@adonis-agora/durable 0.54.0 → 0.54.2

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.
Files changed (66) hide show
  1. package/CHANGELOG.md +90 -0
  2. package/dist/assets/spa/assets/index-9kO8tWjg.js +16 -0
  3. package/dist/assets/spa/assets/index-9kO8tWjg.js.map +1 -0
  4. package/dist/assets/spa/index.html +1 -1
  5. package/dist/src/buffered-child-signals.d.ts +20 -4
  6. package/dist/src/buffered-child-signals.d.ts.map +1 -1
  7. package/dist/src/buffered-child-signals.js +207 -63
  8. package/dist/src/buffered-child-signals.js.map +1 -1
  9. package/dist/src/codec-state-store.d.ts +10 -1
  10. package/dist/src/codec-state-store.d.ts.map +1 -1
  11. package/dist/src/codec-state-store.js +32 -0
  12. package/dist/src/codec-state-store.js.map +1 -1
  13. package/dist/src/commands/worker.d.ts +2 -1
  14. package/dist/src/commands/worker.d.ts.map +1 -1
  15. package/dist/src/commands/worker.js +44 -3
  16. package/dist/src/commands/worker.js.map +1 -1
  17. package/dist/src/dashboard/index.d.ts +1 -1
  18. package/dist/src/dashboard/index.js +1 -1
  19. package/dist/src/engine.d.ts +69 -8
  20. package/dist/src/engine.d.ts.map +1 -1
  21. package/dist/src/engine.js +248 -53
  22. package/dist/src/engine.js.map +1 -1
  23. package/dist/src/index.d.ts +1 -1
  24. package/dist/src/index.js +1 -1
  25. package/dist/src/interfaces.d.ts +47 -0
  26. package/dist/src/interfaces.d.ts.map +1 -1
  27. package/dist/src/interfaces.js.map +1 -1
  28. package/dist/src/stores/lucid-mappers.d.ts +6 -1
  29. package/dist/src/stores/lucid-mappers.d.ts.map +1 -1
  30. package/dist/src/stores/lucid-mappers.js +41 -0
  31. package/dist/src/stores/lucid-mappers.js.map +1 -1
  32. package/dist/src/stores/lucid.d.ts +4 -1
  33. package/dist/src/stores/lucid.d.ts.map +1 -1
  34. package/dist/src/stores/lucid.js +148 -37
  35. package/dist/src/stores/lucid.js.map +1 -1
  36. package/dist/src/testing/in-memory-state-store.d.ts +4 -1
  37. package/dist/src/testing/in-memory-state-store.d.ts.map +1 -1
  38. package/dist/src/testing/in-memory-state-store.js +29 -0
  39. package/dist/src/testing/in-memory-state-store.js.map +1 -1
  40. package/dist/src/testing-kit/run-dispatch-contract.d.ts.map +1 -1
  41. package/dist/src/testing-kit/run-dispatch-contract.js +12 -3
  42. package/dist/src/testing-kit/run-dispatch-contract.js.map +1 -1
  43. package/dist/src/testing-kit/state-store-conformance.d.ts +6 -0
  44. package/dist/src/testing-kit/state-store-conformance.d.ts.map +1 -1
  45. package/dist/src/testing-kit/state-store-conformance.js +230 -0
  46. package/dist/src/testing-kit/state-store-conformance.js.map +1 -1
  47. package/dist/src/transports/adaptive-concurrency.d.ts +2 -0
  48. package/dist/src/transports/adaptive-concurrency.d.ts.map +1 -1
  49. package/dist/src/transports/adaptive-concurrency.js +39 -9
  50. package/dist/src/transports/adaptive-concurrency.js.map +1 -1
  51. package/dist/src/transports/bullmq/bullmq-transport.d.ts.map +1 -1
  52. package/dist/src/transports/bullmq/bullmq-transport.js +8 -4
  53. package/dist/src/transports/bullmq/bullmq-transport.js.map +1 -1
  54. package/dist/src/transports/bullmq/deps.d.ts.map +1 -1
  55. package/dist/src/transports/bullmq/deps.js +18 -1
  56. package/dist/src/transports/bullmq/deps.js.map +1 -1
  57. package/dist/src/transports/db.js +4 -4
  58. package/dist/src/transports/db.js.map +1 -1
  59. package/dist/src/transports/queue.d.ts.map +1 -1
  60. package/dist/src/transports/queue.js +8 -4
  61. package/dist/src/transports/queue.js.map +1 -1
  62. package/package.json +2 -2
  63. package/skills/durable-reliability/SKILL.md +2 -1
  64. package/skills/durable-transports-stores/SKILL.md +15 -2
  65. package/dist/assets/spa/assets/index-BEk3-95l.js +0 -16
  66. package/dist/assets/spa/assets/index-BEk3-95l.js.map +0 -1
@@ -122,6 +122,11 @@ const DEFAULT_TIMER_BATCH = 500;
122
122
  const BUFFERED_SWEEP_PAGE = 500;
123
123
  /** {@link WorkflowEngine.sweepBufferedSignals}: pages per call, so one tick stays short. */
124
124
  const BUFFERED_SWEEP_PAGES_PER_CALL = 4;
125
+ /**
126
+ * {@link WorkflowEngine.sweepBufferedSignals}: wall-clock budget of one call. Past it the sweep
127
+ * starts no further page or parent read, and the next call resumes where this one stopped.
128
+ */
129
+ const BUFFERED_SWEEP_BUDGET_MS = 250;
125
130
  /** {@link WorkflowEngine.sweepBufferedSignals}: rest between two full passes over the table. */
126
131
  const BUFFERED_SWEEP_REST_MS = 10 * 60_000;
127
132
  /** The most results of one run settled by one batched write (see `completeRemoteResult`). */
@@ -402,6 +407,14 @@ export class WorkflowEngine {
402
407
  turnSettledFans = new Map();
403
408
  /** The kept wake reasons each executing turn on this instance serves (see {@link execute}). */
404
409
  turnServedWakes = new Map();
410
+ /**
411
+ * `<runId>:<seq>` of the steps (not of a step fan) whose result this instance settled and whose
412
+ * run it then resumed, with when, kept for a lease. That resume runs a turn after any this instance
413
+ * is executing, or asks the lease holder for one, so the re-check after an in-process turn
414
+ * ({@link recheckWaitSeqs}) has no wake to recover for such a step: resuming again would only
415
+ * replay the run once more.
416
+ */
417
+ settledHere = new Map();
405
418
  /** {@link sweepBufferedSignals}: the id the pass in progress resumes after, `undefined` between passes. */
406
419
  bufferedSweepCursor;
407
420
  bufferedSweepActive = false;
@@ -539,6 +552,7 @@ export class WorkflowEngine {
539
552
  this.latest.get(run.workflow))?.singleton,
540
553
  });
541
554
  this.pool.bind(async (result) => {
555
+ this.heartbeatWrites.delete(result.stepId);
542
556
  // In-memory path (a `timeoutMs` step awaiting on THIS instance): resolve its pending promise.
543
557
  const waiter = this.pending.get(result.stepId);
544
558
  if (waiter) {
@@ -1153,9 +1167,35 @@ export class WorkflowEngine {
1153
1167
  if (this.heartbeatPersists.size > 1024)
1154
1168
  this.heartbeatPersists.clear();
1155
1169
  this.heartbeatPersists.set(id, { at: now, hadProgress: beat.progress !== undefined });
1156
- void this.store
1157
- .recordStepHeartbeat(beat.runId, beat.seq, new Date(beat.at ?? now), beat.progress)
1158
- .catch(() => undefined);
1170
+ this.heartbeatWrites.set(id, {
1171
+ runId: beat.runId,
1172
+ seq: beat.seq,
1173
+ at: new Date(beat.at ?? now),
1174
+ progress: beat.progress,
1175
+ });
1176
+ if (!this.heartbeatWriter) {
1177
+ this.heartbeatWriter = this.writeHeartbeats().finally(() => {
1178
+ this.heartbeatWriter = undefined;
1179
+ });
1180
+ this.trackEffect(this.heartbeatWriter);
1181
+ }
1182
+ }
1183
+ /** Beats {@link persistHeartbeat} let through and {@link writeHeartbeats} has not written yet, by
1184
+ * stepId: a newer beat of the same step replaces the queued one, a step's result drops it. */
1185
+ heartbeatWrites = new Map();
1186
+ heartbeatWriter;
1187
+ /**
1188
+ * Write the queued beats one at a time, oldest first, until none is left. One write in flight keeps
1189
+ * a burst of beats (every step of a wide fan sends one when a worker picks it up) from taking the
1190
+ * store's connections away from the turns and results, and a step that finishes while its beat
1191
+ * waits never has it written: nothing reads a settled step's liveness.
1192
+ */
1193
+ async writeHeartbeats() {
1194
+ const record = this.store.recordStepHeartbeat?.bind(this.store);
1195
+ for (const [id, beat] of this.heartbeatWrites) {
1196
+ this.heartbeatWrites.delete(id);
1197
+ await record?.(beat.runId, beat.seq, beat.at, beat.progress).catch(() => undefined);
1198
+ }
1159
1199
  }
1160
1200
  /**
1161
1201
  * Declare this process a worker: flush consumer subscriptions parked by a transport whose
@@ -1582,13 +1622,18 @@ export class WorkflowEngine {
1582
1622
  *
1583
1623
  * Walks the buffered signals by primary key, {@link BUFFERED_SWEEP_PAGE} rows at a time and at most
1584
1624
  * {@link BUFFERED_SWEEP_PAGES_PER_CALL} pages per call, so a large backlog drains over a few worker
1585
- * ticks without a long query or a big delete. A pass that reaches the end rests
1586
- * {@link BUFFERED_SWEEP_REST_MS} before the next. The worker loop calls it every tick. Skipped while
1587
- * draining, on a namespaced engine (a buffered signal carries no namespace, and a scoped read would
1588
- * report another tenant's parent as gone), and on a store without
1589
- * `listBufferedSignals`/`deleteBufferedSignals`. Returns how many rows it deleted.
1590
- */
1591
- async sweepBufferedSignals(now = this.clock()) {
1625
+ * ticks without a long query or a big delete. Each page is decided set-based (see
1626
+ * `unneededChildSignals`): a handful of store calls per page and per open parent, none of them a
1627
+ * read by checkpoint name for an id that names its seq. A call stops once `budgetMs` of wall-clock
1628
+ * time is spent (default {@link BUFFERED_SWEEP_BUDGET_MS}), between pages or between two parents'
1629
+ * reads, and the next call resumes there. A pass that reaches the end rests
1630
+ * {@link BUFFERED_SWEEP_REST_MS} before the next. The worker loop starts it every tick without
1631
+ * waiting for it; a call made while another is running returns 0. Skipped while draining, on a
1632
+ * namespaced engine (a buffered signal carries no namespace, and a scoped read would report another
1633
+ * tenant's parent as gone), and on a store without `listBufferedSignals`/`deleteBufferedSignals`.
1634
+ * Returns how many rows it deleted.
1635
+ */
1636
+ async sweepBufferedSignals(now = this.clock(), budgetMs = BUFFERED_SWEEP_BUDGET_MS) {
1592
1637
  const { listBufferedSignals, deleteBufferedSignals } = this.store;
1593
1638
  if (!listBufferedSignals || !deleteBufferedSignals)
1594
1639
  return 0;
@@ -1599,15 +1644,24 @@ export class WorkflowEngine {
1599
1644
  return 0;
1600
1645
  }
1601
1646
  this.bufferedSweepActive = true;
1647
+ const deadline = Date.now() + budgetMs;
1602
1648
  let deleted = 0;
1603
1649
  try {
1604
1650
  for (let page = 0; page < BUFFERED_SWEEP_PAGES_PER_CALL; page += 1) {
1651
+ if (page > 0 && (Date.now() >= deadline || this.draining))
1652
+ break;
1605
1653
  const rows = await listBufferedSignals.call(this.store, CHILD_TOKEN_PREFIX, this.bufferedSweepCursor, BUFFERED_SWEEP_PAGE);
1606
- const doomed = await unneededChildSignals(this.store, rows);
1654
+ const { doomed, decided } = await unneededChildSignals(this.store, rows, deadline);
1607
1655
  if (doomed.length > 0) {
1608
1656
  await deleteBufferedSignals.call(this.store, doomed.map((row) => row.id));
1609
1657
  deleted += doomed.length;
1610
1658
  }
1659
+ if (decided < rows.length) {
1660
+ // Out of time mid-page: resume after the last row decided.
1661
+ if (decided > 0)
1662
+ this.bufferedSweepCursor = rows[decided - 1].id;
1663
+ break;
1664
+ }
1611
1665
  const last = rows.at(-1);
1612
1666
  if (rows.length < BUFFERED_SWEEP_PAGE || !last) {
1613
1667
  this.bufferedSweepCursor = undefined;
@@ -2108,13 +2162,19 @@ export class WorkflowEngine {
2108
2162
  * delivery's own resume no-oped against this turn's still-held lease (see the WorkflowSuspended
2109
2163
  * catch in runExecution). Deferred one macrotask so the caller's `finally` releases the lease
2110
2164
  * first; tracked so `drain()` waits for the re-driven turn's writes.
2165
+ *
2166
+ * With `settledHereResumes`, a step in {@link settledHere} needs no re-drive: the turn that
2167
+ * suspended on it read its snapshot before the settle, so the settle's resume queued behind that
2168
+ * turn or asked the lease holder for one. A remote workflow's decision is applied after the turn
2169
+ * that computed it ended, so the settle's resume may already have run, and it passes `false`.
2111
2170
  */
2112
- recheckWaitSeqs(runId, seqs) {
2171
+ recheckWaitSeqs(runId, seqs, settledHereResumes = false) {
2113
2172
  this.trackEffect((async () => {
2114
2173
  await new Promise((r) => setTimeout(r, 0));
2115
2174
  for (const seq of seqs) {
2116
- const cp = await this.store.getCheckpoint(runId, seq).catch(() => null);
2117
- if (cp && (cp.status === 'completed' || cp.status === 'failed')) {
2175
+ if (await this.checkpointSettled(runId, seq).catch(() => false)) {
2176
+ if (settledHereResumes && this.settledHere.has(`${runId}:${seq}`))
2177
+ return;
2118
2178
  await this.resumeImplicit(runId).catch(() => undefined);
2119
2179
  return;
2120
2180
  }
@@ -2140,6 +2200,14 @@ export class WorkflowEngine {
2140
2200
  })());
2141
2201
  }
2142
2202
  /** {@link StateStore.countCheckpoints} of `pending` steps, or a JS count on a store without it. */
2203
+ /** Whether the checkpoint at `seq` is settled: a `COUNT` when the store has one, else the row. */
2204
+ async checkpointSettled(runId, seq) {
2205
+ if (this.store.countCheckpoints) {
2206
+ return (await this.store.countCheckpoints(runId, seq, seq, ['completed', 'failed'])) > 0;
2207
+ }
2208
+ const cp = await this.store.getCheckpoint(runId, seq);
2209
+ return cp?.status === 'completed' || cp?.status === 'failed';
2210
+ }
2143
2211
  async countPendingSteps(runId, from, to) {
2144
2212
  if (this.store.countCheckpoints) {
2145
2213
  return this.store.countCheckpoints(runId, from, to, ['pending']);
@@ -4360,7 +4428,7 @@ export class WorkflowEngine {
4360
4428
  const routePlan = await this.planRoute(registered.convention ? conventionGroup(run.workflow, run.namespace) : remote.group, registered.requires);
4361
4429
  if (routePlan?.status === 'blocked')
4362
4430
  return this.parkBlocked(run, routePlan);
4363
- const checkpoints = await this.store.listCheckpoints(run.id);
4431
+ const checkpoints = await this.replayCheckpoints(run.id);
4364
4432
  const handoff = liveHandoff(run.id, checkpoints);
4365
4433
  if (handoff) {
4366
4434
  // A run that handed its failure to an onError handler only follows the handler from here on.
@@ -4924,6 +4992,42 @@ export class WorkflowEngine {
4924
4992
  /** Earliest clock time a step's lease is worth LOOKING at again (`stepId` → clock ms) — the read
4925
4993
  * throttle for {@link rearmStepLease}, the sibling of {@link heartbeatPersists}. */
4926
4994
  stepLeaseRearms = new Map();
4995
+ /**
4996
+ * {@link stepLeaseRearms} for the steps of a fan this instance sent, by run: one entry per sent
4997
+ * block of consecutive seqs instead of one per step, since they share the lease stamped at send.
4998
+ */
4999
+ fanLeaseRearms = new Map();
5000
+ noteFanLease(runId, seqs, lease) {
5001
+ if (this.remoteRedispatchMs == null || seqs.length === 0)
5002
+ return;
5003
+ let from = seqs[0];
5004
+ let to = from;
5005
+ for (const seq of seqs) {
5006
+ if (seq < from)
5007
+ from = seq;
5008
+ if (seq > to)
5009
+ to = seq;
5010
+ }
5011
+ // A gap would cover steps sent earlier, under an older lease.
5012
+ if (to - from + 1 !== seqs.length)
5013
+ return;
5014
+ if (this.fanLeaseRearms.size > 1024)
5015
+ this.fanLeaseRearms.clear();
5016
+ const blocks = this.fanLeaseRearms.get(runId) ?? [];
5017
+ blocks.push({ from, to, lookAgainAt: lease - Math.ceil(this.remoteRedispatchMs / 2) });
5018
+ this.fanLeaseRearms.set(runId, blocks);
5019
+ }
5020
+ fanLeaseLookAgainAt(runId, seq, now) {
5021
+ const blocks = this.fanLeaseRearms.get(runId);
5022
+ if (!blocks)
5023
+ return undefined;
5024
+ const live = blocks.filter((b) => now < b.lookAgainAt);
5025
+ if (live.length === 0)
5026
+ this.fanLeaseRearms.delete(runId);
5027
+ else if (live.length < blocks.length)
5028
+ this.fanLeaseRearms.set(runId, live);
5029
+ return live.find((b) => seq >= b.from && seq <= b.to)?.lookAgainAt;
5030
+ }
4927
5031
  /**
4928
5032
  * A step-scoped heartbeat for a durably-suspended remote step: push its LEASE forward so a worker
4929
5033
  * that is still working on it is never presumed dead and double-dispatched. The durable counterpart
@@ -4970,7 +5074,7 @@ export class WorkflowEngine {
4970
5074
  return;
4971
5075
  const id = stepId(runId, seq);
4972
5076
  const now = this.clock();
4973
- const lookAgainAt = this.stepLeaseRearms.get(id);
5077
+ const lookAgainAt = this.stepLeaseRearms.get(id) ?? this.fanLeaseLookAgainAt(runId, seq, now);
4974
5078
  if (lookAgainAt !== undefined && now < lookAgainAt)
4975
5079
  return;
4976
5080
  const cp = await this.store.getCheckpoint(runId, seq);
@@ -5378,7 +5482,7 @@ export class WorkflowEngine {
5378
5482
  // Read once at execution start: a checkpoint written AFTER this snapshot (the signal/timer/child
5379
5483
  // this resume wakes on, or one written later in this same execution) is absent from the map, and
5380
5484
  // the ctx falls back to the live store for any absent seq — so replay semantics are unchanged.
5381
- const snapshot = await this.store.listCheckpoints(run.id);
5485
+ const snapshot = await this.replayCheckpoints(run.id);
5382
5486
  const replay = new Map();
5383
5487
  for (const cp of snapshot)
5384
5488
  replay.set(cp.seq, cp);
@@ -5471,7 +5575,7 @@ export class WorkflowEngine {
5471
5575
  // all with reconcileMs: 0). Re-check the seqs this suspension waits on and re-drive if one
5472
5576
  // already settled.
5473
5577
  if (settled.status === 'suspended' && err.waitSeqs?.length) {
5474
- this.recheckWaitSeqs(run.id, err.waitSeqs);
5578
+ this.recheckWaitSeqs(run.id, err.waitSeqs, true);
5475
5579
  }
5476
5580
  if (settled.status === 'suspended' && err.stepFans?.length) {
5477
5581
  this.recheckStepFans(run.id, err.stepFans);
@@ -6292,7 +6396,11 @@ export class WorkflowEngine {
6292
6396
  ? unsent
6293
6397
  : unsent.slice(0, opts.concurrency);
6294
6398
  if (batch.length > 0) {
6295
- await this.sendStepFan(runId, positions, step, inputs, batch, group, opts, runNamespace);
6399
+ const lease = await this.sendStepFan(runId, positions, step, inputs, batch, group, opts, runNamespace);
6400
+ if (lease !== undefined) {
6401
+ const at = this.stepLeaseWakeAt(lease);
6402
+ wakeAt = Math.min(wakeAt ?? at, at);
6403
+ }
6296
6404
  }
6297
6405
  if (inFlight + batch.length > 0) {
6298
6406
  throw new WorkflowSuspended(wakeAt, undefined, undefined, [{ from: first, to: last }]);
@@ -6324,7 +6432,8 @@ export class WorkflowEngine {
6324
6432
  /**
6325
6433
  * Write the `pending` checkpoints of a step fan's new steps, then dispatch them: one
6326
6434
  * {@link StateStore.saveCheckpoints} call and one {@link TransportPool.dispatchMany} for the batch,
6327
- * where the store and transport have them.
6435
+ * where the store and transport have them. With the lost-dispatch self-heal on, each step is
6436
+ * written with its lease already stamped, and the lease is returned for the run to park on.
6328
6437
  */
6329
6438
  async sendStepFan(runId, positions, step, inputs, batch, group, opts, runNamespace) {
6330
6439
  const token = stepGroup(step, runNamespace);
@@ -6336,6 +6445,7 @@ export class WorkflowEngine {
6336
6445
  for (const i of batch)
6337
6446
  validInputs.set(i, await input(i));
6338
6447
  const enqueuedAt = new Date();
6448
+ const lease = this.remoteRedispatchMs == null ? undefined : this.clock() + this.remoteRedispatchMs;
6339
6449
  const checkpoints = batch.map((i) => {
6340
6450
  const seq = positions[i];
6341
6451
  return {
@@ -6352,6 +6462,7 @@ export class WorkflowEngine {
6352
6462
  enqueuedAt,
6353
6463
  startedAt: enqueuedAt,
6354
6464
  finishedAt: enqueuedAt,
6465
+ ...(lease !== undefined ? { wakeAt: lease } : {}),
6355
6466
  };
6356
6467
  });
6357
6468
  if (this.store.saveCheckpoints) {
@@ -6379,6 +6490,10 @@ export class WorkflowEngine {
6379
6490
  for (const cp of checkpoints) {
6380
6491
  this.emit({ type: 'step.started', runId, seq: cp.seq, name: step.name, kind: 'remote' });
6381
6492
  }
6493
+ if (lease !== undefined) {
6494
+ this.noteFanLease(runId, checkpoints.map((cp) => cp.seq), lease);
6495
+ }
6496
+ return lease;
6382
6497
  }
6383
6498
  /**
6384
6499
  * Enqueue a remote step task to its worker group via the transport pool. Shared by the initial
@@ -6399,6 +6514,13 @@ export class WorkflowEngine {
6399
6514
  attempt: task.attempt,
6400
6515
  }, task.transport);
6401
6516
  }
6517
+ /** A turn's read of its run's checkpoints: {@link StateStore.listCheckpointsForReplay}, or every
6518
+ * column through {@link StateStore.listCheckpoints} on a store without it. */
6519
+ replayCheckpoints(runId) {
6520
+ return this.store.listCheckpointsForReplay
6521
+ ? this.store.listCheckpointsForReplay(runId)
6522
+ : this.store.listCheckpoints(runId);
6523
+ }
6402
6524
  /**
6403
6525
  * Complete a durable remote step from its worker result and resume the run — runs on whichever
6404
6526
  * instance receives the result (the dispatching one may be gone), so the run is crash/scale-safe.
@@ -6481,8 +6603,28 @@ export class WorkflowEngine {
6481
6603
  return;
6482
6604
  if (await this.stepFanStillRunning(cp, result, queuedIn))
6483
6605
  return;
6606
+ if (!parseStepFanGroup(cp.parallelGroup))
6607
+ this.noteSettledHere(result.runId, [cp.seq]);
6484
6608
  this.resumeInBackground(result.runId, fanWakeOf(cp, result));
6485
6609
  }
6610
+ /**
6611
+ * Record `seqs` of `runId` in {@link settledHere}, dropping the entries older than a lease. A key is
6612
+ * re-inserted at the end, so the map stays in the order its keys were noted and pruning stops at
6613
+ * the first fresh entry.
6614
+ */
6615
+ noteSettledHere(runId, seqs) {
6616
+ const now = this.clock();
6617
+ for (const seq of seqs) {
6618
+ const key = `${runId}:${seq}`;
6619
+ this.settledHere.delete(key);
6620
+ this.settledHere.set(key, now);
6621
+ }
6622
+ for (const [key, at] of this.settledHere) {
6623
+ if (now - at <= this.leaseMs)
6624
+ break;
6625
+ this.settledHere.delete(key);
6626
+ }
6627
+ }
6486
6628
  resumeInBackground(runId, fan) {
6487
6629
  this.trackEffect(this.resumeImplicit(runId, fan).catch((err) => {
6488
6630
  console.warn(`[adonis-durable] background resume of run ${runId} failed (the run recovers on its next reconcile wake): ${err instanceof Error ? err.message : String(err)}`);
@@ -6546,20 +6688,11 @@ export class WorkflowEngine {
6546
6688
  resumable.push({ cp: read.value, result });
6547
6689
  });
6548
6690
  const finishedAt = new Date();
6549
- const writes = pending.map(({ index, cp }) => {
6550
- const result = results[index];
6551
- return {
6552
- ...cp,
6553
- status: result.status,
6554
- output: result.status === 'completed' ? result.output : cp.output,
6555
- error: result.error,
6556
- events: result.events ?? cp.events,
6557
- startedAt: result.startedAt ? new Date(result.startedAt) : cp.startedAt,
6558
- finishedAt,
6559
- };
6560
- });
6561
6691
  // Claimed, not blindly written: a duplicate delivery may have settled a step first.
6562
- const claims = await this.savePendingCheckpoints(writes);
6692
+ const claims = await this.settlePendingCheckpoints(pending.map(({ index, cp }) => ({
6693
+ cp,
6694
+ settlement: settlementOf(cp, results[index], finishedAt),
6695
+ })));
6563
6696
  for (const [k, claim] of claims.entries()) {
6564
6697
  const { index, cp } = pending[k];
6565
6698
  const result = results[index];
@@ -6605,9 +6738,14 @@ export class WorkflowEngine {
6605
6738
  if (fan)
6606
6739
  highest.set(fan.firstSeq, Math.max(highest.get(fan.firstSeq) ?? cp.seq, cp.seq));
6607
6740
  }
6741
+ // The one resume below replays every step of the batch.
6742
+ const singles = resumable
6743
+ .filter(({ cp }) => !parseStepFanGroup(cp.parallelGroup))
6744
+ .map(({ cp }) => cp.seq);
6608
6745
  for (const { cp, result } of resumable) {
6609
6746
  const fan = parseStepFanGroup(cp.parallelGroup);
6610
6747
  if (!fan || result.status === 'failed') {
6748
+ this.noteSettledHere(runId, singles);
6611
6749
  this.resumeInBackground(runId);
6612
6750
  return outcomes;
6613
6751
  }
@@ -6620,6 +6758,7 @@ export class WorkflowEngine {
6620
6758
  barriers.set(fan.firstSeq, closed);
6621
6759
  }
6622
6760
  if (!(await closed)) {
6761
+ this.noteSettledHere(runId, singles);
6623
6762
  this.resumeInBackground(runId, {
6624
6763
  first: fan.firstSeq,
6625
6764
  seq: highest.get(fan.firstSeq) ?? cp.seq,
@@ -6629,6 +6768,29 @@ export class WorkflowEngine {
6629
6768
  }
6630
6769
  return outcomes;
6631
6770
  }
6771
+ /**
6772
+ * Settle pending steps, each only while its row is still `pending`; per entry, whether it wrote
6773
+ * or the error that kept it from writing. Through {@link StateStore.settleCheckpointsIfPending}
6774
+ * when the store has it: one call, or one per entry when that call rejects (it then wrote
6775
+ * nothing). Else each step's whole row is written through {@link savePendingCheckpoints}.
6776
+ */
6777
+ async settlePendingCheckpoints(entries) {
6778
+ const settle = this.store.settleCheckpointsIfPending?.bind(this.store);
6779
+ if (!settle) {
6780
+ return this.savePendingCheckpoints(entries.map(({ cp, settlement }) => ({ ...cp, ...settlement })));
6781
+ }
6782
+ const settlements = entries.map((e) => e.settlement);
6783
+ try {
6784
+ const wrote = await settle(settlements);
6785
+ return settlements.map((_, i) => wrote[i] === true);
6786
+ }
6787
+ catch (error) {
6788
+ if (settlements.length === 1)
6789
+ return [{ error }];
6790
+ }
6791
+ const settled = await Promise.allSettled(settlements.map((s) => settle([s])));
6792
+ return settled.map((r) => r.status === 'fulfilled' ? r.value[0] === true : { error: r.reason });
6793
+ }
6632
6794
  /**
6633
6795
  * {@link savePendingCheckpoint} for checkpoints at distinct seqs: one
6634
6796
  * {@link StateStore.saveCheckpointsIfPending} call when the store has it, else one write each.
@@ -6665,12 +6827,35 @@ export class WorkflowEngine {
6665
6827
  return true;
6666
6828
  return (await this.countPendingSteps(result.runId, fan.firstSeq, lastSeq)) > 0;
6667
6829
  }
6830
+ /**
6831
+ * {@link StateStore.getCheckpointsForSettle}, when the store can use what it leaves out: only
6832
+ * with {@link StateStore.settleCheckpointsIfPending}, since a settle without it (and the salvage
6833
+ * of a late success onto a failed run) writes back the whole row it read.
6834
+ */
6835
+ settleReader() {
6836
+ return this.store.settleCheckpointsIfPending
6837
+ ? this.store.getCheckpointsForSettle?.bind(this.store)
6838
+ : undefined;
6839
+ }
6668
6840
  /**
6669
6841
  * The checkpoints of one run at `seqs`, each as its own outcome: one
6670
6842
  * {@link StateStore.getCheckpoints} call when the store has it, else one read each. The batch
6671
6843
  * call rejecting falls back to one read each, so a failing read fails only its own entry.
6672
6844
  */
6673
6845
  async readCheckpoints(runId, seqs) {
6846
+ const forSettle = this.settleReader();
6847
+ if (forSettle) {
6848
+ if (seqs.length > 1) {
6849
+ try {
6850
+ const bySeq = new Map((await forSettle(runId, seqs)).map((cp) => [cp.seq, cp]));
6851
+ return seqs.map((seq) => ({ status: 'fulfilled', value: bySeq.get(seq) ?? null }));
6852
+ }
6853
+ catch {
6854
+ // Fall through: one by one.
6855
+ }
6856
+ }
6857
+ return Promise.allSettled(seqs.map(async (seq) => (await forSettle(runId, [seq]))[0] ?? null));
6858
+ }
6674
6859
  if (seqs.length > 1 && this.store.getCheckpoints) {
6675
6860
  try {
6676
6861
  const bySeq = new Map((await this.store.getCheckpoints(runId, seqs)).map((cp) => [cp.seq, cp]));
@@ -6689,7 +6874,10 @@ export class WorkflowEngine {
6689
6874
  * late result for a terminal run — terminal is terminal).
6690
6875
  */
6691
6876
  async settleRemoteCheckpoint(result) {
6692
- const cp = await this.store.getCheckpoint(result.runId, result.seq);
6877
+ const forSettle = this.settleReader();
6878
+ const cp = forSettle
6879
+ ? ((await forSettle(result.runId, [result.seq]))[0] ?? null)
6880
+ : await this.store.getCheckpoint(result.runId, result.seq);
6693
6881
  if (!cp)
6694
6882
  return undefined;
6695
6883
  const run = await this.store.getRun(result.runId);
@@ -6732,31 +6920,25 @@ export class WorkflowEngine {
6732
6920
  if (run && (run.status === 'failed' || run.status === 'dead')) {
6733
6921
  await this.releaseQueueSlot(cp);
6734
6922
  if (result.status === 'completed') {
6735
- await this.store.saveCheckpoint({
6736
- ...cp,
6737
- status: 'completed',
6738
- output: result.output,
6739
- events: result.events ?? cp.events,
6740
- startedAt: result.startedAt ? new Date(result.startedAt) : cp.startedAt,
6741
- finishedAt: new Date(),
6742
- });
6923
+ const salvage = { ...settlementOf(cp, result, new Date()), error: cp.error };
6924
+ if (this.store.settleCheckpointsIfPending) {
6925
+ await this.store.settleCheckpointsIfPending([salvage]);
6926
+ }
6927
+ else {
6928
+ await this.store.saveCheckpoint({ ...cp, ...salvage });
6929
+ }
6743
6930
  }
6744
6931
  return undefined;
6745
6932
  }
6746
6933
  const finishedAt = new Date();
6747
- const startedAt = result.startedAt ? new Date(result.startedAt) : cp.startedAt;
6934
+ const settlement = settlementOf(cp, result, finishedAt);
6935
+ const startedAt = settlement.startedAt;
6748
6936
  // Claimed, not blindly written: another delivery settled it first, which is the redelivery case
6749
6937
  // above — re-drive the resume rather than settle twice.
6750
- const claimed = await this.savePendingCheckpoint({
6751
- ...cp,
6752
- status: result.status,
6753
- output: result.status === 'completed' ? result.output : cp.output,
6754
- error: result.error,
6755
- events: result.events ?? cp.events,
6756
- startedAt,
6757
- finishedAt,
6758
- });
6759
- if (!claimed)
6938
+ const [claim] = await this.settlePendingCheckpoints([{ cp, settlement }]);
6939
+ if (typeof claim === 'object')
6940
+ throw claim.error;
6941
+ if (!claim)
6760
6942
  return terminal ? undefined : cp;
6761
6943
  await this.releaseQueueSlot(cp);
6762
6944
  this.emit({
@@ -7041,6 +7223,19 @@ export class WorkflowEngine {
7041
7223
  function conventionGroup(name, namespace) {
7042
7224
  return tenantGroup(sanitizeQueueToken(name), namespace);
7043
7225
  }
7226
+ /** What `result` settles onto its step's pending checkpoint `cp`. */
7227
+ function settlementOf(cp, result, finishedAt) {
7228
+ return {
7229
+ runId: cp.runId,
7230
+ seq: cp.seq,
7231
+ status: result.status,
7232
+ output: result.status === 'completed' ? result.output : cp.output,
7233
+ error: result.error,
7234
+ events: result.events ?? cp.events,
7235
+ startedAt: result.startedAt ? new Date(result.startedAt) : cp.startedAt,
7236
+ finishedAt,
7237
+ };
7238
+ }
7044
7239
  /** Key of an inline remote turn's liveness reset in `heartbeatResets` (step keys are `runId:seq`). */
7045
7240
  function turnBeatKey(runId) {
7046
7241
  return `turn:${runId}`;