaitStatus::Waiting => return self.snapshot(run_id).await,
802                                WaitStatus::Cancelled => {
803                                    return Err(FlowError::InvalidTransition(format!(
804                                        "workflow rescheduled cancelled wait {wait_id}; cancellation cleanup must use a distinct stable identity"
805                                    )))
806                                }
807                            }
808                        }
809                        None => {
810                            match self
811                                .record_event_at(
812                                    run_id,
813                                    snapshot.last_sequence,
814                                    FlowEvent::WaitCreated { wait_id, resume_at },
815                                )
816                                .await
817                            {
818                                Ok(_) => {}
819                                Err(err) if is_event_conflict(&err) => continue,
820                                Err(err) => return Err(err),
821                            }
822                            return self.snapshot(run_id).await;
823                        }
824                    }
825                }
826                RuntimeCommand::CreateHook {
827                    hook_id,
828                    token,
829                    metadata,
830                } => match snapshot.hooks.get(&hook_id) {
831                    Some(hook) => {
832                        ensure_hook_command_matches(run_id, hook, &token, &metadata)?;
833                        match hook.status {
834                            HookStatus::Received | HookStatus::Disposed => continue,
835                            HookStatus::Active => return self.snapshot(run_id).await,
836                            HookStatus::Cancelled => {
837                                return Err(FlowError::InvalidTransition(format!(
838                                    "workflow rescheduled cancelled hook {hook_id}; cancellation cleanup must use a distinct stable identity"
839                                )))
840                            }
841                        }
842                    }
843                    None => {
844                        self.ensure_hook_token_available(run_id, &hook_id, &token)
845                            .await?;
846                        match self
847                            .record_event_at(
848                                run_id,
849                                snapshot.last_sequence,
850                                FlowEvent::HookCreated {
851                                    hook_id,
852                                    token,
853                                    metadata,
854                                },
855                            )
856                            .await
857                        {
858                            Ok(_) => {}
859                            Err(err) if is_event_conflict(&err) => continue,
860                            Err(err) => return Err(err),
861                        }
862                        return self.snapshot(run_id).await;
863                    }
864                },
865            }
866        }
867
868        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
869    }
870
871    async fn terminate_run(&self, run_id: &str, event: FlowEvent) -> Result<()> {
872        for _ in 0..self.max_replay_iterations {
873            let snapshot = self.snapshot(run_id).await?;
874            if snapshot.status.is_terminal() {
875                return Ok(());
876            }
877            match self
878                .record_event_at(run_id, snapshot.last_sequence, event.clone())
879                .await
880            {
881                Ok(_) => return Ok(()),
882                Err(err) if is_event_conflict(&err) => continue,
883                Err(err) => return Err(err),
884            }
885        }
886        Err(FlowError::ReplayLimitExceeded(self.max_replay_iterations))
887    }
888
889    async fn record_event_at(
890        &self,
891        run_id: &str,
892        expected_sequence: u64,
893        event: FlowEvent,
894    ) -> Result<FlowEventEnvelope> {
895        let envelope = self
896            .store
897            .append_if_sequence(run_id, expected_sequence, event)
898            .await?;
899        self.observer.observe(envelope.clone()).await;
900        Ok(envelope)
901    }
902
903    async fn ensure_hook_token_available(
904        &self,
905        run_id: &str,
906        hook_id: &str,
907        token: &str,
908    ) -> Result<()> {
909        for active in self.store.find_active_hooks_by_token(token).await? {
910            if active.run_id == run_id && active.hook.hook_id == hook_id {
911                continue;
912            }
913            return Err(FlowError::HookTokenConflict {
914                token: token.to_string(),
915                existing_run_id: active.run_id,
916                existing_hook_id: active.hook.hook_id,
917            });
918        }
919        Ok(())
920    }
921}
a3s_flow/engine/mod.rs

a3s_flow/engine/
mod.rs
