@dudousxd/nestjs-durable-core 0.76.0 → 0.77.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.cjs CHANGED
@@ -1250,6 +1250,20 @@ function allGroup(firstSeq, mode) {
1250
1250
  return mode === "failFast" ? `all:${firstSeq}:failFast` : `all:${firstSeq}`;
1251
1251
  }
1252
1252
  __name(allGroup, "allGroup");
1253
+ function stepFanGroup(firstSeq, size, mode) {
1254
+ const group = `steps:${firstSeq}:${size}`;
1255
+ return mode === "failFast" ? `${group}:failFast` : group;
1256
+ }
1257
+ __name(stepFanGroup, "stepFanGroup");
1258
+ function parseStepFanGroup(group) {
1259
+ const m = group ? /^steps:(\d+):(\d+)(?::failFast)?$/.exec(group) : null;
1260
+ if (!m) return void 0;
1261
+ return {
1262
+ firstSeq: Number(m[1]),
1263
+ size: Number(m[2])
1264
+ };
1265
+ }
1266
+ __name(parseStepFanGroup, "parseStepFanGroup");
1253
1267
  function allWait(runId, firstSeq, wave) {
1254
1268
  const base = allBase(runId, firstSeq, wave);
1255
1269
  return {
@@ -3291,6 +3305,9 @@ function createWorkflowCtx(host, runId, compensations, workflow = "", searchAttr
3291
3305
  }, "child");
3292
3306
  const all = /* @__PURE__ */ __name(async (workflow2, inputs, opts) => {
3293
3307
  if (inputs.length === 0) return [];
3308
+ if (stepNameOf(workflow2) !== void 0) {
3309
+ return allSteps(workflow2, inputs, opts);
3310
+ }
3294
3311
  const mode = opts?.mode ?? "waitAll";
3295
3312
  const name = workflowName(workflow2);
3296
3313
  const waves = allWaves(inputs.length, opts?.concurrency);
@@ -3430,6 +3447,26 @@ function createWorkflowCtx(host, runId, compensations, workflow = "", searchAttr
3430
3447
  if (failures.length > 0) throw new GatherError(failures, inputs.length);
3431
3448
  return outputs;
3432
3449
  }, "all");
3450
+ const allSteps = /* @__PURE__ */ __name(async (ref, inputs, opts) => {
3451
+ if (opts?.queue !== void 0) {
3452
+ throw new Error("ctx.all: a flow-control queue is not supported on a step fan");
3453
+ }
3454
+ const name = stepNameOf(ref);
3455
+ const def = {
3456
+ name,
3457
+ ...resolveStepPolicy(stepConfigOf(ref), opts)
3458
+ };
3459
+ const positions = inputs.map(() => pos.next());
3460
+ return host.callRemoteAll(runId, positions, def, inputs, {
3461
+ mode: opts?.mode ?? "waitAll",
3462
+ concurrency: opts?.concurrency,
3463
+ transport: opts?.transport,
3464
+ admission: {
3465
+ priority: opts?.priority,
3466
+ fairnessKey: opts?.fairnessKey
3467
+ }
3468
+ });
3469
+ }, "allSteps");
3433
3470
  const readFan = /* @__PURE__ */ __name(async (positions, prefixes) => {
3434
3471
  const checkpoints = positions.map((seq) => replay?.get(seq) ?? null);
3435
3472
  const folded = positions.map(() => false);
@@ -3766,6 +3803,18 @@ var BREAKPOINT = "breakpoint";
3766
3803
  var isBreakpoint = /* @__PURE__ */ __name((cp) => cp.status === "pending" && cp.name.startsWith(BREAKPOINT), "isBreakpoint");
3767
3804
  var VERSION_UNDECLARED_TAG = "version:undeclared";
3768
3805
  var versionKey = /* @__PURE__ */ __name((name, version) => `${name}@${version}`, "versionKey");
3806
+ var STEP_FAN_CHUNK = 200;
3807
+ function shuffle(items) {
3808
+ for (let i = items.length - 1; i > 0; i -= 1) {
3809
+ const j = Math.floor(Math.random() * (i + 1));
3810
+ [items[i], items[j]] = [
3811
+ items[j],
3812
+ items[i]
3813
+ ];
3814
+ }
3815
+ return items;
3816
+ }
3817
+ __name(shuffle, "shuffle");
3769
3818
  function stepGroup(step, runNamespace) {
3770
3819
  return tenantGroup(sanitizeQueueToken(step.name), step.partition ?? runNamespace);
3771
3820
  }
@@ -3807,6 +3856,7 @@ var WorkflowEngine = class {
3807
3856
  maxRecoveryAttempts;
3808
3857
  remoteAdvanceSilenceMs;
3809
3858
  reconcileMs;
3859
+ timerBatchSize;
3810
3860
  eventBufferTtlMs;
3811
3861
  remoteRedispatchMs;
3812
3862
  remoteRedispatchMax;
@@ -3867,6 +3917,21 @@ var WorkflowEngine = class {
3867
3917
  * {@link redriveSettledSteps}).
3868
3918
  */
3869
3919
  awaitedRemoteSteps = /* @__PURE__ */ new Map();
3920
+ /**
3921
+ * When THIS instance settled a remote step from its result, by `runId:seq`. That result's resume
3922
+ * runs here, and `execute` lets it into a lease this instance holds, so its wake is never the
3923
+ * dropped one {@link redriveSettledSteps} recovers. Pruned by age (see {@link noteSettledHere}).
3924
+ */
3925
+ settledHere = /* @__PURE__ */ new Map();
3926
+ /**
3927
+ * The step fans (`ctx.all` over a `@Step`) a turn suspended on, by run, while that turn executes on
3928
+ * THIS instance — re-checked once its lease is released (see {@link redriveSettledStepFans}).
3929
+ */
3930
+ awaitedStepFans = /* @__PURE__ */ new Map();
3931
+ /** Step fans whose join barrier this instance opened, by `runId:firstSeq` — see {@link settledHere}. */
3932
+ stepFansFiredHere = /* @__PURE__ */ new Map();
3933
+ /** The tail of each run's queue of results being handled on this instance — see {@link serializedPerRun}. */
3934
+ resultChains = /* @__PURE__ */ new Map();
3870
3935
  /** Flow-control admission backend for remote steps (see {@link registerQueue}). */
3871
3936
  admission;
3872
3937
  /** Runs on THIS instance blocked on admission, by queue — woken early on a freed-slot signal. */
@@ -3898,6 +3963,8 @@ var WorkflowEngine = class {
3898
3963
  this.maxRecoveryAttempts = deps.maxRecoveryAttempts;
3899
3964
  this.remoteAdvanceSilenceMs = deps.remoteAdvanceSilenceMs;
3900
3965
  this.reconcileMs = deps.reconcileMs === void 0 ? 3e5 : deps.reconcileMs || void 0;
3966
+ const batch = deps.timerBatchSize;
3967
+ this.timerBatchSize = batch !== void 0 && Number.isFinite(batch) && batch >= 1 ? Math.floor(batch) : 500;
3901
3968
  this.eventBufferTtlMs = deps.eventBufferTtlMs || void 0;
3902
3969
  this.remoteRedispatchMs = deps.remoteRedispatchMs || void 0;
3903
3970
  this.remoteRedispatchMax = Math.max(1, deps.remoteRedispatchMax ?? 10);
@@ -3950,7 +4017,11 @@ var WorkflowEngine = class {
3950
4017
  }
3951
4018
  return;
3952
4019
  }
3953
- await this.completeRemoteResult(result);
4020
+ if (!await this.completeRemoteResult(result)) return;
4021
+ await this.serializedPerRun(result.runId, () => this.resumeImplicit(result.runId).then(() => void 0, (err) => {
4022
+ if (err instanceof NamespaceMismatch) return;
4023
+ throw err;
4024
+ }));
3954
4025
  },
3955
4026
  // A heartbeat resets the liveness window for whatever it targets: an in-flight long STEP (keyed
3956
4027
  // by stepId, see callRemote) or — when stepId is absent — the in-flight workflow TURN (keyed by
@@ -4810,7 +4881,7 @@ var WorkflowEngine = class {
4810
4881
  * scan, filtered to just this tick's batch, keeps it cheap regardless of store size.
4811
4882
  */
4812
4883
  async resumeDueTimers(nowMs = this.clock()) {
4813
- const due = await this.store.listDueTimers(nowMs, this.namespace);
4884
+ const due = shuffle(await this.store.listDueTimers(nowMs, this.namespace, this.timerBatchSize));
4814
4885
  const blocked = await this.dueBlockedRuns(nowMs);
4815
4886
  if (due.length === 0 && blocked.length === 0) return [];
4816
4887
  const waiterByRun = indexWaitersByRun(await listSignalWaitersOfRuns(this.store, [
@@ -5294,11 +5365,47 @@ var WorkflowEngine = class {
5294
5365
  for (const seq of seqs) {
5295
5366
  const cp = await this.store.getCheckpoint(runId, seq);
5296
5367
  if (cp && (cp.status === "completed" || cp.status === "failed")) {
5368
+ if (this.settledHere.has(`${runId}:${seq}`)) return;
5369
+ const nowMs = this.clock();
5370
+ if (!await this.store.tryLockRun(runId, this.instanceId, nowMs + this.leaseMs, nowMs)) {
5371
+ return;
5372
+ }
5373
+ const run = await this.store.getRun(runId);
5374
+ if (run?.status !== "suspended") {
5375
+ await this.store.releaseRunLock(runId);
5376
+ return;
5377
+ }
5297
5378
  await this.resumeImplicit(runId);
5298
5379
  return;
5299
5380
  }
5300
5381
  }
5301
5382
  }
5383
+ /**
5384
+ * {@link redriveSettledSteps} for a step fan: once the turn that suspended on it released its lease,
5385
+ * re-drive the run if no step of the fan is still `pending` — the last result landed on another
5386
+ * instance while this one held the lease, and its wake was dropped there. A fan whose barrier this
5387
+ * instance opened is skipped, for the reason {@link settledHere} gives.
5388
+ */
5389
+ async redriveSettledStepFans(runId, fans) {
5390
+ for (const fan of fans) {
5391
+ if (this.stepFansFiredHere.has(`${runId}:${fan.from}`)) continue;
5392
+ const pending = this.store.countCheckpoints ? await this.store.countCheckpoints(runId, fan.from, fan.to, [
5393
+ "pending"
5394
+ ]) : (await this.store.listCheckpoints(runId)).filter((c) => c.seq >= fan.from && c.seq <= fan.to && c.status === "pending").length;
5395
+ if (pending > 0) continue;
5396
+ const nowMs = this.clock();
5397
+ if (!await this.store.tryLockRun(runId, this.instanceId, nowMs + this.leaseMs, nowMs)) {
5398
+ return;
5399
+ }
5400
+ const run = await this.store.getRun(runId);
5401
+ if (run?.status !== "suspended") {
5402
+ await this.store.releaseRunLock(runId);
5403
+ return;
5404
+ }
5405
+ await this.resumeImplicit(runId);
5406
+ return;
5407
+ }
5408
+ }
5302
5409
  async redriveSettledChildren(runId, fan) {
5303
5410
  if (!this.store.hasSignalWaiter) return;
5304
5411
  for (const wait of fan.waits) {
@@ -6756,6 +6863,8 @@ var WorkflowEngine = class {
6756
6863
  let suspendedOnSteps = false;
6757
6864
  const awaitedSteps = /* @__PURE__ */ new Set();
6758
6865
  this.awaitedRemoteSteps.set(run.id, awaitedSteps);
6866
+ const awaitedFans = [];
6867
+ this.awaitedStepFans.set(run.id, awaitedFans);
6759
6868
  const ctx = createWorkflowCtx(this.ctxHostFor(replay, run.namespace), run.id, compensations, run.workflow, registered?.searchAttributesSchema);
6760
6869
  try {
6761
6870
  const output = await runInWorkflowCtx(ctx, () => fn(ctx, run.input));
@@ -6835,6 +6944,7 @@ var WorkflowEngine = class {
6835
6944
  if (this.awaitedRemoteSteps.get(run.id) === awaitedSteps) {
6836
6945
  this.awaitedRemoteSteps.delete(run.id);
6837
6946
  }
6947
+ if (this.awaitedStepFans.get(run.id) === awaitedFans) this.awaitedStepFans.delete(run.id);
6838
6948
  await this.store.releaseRunLock(run.id);
6839
6949
  if (suspendedOnFan) {
6840
6950
  void this.redriveSettledChildren(run.id, suspendedOnFan).catch(() => void 0);
@@ -6842,6 +6952,9 @@ var WorkflowEngine = class {
6842
6952
  if (suspendedOnSteps && awaitedSteps.size > 0) {
6843
6953
  void this.redriveSettledSteps(run.id, awaitedSteps).catch(() => void 0);
6844
6954
  }
6955
+ if (suspendedOnSteps && awaitedFans.length > 0) {
6956
+ void this.redriveSettledStepFans(run.id, awaitedFans).catch(() => void 0);
6957
+ }
6845
6958
  }
6846
6959
  }
6847
6960
  /**
@@ -7064,6 +7177,9 @@ var WorkflowEngine = class {
7064
7177
  callRemote: /* @__PURE__ */ __name((runId, seq, step, input, queue, transport, admission) => this.callRemote(runId, seq, step, input, queue, transport, replay, admission, {
7065
7178
  namespace: parentNamespace
7066
7179
  }), "callRemote"),
7180
+ callRemoteAll: /* @__PURE__ */ __name((runId, positions, step, inputs, opts) => this.callRemoteAll(runId, positions, step, inputs, opts, {
7181
+ namespace: parentNamespace
7182
+ }), "callRemoteAll"),
7067
7183
  // Defer so a fast child can't reentrantly resume a still-running parent. Stamp the child with
7068
7184
  // the PARENT run's namespace (not this engine's) so a child belongs to the tenant/partition of
7069
7185
  // the run it was spawned from — even when an operator (namespace: undefined, drives every
@@ -7365,6 +7481,9 @@ var WorkflowEngine = class {
7365
7481
  input,
7366
7482
  attempts: attempt,
7367
7483
  workerGroup: group,
7484
+ ...routing?.parallelGroup !== void 0 ? {
7485
+ parallelGroup: routing.parallelGroup
7486
+ } : {},
7368
7487
  enqueuedAt,
7369
7488
  startedAt: enqueuedAt,
7370
7489
  finishedAt: enqueuedAt
@@ -7390,6 +7509,149 @@ var WorkflowEngine = class {
7390
7509
  this.awaitedRemoteSteps.get(runId)?.add(seq);
7391
7510
  throw new WorkflowSuspended();
7392
7511
  }
7512
+ /**
7513
+ * `ctx.all` over a `@Step`: one durable step per input, dispatched together, joined once. The ctx
7514
+ * reserves the whole position block up front, so replay re-derives the same seqs.
7515
+ *
7516
+ * Each turn reads the fan's checkpoints fresh, hands every step that already has one to
7517
+ * {@link callRemote} (which owns the completed / pending-lease / failed-retry rules), and sends
7518
+ * the steps nobody has sent yet: all of them, or with `concurrency`, the next wave once the last
7519
+ * one has drained. New steps are written `pending` BEFORE any is dispatched, so a turn that runs
7520
+ * concurrently with this one finds them already sent instead of sending them again. The join
7521
+ * barrier ({@link stepFanStillRunning}) then wakes the parent once per wave, not once per result.
7522
+ */
7523
+ async callRemoteAll(runId, positions, step, inputs, opts, routing) {
7524
+ if (step.timeoutMs) {
7525
+ throw new Error(`ctx.all: step "${step.name}" has a liveness timeoutMs; a step fan needs durable steps`);
7526
+ }
7527
+ if (opts.concurrency !== void 0 && (!Number.isInteger(opts.concurrency) || opts.concurrency < 1)) {
7528
+ throw new Error(`ctx.all: concurrency must be a positive integer, got ${opts.concurrency}`);
7529
+ }
7530
+ if (this.pool.size === 0) throw new Error("remote steps require a Transport");
7531
+ const first = positions[0];
7532
+ const last = positions[positions.length - 1];
7533
+ const group = stepFanGroup(first, positions.length, opts.mode);
7534
+ const bySeq = /* @__PURE__ */ new Map();
7535
+ for (const cp of await this.store.listCheckpoints(runId)) {
7536
+ if (cp.seq >= first && cp.seq <= last) bySeq.set(cp.seq, cp);
7537
+ }
7538
+ const outputs = new Array(inputs.length);
7539
+ const failures = [];
7540
+ const unsent = [];
7541
+ let inFlight = 0;
7542
+ let wakeAt;
7543
+ for (let i = 0; i < inputs.length; i += 1) {
7544
+ const seq = positions[i];
7545
+ const cp = bySeq.get(seq);
7546
+ if (!cp) {
7547
+ unsent.push(i);
7548
+ continue;
7549
+ }
7550
+ if (cp.name !== step.name) throw new NonDeterminismError(runId, seq, step.name, cp.name);
7551
+ if (cp.status === "completed") {
7552
+ outputs[i] = cp.output;
7553
+ continue;
7554
+ }
7555
+ try {
7556
+ outputs[i] = await this.callRemote(runId, seq, step, inputs[i], void 0, opts.transport, void 0, opts.admission, {
7557
+ ...routing,
7558
+ parallelGroup: group
7559
+ });
7560
+ } catch (err) {
7561
+ if (err instanceof WorkflowSuspended) {
7562
+ inFlight += 1;
7563
+ if (err.wakeAt !== void 0) wakeAt = Math.min(wakeAt ?? err.wakeAt, err.wakeAt);
7564
+ } else if (err instanceof RemoteStepError) {
7565
+ const failure = {
7566
+ index: i,
7567
+ id: stepId(runId, seq),
7568
+ error: err.message
7569
+ };
7570
+ if (opts.mode === "failFast") throw new GatherError([
7571
+ failure
7572
+ ], inputs.length);
7573
+ failures.push(failure);
7574
+ } else {
7575
+ throw err;
7576
+ }
7577
+ } finally {
7578
+ this.awaitedRemoteSteps.get(runId)?.delete(seq);
7579
+ }
7580
+ }
7581
+ const batch = inFlight > 0 ? [] : opts.concurrency === void 0 ? unsent : unsent.slice(0, opts.concurrency);
7582
+ if (batch.length > 0) await this.sendStepFan(runId, positions, step, inputs, batch, group, opts, routing);
7583
+ if (inFlight + batch.length > 0) {
7584
+ this.awaitedStepFans.get(runId)?.push({
7585
+ from: first,
7586
+ to: last
7587
+ });
7588
+ throw new WorkflowSuspended(wakeAt);
7589
+ }
7590
+ if (failures.length > 0) throw new GatherError(failures, inputs.length);
7591
+ return outputs;
7592
+ }
7593
+ /** Write the `pending` checkpoints of a step fan's new steps, then dispatch them. */
7594
+ async sendStepFan(runId, positions, step, inputs, batch, group, opts, routing) {
7595
+ const workerGroup = stepGroup(step, routing?.namespace);
7596
+ await this.ensureRoutable(workerGroup, step.requires);
7597
+ const enqueuedAt = /* @__PURE__ */ new Date();
7598
+ for (let at = 0; at < batch.length; at += STEP_FAN_CHUNK) {
7599
+ await Promise.all(batch.slice(at, at + STEP_FAN_CHUNK).map((i) => {
7600
+ const seq = positions[i];
7601
+ return this.store.saveCheckpoint({
7602
+ runId,
7603
+ seq,
7604
+ name: step.name,
7605
+ kind: "remote",
7606
+ stepId: stepId(runId, seq),
7607
+ status: "pending",
7608
+ input: inputs[i],
7609
+ attempts: 1,
7610
+ workerGroup,
7611
+ parallelGroup: group,
7612
+ enqueuedAt,
7613
+ startedAt: enqueuedAt,
7614
+ finishedAt: enqueuedAt
7615
+ });
7616
+ }));
7617
+ }
7618
+ for (let at = 0; at < batch.length; at += STEP_FAN_CHUNK) {
7619
+ await Promise.all(batch.slice(at, at + STEP_FAN_CHUNK).map(async (i) => {
7620
+ const seq = positions[i];
7621
+ await this.dispatchRemoteTask({
7622
+ runId,
7623
+ seq,
7624
+ name: step.name,
7625
+ stepId: stepId(runId, seq),
7626
+ group: workerGroup,
7627
+ input: inputs[i],
7628
+ priority: opts.admission?.priority,
7629
+ attempt: 1,
7630
+ transport: opts.transport
7631
+ });
7632
+ this.emit({
7633
+ type: "step.started",
7634
+ runId,
7635
+ seq,
7636
+ name: step.name,
7637
+ kind: "remote"
7638
+ });
7639
+ }));
7640
+ }
7641
+ }
7642
+ /** Remember a step settled here, forgetting entries older than a lease so the map stays small. */
7643
+ noteSettledHere(key) {
7644
+ this.noteRecent(this.settledHere, key);
7645
+ }
7646
+ /** Record `key` in a "happened here recently" map, pruning entries older than a lease. */
7647
+ noteRecent(map, key) {
7648
+ const now = this.clock();
7649
+ map.set(key, now);
7650
+ if (map.size < 1024) return;
7651
+ for (const [k, at] of map) {
7652
+ if (now - at > this.leaseMs) map.delete(k);
7653
+ }
7654
+ }
7393
7655
  /** {@link StateStore.saveCheckpointIfPending}, or a plain save on a store without it. */
7394
7656
  async savePendingCheckpoint(checkpoint) {
7395
7657
  if (this.store.saveCheckpointIfPending) return this.store.saveCheckpointIfPending(checkpoint);
@@ -7414,16 +7676,33 @@ var WorkflowEngine = class {
7414
7676
  }, task.transport);
7415
7677
  }
7416
7678
  /**
7417
- * Complete a durable remote step from its worker result and resume the run — runs on whichever
7418
- * instance receives the result (the dispatching one may be gone), so the run is crash/scale-safe.
7419
- * A no-op if the checkpoint isn't `pending` (a duplicate or late delivery).
7679
+ * Run `work` after every earlier `work` for the same run on this instance has settled. A result
7680
+ * resumes its run inline, and `execute` lets this instance into a lease it already holds — so two
7681
+ * results of one run handled at once here would run two turns of it concurrently. Results of
7682
+ * different runs still go in parallel, which is what a results consumer with concurrency > 1 is for.
7683
+ */
7684
+ serializedPerRun(runId, work) {
7685
+ const previous = this.resultChains.get(runId) ?? Promise.resolve();
7686
+ const next = previous.then(work, work);
7687
+ const tail = next.catch(() => void 0);
7688
+ this.resultChains.set(runId, tail);
7689
+ void tail.then(() => {
7690
+ if (this.resultChains.get(runId) === tail) this.resultChains.delete(runId);
7691
+ });
7692
+ return next;
7693
+ }
7694
+ /**
7695
+ * Complete a durable remote step from its worker result — runs on whichever instance receives the
7696
+ * result (the dispatching one may be gone), so the run is crash/scale-safe. Resolves whether the
7697
+ * run should now be resumed: not for a duplicate or late delivery (the checkpoint isn't `pending`),
7698
+ * a run already finished, or a step fan whose barrier is still closed.
7420
7699
  */
7421
7700
  async completeRemoteResult(result) {
7422
7701
  const cp = await this.store.getCheckpoint(result.runId, result.seq);
7423
- if (!cp || cp.status !== "pending") return;
7702
+ if (!cp || cp.status !== "pending") return false;
7424
7703
  void this.releaseQueueSlot(cp.stepId).catch(() => void 0);
7425
7704
  const run = await this.store.getRun(result.runId);
7426
- if (run && (run.status === "cancelled" || run.status === "completed")) return;
7705
+ if (run && (run.status === "cancelled" || run.status === "completed")) return false;
7427
7706
  const finishedAt = /* @__PURE__ */ new Date();
7428
7707
  const startedAt = result.startedAt ? new Date(result.startedAt) : cp.startedAt;
7429
7708
  const claimed = await this.savePendingCheckpoint({
@@ -7435,7 +7714,8 @@ var WorkflowEngine = class {
7435
7714
  startedAt,
7436
7715
  finishedAt
7437
7716
  });
7438
- if (!claimed) return;
7717
+ if (!claimed) return false;
7718
+ this.noteSettledHere(`${result.runId}:${result.seq}`);
7439
7719
  this.emit({
7440
7720
  type: result.status === "completed" ? "step.completed" : "step.failed",
7441
7721
  runId: result.runId,
@@ -7447,10 +7727,25 @@ var WorkflowEngine = class {
7447
7727
  queueMs: startedAt.getTime() - cp.enqueuedAt.getTime(),
7448
7728
  durationMs: finishedAt.getTime() - startedAt.getTime()
7449
7729
  });
7450
- await this.resumeImplicit(result.runId).catch((err) => {
7451
- if (err instanceof NamespaceMismatch) return null;
7452
- throw err;
7453
- });
7730
+ return !await this.stepFanStillRunning(cp, result);
7731
+ }
7732
+ /**
7733
+ * The join barrier of a step fan: a result for one of its steps resumes the parent only once no
7734
+ * step of the fan is still `pending` (a settled wave), or on a failure (failFast surfaces it, and a
7735
+ * retry needs the parent to schedule it). Without it every result would replay the whole parent,
7736
+ * which makes a fan of N steps cost O(N²) checkpoint reads.
7737
+ */
7738
+ async stepFanStillRunning(cp, result) {
7739
+ const fan = parseStepFanGroup(cp.parallelGroup);
7740
+ if (!fan) return false;
7741
+ if (result.status === "failed") return false;
7742
+ const to = fan.firstSeq + fan.size - 1;
7743
+ const pending = this.store.countCheckpoints ? await this.store.countCheckpoints(result.runId, fan.firstSeq, to, [
7744
+ "pending"
7745
+ ]) : (await this.store.listCheckpoints(result.runId)).filter((c) => c.seq >= fan.firstSeq && c.seq <= to && c.status === "pending").length;
7746
+ if (pending > 0) return true;
7747
+ this.noteRecent(this.stepFansFiredHere, `${result.runId}:${fan.firstSeq}`);
7748
+ return false;
7454
7749
  }
7455
7750
  /**
7456
7751
  * Apply a remote workflow-turn {@link WorkflowDecision} durably — runs on WHATEVER instance receives
@@ -7831,6 +8126,8 @@ var CodecStateStore = class {
7831
8126
  /** Encodes like {@link saveCheckpoint}. Bound only when the inner store has it, so the engine's
7832
8127
  * plain-save fallback still kicks in through the wrapper. */
7833
8128
  saveCheckpointIfPending;
8129
+ /** Forwarded verbatim — a count reads no payload. Bound only when the inner store has it. */
8130
+ countCheckpoints;
7834
8131
  /** Forwarded verbatim — retention selects by status/updatedAt/scope, never by payload. Bound only
7835
8132
  * when the inner store prunes, so the retention poller still sees (and warns about) its absence. */
7836
8133
  pruneTerminalRuns;
@@ -7860,6 +8157,8 @@ var CodecStateStore = class {
7860
8157
  if (deleteRuns) this.deleteRuns = (ids) => deleteRuns.call(inner, ids);
7861
8158
  const saveIfPending = inner.saveCheckpointIfPending;
7862
8159
  if (saveIfPending) this.saveCheckpointIfPending = (cp) => saveIfPending.call(inner, this.encCp(cp));
8160
+ const countCps = inner.countCheckpoints;
8161
+ if (countCps) this.countCheckpoints = (runId, from, to, statuses) => countCps.call(inner, runId, from, to, statuses);
7863
8162
  const prune = inner.pruneTerminalRuns;
7864
8163
  if (prune) this.pruneTerminalRuns = (policy, now, limit) => prune.call(inner, policy, now, limit);
7865
8164
  const count = inner.countRuns;
@@ -7990,8 +8289,8 @@ var CodecStateStore = class {
7990
8289
  async listPendingRuns(limit, namespace) {
7991
8290
  return (await this.inner.listPendingRuns(limit, namespace)).map((r) => this.decRun(r));
7992
8291
  }
7993
- async listDueTimers(nowMs, namespace) {
7994
- return (await this.inner.listDueTimers(nowMs, namespace)).map((r) => this.decRun(r));
8292
+ async listDueTimers(nowMs, namespace, limit) {
8293
+ return (await this.inner.listDueTimers(nowMs, namespace, limit)).map((r) => this.decRun(r));
7995
8294
  }
7996
8295
  tryLockRun(runId, owner, leaseUntilMs, nowMs) {
7997
8296
  return this.inner.tryLockRun(runId, owner, leaseUntilMs, nowMs);
@@ -8445,10 +8744,11 @@ var InMemoryStateStore = class {
8445
8744
  ...r
8446
8745
  }));
8447
8746
  }
8448
- async listDueTimers(nowMs, namespace) {
8449
- return [
8747
+ async listDueTimers(nowMs, namespace, limit) {
8748
+ const due = [
8450
8749
  ...this.runs.values()
8451
- ].filter((r) => r.status === "suspended" && r.wakeAt !== void 0 && r.wakeAt <= nowMs && (namespace === void 0 || r.namespace === namespace)).map((r) => ({
8750
+ ].filter((r) => r.status === "suspended" && r.wakeAt !== void 0 && r.wakeAt <= nowMs && (namespace === void 0 || r.namespace === namespace)).sort((a, b) => (a.wakeAt ?? 0) - (b.wakeAt ?? 0));
8751
+ return (limit === void 0 ? due : due.slice(0, limit)).map((r) => ({
8452
8752
  ...r
8453
8753
  }));
8454
8754
  }
@@ -8722,6 +9022,13 @@ var InMemoryStateStore = class {
8722
9022
  ...latest
8723
9023
  } : void 0;
8724
9024
  }
9025
+ async countCheckpoints(runId, fromSeq, toSeq, statuses) {
9026
+ let n = 0;
9027
+ for (const cp of this.checkpoints.values()) {
9028
+ if (cp.runId === runId && cp.seq >= fromSeq && cp.seq <= toSeq && statuses.includes(cp.status)) n += 1;
9029
+ }
9030
+ return n;
9031
+ }
8725
9032
  async listCheckpointsByNamePrefix(runId, prefixes) {
8726
9033
  const rows = [
8727
9034
  ...this.checkpoints.values()