deepline 0.3.313 → 0.3.315

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 (109) hide show
  1. package/dist/bundling-sources/sdk/src/client.ts +12 -0
  2. package/dist/bundling-sources/sdk/src/release.ts +1 -1
  3. package/dist/bundling-sources/sdk/src/types.ts +10 -0
  4. package/dist/bundling-sources/shared_libs/play-runtime/play-run-recovery-policy.ts +50 -0
  5. package/dist/bundling-sources/shared_libs/play-runtime/run-event-query.ts +15 -2
  6. package/dist/bundling-sources/shared_libs/play-runtime/runner-backends/backends/daytona-launch-transport.ts +210 -2
  7. package/dist/bundling-sources/shared_libs/play-runtime/runner-backends/backends/daytona-lifecycle.ts +226 -14
  8. package/dist/bundling-sources/shared_libs/play-runtime/runner-backends/backends/daytona.ts +4 -18
  9. package/dist/bundling-sources/shared_libs/play-runtime/scheduler-backends/absurd-runtime-worker.ts +5 -0
  10. package/dist/bundling-sources/shared_libs/play-runtime/scheduler-backends/postgres.ts +94 -14
  11. package/dist/bundling-sources/shared_libs/plays/docflow-binding-owner.ts +61 -11
  12. package/dist/cli/{auth-LNA6WCQA.mjs → auth-OVWOO5NP.mjs} +4 -4
  13. package/dist/cli/auth-P2TTZU36.js +31 -0
  14. package/dist/cli/{billing-GJJUUCOZ.js → billing-SRZL7UOH.js} +70 -70
  15. package/dist/cli/{billing-D37QNIE3.mjs → billing-W3OXKWNE.mjs} +4 -4
  16. package/dist/cli/{chunk-HY2SSFRI.js → chunk-6BJACIVK.js} +1 -1
  17. package/dist/cli/{chunk-BS4GAUMH.mjs → chunk-AO4IWPP7.mjs} +33 -8
  18. package/dist/cli/{chunk-DOI3UBGV.mjs → chunk-AZWQJAQF.mjs} +1 -1
  19. package/dist/cli/{chunk-XFEPRIW7.mjs → chunk-B473QUQU.mjs} +4 -4
  20. package/dist/cli/{chunk-P4N7233B.js → chunk-C5WSTXUB.js} +247 -222
  21. package/dist/cli/{chunk-4UXNEC5C.js → chunk-FGAK6FET.js} +26 -26
  22. package/dist/cli/{chunk-3ZC4PHJ6.mjs → chunk-FTJQO757.mjs} +1 -1
  23. package/dist/cli/{chunk-UQBMMN4U.mjs → chunk-GK67FYMS.mjs} +5 -5
  24. package/dist/cli/{chunk-CSH4PKPJ.mjs → chunk-HNKHMHJ7.mjs} +1 -1
  25. package/dist/cli/{chunk-YZ5PP6TX.mjs → chunk-IGE7EJJY.mjs} +23 -3
  26. package/dist/cli/{chunk-HM2TSTCT.mjs → chunk-JF4O5OMH.mjs} +1 -1
  27. package/dist/cli/{chunk-N2DXWS6W.mjs → chunk-L5TSFCJ6.mjs} +2 -2
  28. package/dist/cli/{chunk-6G7ZYA6E.js → chunk-LOWURQMF.js} +50 -50
  29. package/dist/cli/{chunk-FJUIXWZF.js → chunk-OJP4FCCX.js} +48 -48
  30. package/dist/cli/{chunk-BLMO3JMZ.mjs → chunk-OUIIYFG3.mjs} +2 -2
  31. package/dist/cli/{chunk-MHETTUWR.js → chunk-OYJCNCUW.js} +2 -2
  32. package/dist/cli/{chunk-JG2ZTDZV.js → chunk-PPONZ6IE.js} +15 -15
  33. package/dist/cli/{chunk-QCXDLOGA.mjs → chunk-R3IFZDJL.mjs} +2 -2
  34. package/dist/cli/{chunk-KNGTN7NE.js → chunk-SNFSO2AN.js} +6 -6
  35. package/dist/cli/{chunk-RDUS2EDE.mjs → chunk-TQ3F77AC.mjs} +3 -3
  36. package/dist/cli/{chunk-I4FHFGMA.js → chunk-TQK662WG.js} +60 -60
  37. package/dist/cli/{chunk-OTA5C7DD.mjs → chunk-TZ5KUDPP.mjs} +1 -1
  38. package/dist/cli/{chunk-FSZTWQFK.js → chunk-URVZGYGG.js} +4 -4
  39. package/dist/cli/{chunk-VSAYSOVC.js → chunk-VYK23G2C.js} +222 -202
  40. package/dist/cli/{chunk-GQXEHX5X.js → chunk-Y42UERUY.js} +6 -6
  41. package/dist/cli/{chunk-QQMKX7JW.js → chunk-YL42TF53.js} +6 -6
  42. package/dist/cli/client-CXFPFCIG.js +20 -0
  43. package/dist/cli/{client-6RVRZ4OY.mjs → client-WQ3J5BSH.mjs} +3 -3
  44. package/dist/cli/{csv-SBOHBKEG.mjs → csv-36ARCBNQ.mjs} +3 -3
  45. package/dist/cli/{csv-2UHRG67O.js → csv-CENMEWDH.js} +12 -12
  46. package/dist/cli/{db-VRE7CLIQ.js → db-6DQZG6NB.js} +20 -20
  47. package/dist/cli/{db-TUFTPYEA.mjs → db-JU7XLKQM.mjs} +4 -4
  48. package/dist/cli/{describe-WCBZTSOF.mjs → describe-34ALEEFO.mjs} +8 -8
  49. package/dist/cli/{describe-GS5A2XH4.js → describe-VUD4AFBY.js} +15 -15
  50. package/dist/cli/{executions-ENMOOKWH.js → executions-H2UXS6FP.js} +6 -6
  51. package/dist/cli/{executions-UTN3O372.mjs → executions-H66BMN6N.mjs} +4 -4
  52. package/dist/cli/{feedback-PXBA3AG7.mjs → feedback-Q7LXXLSH.mjs} +3 -3
  53. package/dist/cli/{feedback-BLN6SIRG.js → feedback-YSYFQSAB.js} +6 -6
  54. package/dist/cli/git-SYFCYIFP.js +23 -0
  55. package/dist/cli/{git-YDMIRW3Q.mjs → git-XV43NSWN.mjs} +4 -4
  56. package/dist/cli/index.js +74 -74
  57. package/dist/cli/index.mjs +31 -31
  58. package/dist/cli/{legacy-noops-CWTUWTDB.js → legacy-noops-RGDW6SCA.js} +7 -7
  59. package/dist/cli/{legacy-noops-OVG2SICQ.mjs → legacy-noops-Z2DTZQN3.mjs} +4 -4
  60. package/dist/cli/{monitors-T63OPBTY.js → monitors-I75KZ2BP.js} +85 -85
  61. package/dist/cli/{monitors-ATIDXJA4.mjs → monitors-OFQ56O4S.mjs} +4 -4
  62. package/dist/cli/{org-7RVAPMRN.js → org-75AH6RIB.js} +60 -60
  63. package/dist/cli/{org-NXKNIM3H.mjs → org-ZYMCU6XO.mjs} +4 -4
  64. package/dist/cli/{play-B62OVHD4.mjs → play-BKXM76DV.mjs} +7 -7
  65. package/dist/cli/play-OKXUWPMO.js +152 -0
  66. package/dist/cli/{providers-U7B7JEYY.mjs → providers-GWG2WFIY.mjs} +4 -4
  67. package/dist/cli/{providers-EITVEGXD.js → providers-Z77MM7ID.js} +7 -7
  68. package/dist/cli/{quickstart-VSWTYNJM.mjs → quickstart-BGVIV374.mjs} +3 -3
  69. package/dist/cli/{quickstart-G64F2XRU.js → quickstart-I4RECIQC.js} +5 -5
  70. package/dist/cli/search-3KISUDDT.js +26 -0
  71. package/dist/cli/{search-SKYTFKKF.mjs → search-TIKBP3RE.mjs} +5 -5
  72. package/dist/cli/{secrets-JSFHSCGM.js → secrets-SNGRI4KS.js} +13 -13
  73. package/dist/cli/{secrets-4DNS7KWS.mjs → secrets-YCHKLWKC.mjs} +4 -4
  74. package/dist/cli/{sessions-HB7RQ4XN.mjs → sessions-IRDG6ZRX.mjs} +4 -4
  75. package/dist/cli/{sessions-MLTYPPCG.js → sessions-WXLADGDX.js} +5 -5
  76. package/dist/cli/{settings-3PCKNBPU.js → settings-72J4BVF6.js} +74 -74
  77. package/dist/cli/{settings-BT5WCU65.mjs → settings-KZALJDXW.mjs} +7 -7
  78. package/dist/cli/{setup-EIWDTMCK.mjs → setup-4KCA7ESV.mjs} +6 -6
  79. package/dist/cli/{setup-EYZWTDJI.js → setup-FFB3KLFD.js} +41 -41
  80. package/dist/cli/{skills-AQ3ORN2C.mjs → skills-MQZPYUQ3.mjs} +4 -4
  81. package/dist/cli/skills-XGOK4GRW.js +28 -0
  82. package/dist/cli/tools-LEAJ7EPW.js +38 -0
  83. package/dist/cli/{tools-OFTB2BGP.mjs → tools-M3WKRZ5T.mjs} +6 -6
  84. package/dist/cli/{update-ERQTFSAR.mjs → update-QJMNYGNF.mjs} +6 -6
  85. package/dist/cli/update-ZN6LDUHS.js +28 -0
  86. package/dist/cli/{workflow-6D6NHUH4.js → workflow-35UUOZ5E.js} +15 -15
  87. package/dist/cli/{workflow-SRLWGGBM.mjs → workflow-6T2EUFCN.mjs} +5 -5
  88. package/dist/{client-CtVdLZux.d.mts → client-CvpZsBqt.d.mts} +12 -0
  89. package/dist/{client-OBwghxPC.d.ts → client-Dh_4Ph1R.d.ts} +12 -0
  90. package/dist/index.d.mts +2 -2
  91. package/dist/index.d.ts +2 -2
  92. package/dist/index.js +6 -6
  93. package/dist/index.mjs +3 -3
  94. package/dist/install-integrity.json +76 -76
  95. package/dist/plays/run-request.d.mts +1 -1
  96. package/dist/plays/run-request.d.ts +1 -1
  97. package/dist/release.d.mts +1 -1
  98. package/dist/release.d.ts +1 -1
  99. package/dist/release.js +2 -2
  100. package/dist/release.mjs +1 -1
  101. package/package.json +1 -1
  102. package/dist/cli/auth-56NYRTQK.js +0 -31
  103. package/dist/cli/client-XUTMYJDL.js +0 -20
  104. package/dist/cli/git-26D3VD2I.js +0 -23
  105. package/dist/cli/play-WM3NZVE7.js +0 -152
  106. package/dist/cli/search-TXQJJQWY.js +0 -26
  107. package/dist/cli/skills-73T4BSDE.js +0 -28
  108. package/dist/cli/tools-ZGSOHWBY.js +0 -38
  109. package/dist/cli/update-L4V5TF7I.js +0 -28
@@ -1,5 +1,13 @@
1
1
  import type { CreateSandboxFromSnapshotParams, Sandbox } from '@daytonaio/sdk';
2
- import { DaytonaLaunchOutcomeUnknownError } from './daytona-launch-transport';
2
+ import {
3
+ boundedDaytonaProviderTextForLog,
4
+ type DaytonaLaunchOutcomeDiagnostic,
5
+ DaytonaLaunchOutcomeUnknownError,
6
+ daytonaErrorClassForLog,
7
+ daytonaErrorCodeForLog,
8
+ daytonaErrorStatusForLog,
9
+ daytonaSandboxStateForLog,
10
+ } from './daytona-launch-transport';
3
11
  import type {
4
12
  PlayRunnerExecutionConfig,
5
13
  PlayRunnerResult,
@@ -110,7 +118,15 @@ export class DaytonaSandboxAcquisitionUnavailableError extends Error {
110
118
  * creating another sandbox could execute the Play twice.
111
119
  */
112
120
  export class DaytonaSandboxTransitionPendingError extends Error {
113
- constructor(readonly sandboxName: string) {
121
+ constructor(
122
+ readonly sandboxName: string,
123
+ /**
124
+ * What Daytona reported when this decision was made. Log-only: a 404, a
125
+ * lookup error, and every non-started state all reach this same error, so
126
+ * the scheduler logs this to tell them apart. It never changes recovery.
127
+ */
128
+ readonly recoveryLogDetail: DaytonaSandboxTransitionPendingLogDetail | null = null,
129
+ ) {
114
130
  super(
115
131
  `Daytona sandbox ${sandboxName} is still transitioning after create timed out.`,
116
132
  );
@@ -118,6 +134,140 @@ export class DaytonaSandboxTransitionPendingError extends Error {
118
134
  }
119
135
  }
120
136
 
137
+ /** One provider lookup of the named sandbox, reduced to bounded log facts. */
138
+ type DaytonaNamedSandboxObservation = {
139
+ lookup: 'found' | 'not_found' | 'error' | 'unsupported';
140
+ sandboxState: string | null;
141
+ /** Not `sandboxId`: that key means an acquired sandbox to lifecycle tooling. */
142
+ observedSandboxId: string | null;
143
+ providerErrorReason: string | null;
144
+ lookupErrorClass: string | null;
145
+ lookupErrno: string | null;
146
+ lookupErrorStatus: number | null;
147
+ };
148
+
149
+ /**
150
+ * Flat, bounded scalars only: enumerated states and classes, one provider
151
+ * sandbox id, counts, durations, and one capped provider text field.
152
+ */
153
+ export type DaytonaSandboxTransitionPendingLogDetail =
154
+ DaytonaNamedSandboxObservation & {
155
+ phase: 'create_reconcile' | 'resume';
156
+ /** Names the case; all of them map to the same recovery decision. */
157
+ reasonDetail: string;
158
+ providerAttempt: number;
159
+ /** Lookups issued by this decision's reconcile ladder (at most 7). */
160
+ lookups: number;
161
+ lookupErrors: number;
162
+ deleteConflicts: number;
163
+ elapsedMs: number;
164
+ /** Known only in the process that sent the create. */
165
+ createResponseReceived: boolean | null;
166
+ createLastObservedState: string | null;
167
+ };
168
+
169
+ const NO_DAYTONA_NAMED_SANDBOX_LOOKUP: DaytonaNamedSandboxObservation = {
170
+ lookup: 'unsupported',
171
+ sandboxState: null,
172
+ observedSandboxId: null,
173
+ providerErrorReason: null,
174
+ lookupErrorClass: null,
175
+ lookupErrno: null,
176
+ lookupErrorStatus: null,
177
+ };
178
+
179
+ // Both observers are total: they run between a provider lookup and the
180
+ // decision it feeds, so a failure to describe the lookup must not surface.
181
+ function observeDaytonaNamedSandbox(
182
+ sandbox: DaytonaSandbox,
183
+ ): DaytonaNamedSandboxObservation {
184
+ try {
185
+ const record = sandbox as { id?: unknown; errorReason?: unknown };
186
+ return {
187
+ ...NO_DAYTONA_NAMED_SANDBOX_LOOKUP,
188
+ lookup: 'found',
189
+ sandboxState: daytonaSandboxStateForLog(sandbox.state) ?? 'missing',
190
+ observedSandboxId:
191
+ typeof record.id === 'string' && record.id
192
+ ? record.id.slice(0, 64)
193
+ : null,
194
+ providerErrorReason: boundedDaytonaProviderTextForLog(record.errorReason),
195
+ };
196
+ } catch {
197
+ return {
198
+ ...NO_DAYTONA_NAMED_SANDBOX_LOOKUP,
199
+ lookup: 'found',
200
+ sandboxState: 'unreadable',
201
+ };
202
+ }
203
+ }
204
+
205
+ function observeDaytonaNamedSandboxLookupFailure(
206
+ error: unknown,
207
+ ): DaytonaNamedSandboxObservation {
208
+ try {
209
+ return {
210
+ ...NO_DAYTONA_NAMED_SANDBOX_LOOKUP,
211
+ lookup: isDaytonaSandboxNotFound(error) ? 'not_found' : 'error',
212
+ lookupErrorClass: daytonaErrorClassForLog(error),
213
+ lookupErrno: daytonaErrorCodeForLog(error),
214
+ lookupErrorStatus: daytonaErrorStatusForLog(error),
215
+ };
216
+ } catch {
217
+ return { ...NO_DAYTONA_NAMED_SANDBOX_LOOKUP, lookup: 'error' };
218
+ }
219
+ }
220
+
221
+ function daytonaTransitionPendingReasonDetail(
222
+ phase: DaytonaSandboxTransitionPendingLogDetail['phase'],
223
+ observation: DaytonaNamedSandboxObservation,
224
+ deleteConflicts: number,
225
+ ): string {
226
+ if (observation.lookup !== 'found') {
227
+ return `${phase}:lookup_${observation.lookup}`;
228
+ }
229
+ // Reconcile reaches a pending decision with a visible sandbox only after
230
+ // every delete of it reported an in-progress state change.
231
+ return deleteConflicts > 0
232
+ ? `${phase}:delete_conflict:${observation.sandboxState}`
233
+ : `${phase}:sandbox_state:${observation.sandboxState}`;
234
+ }
235
+
236
+ /**
237
+ * The log detail for one pending decision. Total by construction: failing to
238
+ * describe a decision must never replace the decision.
239
+ */
240
+ function daytonaTransitionPendingLogDetail(input: {
241
+ phase: DaytonaSandboxTransitionPendingLogDetail['phase'];
242
+ observed: TimedOutDaytonaSandboxReconcileObserved;
243
+ providerAttempt: number;
244
+ launchOutcome: DaytonaLaunchOutcomeDiagnostic | null;
245
+ }): DaytonaSandboxTransitionPendingLogDetail | null {
246
+ try {
247
+ const { last, lookups, lookupErrors, deleteConflicts, elapsedMs } =
248
+ input.observed;
249
+ return {
250
+ ...last,
251
+ phase: input.phase,
252
+ reasonDetail: daytonaTransitionPendingReasonDetail(
253
+ input.phase,
254
+ last,
255
+ deleteConflicts,
256
+ ),
257
+ providerAttempt: input.providerAttempt,
258
+ lookups,
259
+ lookupErrors,
260
+ deleteConflicts,
261
+ elapsedMs,
262
+ createResponseReceived:
263
+ input.launchOutcome?.createResponseReceived ?? null,
264
+ createLastObservedState: input.launchOutcome?.lastObservedState ?? null,
265
+ };
266
+ } catch {
267
+ return null;
268
+ }
269
+ }
270
+
121
271
  export function resolveDaytonaSandboxAcquisitionUnavailableReason(
122
272
  errors: readonly string[],
123
273
  ): DaytonaSandboxAcquisitionUnavailableReason | null {
@@ -369,11 +519,21 @@ async function createOneShotDaytonaSandbox(input: {
369
519
  });
370
520
  }
371
521
 
372
- type TimedOutDaytonaSandboxReconciliation =
522
+ /** Log-only tally of one reconcile ladder; it never selects the result. */
523
+ type TimedOutDaytonaSandboxReconcileObserved = {
524
+ last: DaytonaNamedSandboxObservation;
525
+ lookups: number;
526
+ lookupErrors: number;
527
+ deleteConflicts: number;
528
+ elapsedMs: number;
529
+ };
530
+
531
+ type TimedOutDaytonaSandboxReconciliation = (
373
532
  | { kind: 'started'; sandbox: DaytonaSandbox }
374
533
  | { kind: 'deleted' }
375
534
  | { kind: 'not_found' }
376
- | { kind: 'transition_pending' };
535
+ | { kind: 'transition_pending' }
536
+ ) & { observed: TimedOutDaytonaSandboxReconcileObserved };
377
537
 
378
538
  async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
379
539
  daytona: DaytonaClient;
@@ -386,7 +546,19 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
386
546
  runId?: string;
387
547
  };
388
548
  }): Promise<TimedOutDaytonaSandboxReconciliation> {
389
- if (!input.daytona.get) return { kind: 'not_found' };
549
+ const reconcileStartedAt = Date.now();
550
+ let last = NO_DAYTONA_NAMED_SANDBOX_LOOKUP;
551
+ let lookups = 0;
552
+ let lookupErrors = 0;
553
+ let deleteConflicts = 0;
554
+ const observed = (): TimedOutDaytonaSandboxReconcileObserved => ({
555
+ last,
556
+ lookups,
557
+ lookupErrors,
558
+ deleteConflicts,
559
+ elapsedMs: Date.now() - reconcileStartedAt,
560
+ });
561
+ if (!input.daytona.get) return { kind: 'not_found', observed: observed() };
390
562
  let lastDeleteConflict: unknown = null;
391
563
  for (const delayMs of DAYTONA_TIMED_OUT_CREATE_RECONCILE_DELAYS_MS) {
392
564
  input.cancellationSignal?.throwIfAborted();
@@ -394,9 +566,11 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
394
566
  await waitForAcquisitionRetry(delayMs, input.cancellationSignal);
395
567
  }
396
568
  let sandbox: DaytonaSandbox;
569
+ lookups += 1;
397
570
  try {
398
571
  sandbox = await input.daytona.get(input.sandboxName);
399
572
  } catch (error) {
573
+ last = observeDaytonaNamedSandboxLookupFailure(error);
400
574
  const providerStatus =
401
575
  error && typeof error === 'object' && 'statusCode' in error
402
576
  ? error.statusCode
@@ -410,7 +584,10 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
410
584
  // A current absence is not proof that the sent create cannot commit.
411
585
  // Return it to the durable operation owner without another create or
412
586
  // releasing capacity; later recovery re-reads the same identity.
413
- if (isDaytonaSandboxNotFound(error)) return { kind: 'not_found' };
587
+ if (isDaytonaSandboxNotFound(error)) {
588
+ return { kind: 'not_found', observed: observed() };
589
+ }
590
+ lookupErrors += 1;
414
591
  continue;
415
592
  }
416
593
  input.emitStage?.('create:reconcile_state', {
@@ -423,8 +600,9 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
423
600
  // is safer and faster than attempting a conflicting delete or creating a
424
601
  // second sandbox.
425
602
  if (sandbox.state === 'started') {
426
- return { kind: 'started', sandbox };
603
+ return { kind: 'started', sandbox, observed: observed() };
427
604
  }
605
+ last = observeDaytonaNamedSandbox(sandbox);
428
606
  if (input.operationIdentity) {
429
607
  const expected = input.operationIdentity;
430
608
  const configuredOrganizationId =
@@ -455,6 +633,7 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
455
633
  // sandbox while the first one is still becoming ready.
456
634
  if (isDaytonaSandboxStateChangeConflict(error)) {
457
635
  lastDeleteConflict = error;
636
+ deleteConflicts += 1;
458
637
  continue;
459
638
  }
460
639
  throw new Error(
@@ -462,12 +641,12 @@ async function reconcileAndDeleteTimedOutDaytonaSandbox(input: {
462
641
  { cause: error },
463
642
  );
464
643
  }
465
- return { kind: 'deleted' };
644
+ return { kind: 'deleted', observed: observed() };
466
645
  }
467
646
  if (lastDeleteConflict) {
468
- return { kind: 'transition_pending' };
647
+ return { kind: 'transition_pending', observed: observed() };
469
648
  }
470
- return { kind: 'not_found' };
649
+ return { kind: 'not_found', observed: observed() };
471
650
  }
472
651
 
473
652
  function daytonaSandboxName(providerOperationToken?: string | null): string {
@@ -553,8 +732,17 @@ async function createRetriedOneShotDaytonaSandbox(input: {
553
732
  );
554
733
  }
555
734
  // Neither absence nor an ongoing transition proves that create cannot
556
- // commit. Retain the durable identity and its capacity reservation.
557
- throw new DaytonaSandboxTransitionPendingError(sandboxName);
735
+ // commit. Retain the durable identity and its capacity reservation. The
736
+ // detail only records which of the two Daytona reported, for the log.
737
+ throw new DaytonaSandboxTransitionPendingError(
738
+ sandboxName,
739
+ daytonaTransitionPendingLogDetail({
740
+ phase: 'resume',
741
+ observed: reconciliation.observed,
742
+ providerAttempt: attempt,
743
+ launchOutcome: null,
744
+ }),
745
+ );
558
746
  }
559
747
  let sandbox: DaytonaSandbox;
560
748
  try {
@@ -579,17 +767,29 @@ async function createRetriedOneShotDaytonaSandbox(input: {
579
767
  errorClass: daytonaCreateErrorClass(error),
580
768
  });
581
769
  errors.push(message);
770
+ // Log-only: whether the create request returned before the deadline and
771
+ // what Daytona last reported, so an unknown outcome is attributable.
772
+ const launchOutcome =
773
+ error instanceof DaytonaLaunchOutcomeUnknownError
774
+ ? error.diagnostic
775
+ : null;
776
+ const createFailureLogDetail = {
777
+ createErrorClass: daytonaCreateErrorClass(error),
778
+ ...(launchOutcome ?? {}),
779
+ };
582
780
  input.emitStage('create:attempt_failed', {
583
781
  attempt,
584
782
  elapsedMs: Date.now() - input.startedAt,
585
783
  attemptElapsedMs: Date.now() - attemptStartedAt,
586
784
  error: message,
785
+ ...createFailureLogDetail,
587
786
  });
588
787
  console.warn('[play-runner.daytona.create_attempt_failed]', {
589
788
  workflowId: input.context.workflowId ?? null,
590
789
  runId: input.context.runId ?? null,
591
790
  attempt,
592
791
  error: message,
792
+ ...createFailureLogDetail,
593
793
  });
594
794
  const immediateFallbackReason =
595
795
  resolveDaytonaSandboxAcquisitionUnavailableReason([message]);
@@ -640,9 +840,21 @@ async function createRetriedOneShotDaytonaSandbox(input: {
640
840
  };
641
841
  }
642
842
  await recordCreateFailure();
843
+ // Same decision as before; the detail records what the ladder saw.
844
+ const reconcileObserved = reconciliation.observed;
845
+ const reconcilePending = (): DaytonaSandboxTransitionPendingError =>
846
+ new DaytonaSandboxTransitionPendingError(
847
+ sandboxName,
848
+ daytonaTransitionPendingLogDetail({
849
+ phase: 'create_reconcile',
850
+ observed: reconcileObserved,
851
+ providerAttempt: attempt,
852
+ launchOutcome,
853
+ }),
854
+ );
643
855
  if (reconciliation.kind === 'transition_pending') {
644
856
  if (reservation?.providerOperationToken) {
645
- throw new DaytonaSandboxTransitionPendingError(sandboxName);
857
+ throw reconcilePending();
646
858
  }
647
859
  throw new Error(
648
860
  `Daytona timed-out sandbox ${sandboxName} remained in a state transition across the bounded cleanup window. Modal fallback suppressed.`,
@@ -663,7 +875,7 @@ async function createRetriedOneShotDaytonaSandbox(input: {
663
875
  );
664
876
  }
665
877
  if (reservation?.providerOperationToken) {
666
- throw new DaytonaSandboxTransitionPendingError(sandboxName);
878
+ throw reconcilePending();
667
879
  }
668
880
  throw new Error(
669
881
  `${message} Daytona did not return a cleanup-addressable sandbox identity; Modal fallback suppressed. The named ephemeral sandbox retains Daytona's auto-delete-on-stop backstop.`,
@@ -1,4 +1,7 @@
1
- import { createDaytonaLaunchClient } from './daytona-launch-transport';
1
+ import {
2
+ createDaytonaLaunchClient,
3
+ redactDaytonaErrorDetail,
4
+ } from './daytona-launch-transport';
2
5
  import type { DaytonaClientOptions } from '../../daytona-runtime-config';
3
6
  import {
4
7
  createDaytonaControlClient,
@@ -328,23 +331,6 @@ function numberProperty(value: unknown, key: string): number | null {
328
331
  : null;
329
332
  }
330
333
 
331
- function redactDaytonaErrorDetail(value: string): string {
332
- return value
333
- .replace(
334
- /authorization\s*[=:]\s*bearer\s+[^\s,;]+/gi,
335
- 'Authorization=Bearer [redacted]',
336
- )
337
- .replace(
338
- /(authorization\s*[=:]\s*)(?!bearer\s+\[redacted\])[^\s,;]+/gi,
339
- '$1[redacted]',
340
- )
341
- .replace(/(bearer\s+)[^\s,;]+/gi, '$1[redacted]')
342
- .replace(/((?:api[_-]?key|token)\s*[=:]\s*)[^\s,;]+/gi, '$1[redacted]')
343
- .replace(/\s+/g, ' ')
344
- .trim()
345
- .slice(0, 1_000);
346
- }
347
-
348
334
  /**
349
335
  * Preserve the actionable shape of a Daytona control-plane failure without
350
336
  * copying credentials or an arbitrarily large upstream response into logs.
@@ -48,6 +48,7 @@ import { RUNTIME_RELIABILITY_POLICY } from '../runtime-reliability-policy';
48
48
  import type { RuntimeTrafficPolicy } from '../runtime-traffic-policy';
49
49
  import { runtimeLimitExceededCause } from '../run-failure';
50
50
  import {
51
+ playRunRecoveryLogDetail,
51
52
  recoveryForDaytonaSandboxAcquisitionUnavailable,
52
53
  recoveryForRuntimeInitializationTransportFailure,
53
54
  recoveryForRuntimeSchedulerAdmissionFailure,
@@ -4359,6 +4360,9 @@ async function executeAssignedAbsurdAttempt(input: {
4359
4360
  }),
4360
4361
  );
4361
4362
  if (recovery.kind === 'deferred') {
4363
+ // Log-only. The persisted reason groups several provider conditions
4364
+ // into one decision; the detail names the one that was observed.
4365
+ const detail = playRunRecoveryLogDetail(recoveryError);
4362
4366
  console.warn('[absurd-runtime.worker] pre_code_recovery_deferred', {
4363
4367
  runId,
4364
4368
  generation,
@@ -4367,6 +4371,7 @@ async function executeAssignedAbsurdAttempt(input: {
4367
4371
  deferCount: recovery.deferCount,
4368
4372
  deadlineAtMs: recovery.deadlineAtMs,
4369
4373
  delaySeconds: recovery.delaySeconds,
4374
+ ...(detail ? { detail } : {}),
4370
4375
  });
4371
4376
  throw new DeferTask(
4372
4377
  `Platform recovery deferred for ${recovery.dependency}: ${recovery.reason}`,
@@ -15682,6 +15682,34 @@ export async function getPostgresSchedulerObservations(
15682
15682
  return result.rows.map(mapObservationRow);
15683
15683
  }
15684
15684
 
15685
+ /** Read the newest settled observation for one node across a bounded page of runs. */
15686
+ export async function getPostgresSchedulerObservationsForRuns(
15687
+ client: PostgresSchedulerQueryClient,
15688
+ input: { runIds: readonly string[]; nodeId: string },
15689
+ options: PostgresSchedulerReceiptOptions,
15690
+ ): Promise<DocflowObservationRecord[]> {
15691
+ const runIds = [
15692
+ ...new Set(input.runIds.map((runId) => runId.trim()).filter(Boolean)),
15693
+ ];
15694
+ if (runIds.length === 0) return [];
15695
+ const result = await client.query<ObservationRow>(
15696
+ `
15697
+ SELECT DISTINCT ON (run_id)
15698
+ run_id, attempt, node_id, invocation_seq, invocation_id, status,
15699
+ inputs_json, outputs_json, inputs_truncated, outputs_truncated,
15700
+ error, started_at, settled_at
15701
+ FROM ${tableName('observations', options)}
15702
+ WHERE org_id = $1
15703
+ AND run_id = ANY($2::text[])
15704
+ AND node_id = $3
15705
+ AND status IN ('completed', 'failed')
15706
+ ORDER BY run_id, attempt DESC, invocation_seq DESC
15707
+ `,
15708
+ [receiptOrgId(options), runIds, input.nodeId],
15709
+ );
15710
+ return result.rows.map(mapObservationRow);
15711
+ }
15712
+
15685
15713
  export async function countPostgresSchedulerReceipts(
15686
15714
  client: PostgresSchedulerQueryClient,
15687
15715
  input: {
@@ -18151,6 +18179,20 @@ async function claimPostgresSchedulerOutboxPrefix(
18151
18179
  : supportsTerminalBatchMarker && input.allowIngressPrefix === true
18152
18180
  ? `(head.event_type = ${RUN_SCHEDULER_OUTBOX_EVENT_TYPE_SQL_LITERAL.runEvents})`
18153
18181
  : 'false';
18182
+ const findSearchBlockersOnce = seekRunHead && input.searchOnly === true;
18183
+ const blockingPredecessorSql = (alias: string) => `(
18184
+ (${alias}.event_type = ${RUN_SCHEDULER_OUTBOX_EVENT_TYPE_SQL_LITERAL.runEvents}
18185
+ AND NOT (${ingressPrefixSql}))
18186
+ OR (${supportsTerminalBatchMarker && input.allowIngressPrefix === true ? `head.delivery_batch_started_at IS NOT NULL AND ${alias}.delivery_batch_started_at IS NULL` : 'false'})
18187
+ OR ${alias}.failed_at IS NOT NULL
18188
+ OR coalesce(${alias}.next_attempt_at, ${alias}.created_at) > now()
18189
+ OR (
18190
+ $3::text IS NOT NULL
18191
+ AND ${alias}.lease_expires_at IS NOT NULL
18192
+ AND ${alias}.lease_expires_at > now()
18193
+ AND ${alias}.lease_owner IS DISTINCT FROM $3::text
18194
+ )
18195
+ )`;
18154
18196
  return await client.query<{
18155
18197
  row_kind: 'claimed' | 'delivered';
18156
18198
  event_id: string;
@@ -18193,6 +18235,7 @@ async function claimPostgresSchedulerOutboxPrefix(
18193
18235
  : ''
18194
18236
  } head AS MATERIALIZED (
18195
18237
  SELECT candidate.event_id, candidate.run_id, candidate.event_type,
18238
+ ${findSearchBlockersOnce ? 'candidate.created_at,' : ''}
18196
18239
  ${supportsTerminalBatchMarker ? 'candidate.delivery_batch_started_at' : 'NULL::timestamptz'} AS delivery_batch_started_at
18197
18240
  FROM ${tableName('outbox', options)} AS candidate
18198
18241
  ${seekRunHead ? 'JOIN pending_head ON pending_head.event_id = candidate.event_id' : ''}
@@ -18274,7 +18317,41 @@ async function claimPostgresSchedulerOutboxPrefix(
18274
18317
  }
18275
18318
  FOR UPDATE SKIP LOCKED
18276
18319
  LIMIT 1
18277
- ), candidates AS MATERIALIZED (
18320
+ ), ${
18321
+ findSearchBlockersOnce
18322
+ ? `first_blocker AS MATERIALIZED (
18323
+ SELECT predecessor.event_type, predecessor.created_at, predecessor.event_id
18324
+ FROM ${tableName('outbox', options)} AS predecessor
18325
+ JOIN head ON head.run_id = predecessor.run_id
18326
+ WHERE predecessor.run_id = $2::text
18327
+ AND ${pending('predecessor')}
18328
+ AND ${searchLaneSql('predecessor')} = true
18329
+ -- The oldest pending head passed eligibility above. Only later
18330
+ -- facts can introduce an ordinary blocker for this prefix.
18331
+ AND ${runLedgerPositionSql('predecessor')} > (
18332
+ (SELECT CASE head.event_type WHEN ${RUN_EVENT_KIND_SQL_LITERAL.runCreated} THEN 0 WHEN ${RUN_EVENT_KIND_SQL_LITERAL.runStarted} THEN 1 ELSE 2 END FROM head),
18333
+ (SELECT head.created_at FROM head),
18334
+ (SELECT head.event_id FROM head)
18335
+ )
18336
+ AND ${blockingPredecessorSql('predecessor')}
18337
+ ORDER BY ${runLedgerOrderSql('predecessor')}
18338
+ LIMIT 1
18339
+ ), first_started_terminal AS MATERIALIZED (
18340
+ SELECT terminal_predecessor.event_type, terminal_predecessor.created_at, terminal_predecessor.event_id
18341
+ FROM ${tableName('outbox', options)} AS terminal_predecessor
18342
+ JOIN head ON head.run_id = terminal_predecessor.run_id
18343
+ WHERE terminal_predecessor.run_id = $2::text
18344
+ AND ${pending('terminal_predecessor')}
18345
+ AND ${searchLaneSql('terminal_predecessor')} = true
18346
+ AND ${supportsTerminalBatchMarker ? 'terminal_predecessor.delivery_batch_started_at IS NOT NULL' : 'false'}
18347
+ AND terminal_predecessor.event_type IN (
18348
+ ${RUN_EVENT_KIND_SQL_LITERAL.runCompleted}, ${RUN_EVENT_KIND_SQL_LITERAL.runFailed}, ${RUN_EVENT_KIND_SQL_LITERAL.runCancelled}
18349
+ )
18350
+ ORDER BY ${runLedgerOrderSql('terminal_predecessor')}
18351
+ LIMIT 1
18352
+ ),`
18353
+ : ''
18354
+ }candidates AS MATERIALIZED (
18278
18355
  SELECT outbox.event_id
18279
18356
  FROM ${tableName('outbox', options)} AS outbox
18280
18357
  JOIN head ON head.run_id = outbox.run_id
@@ -18313,7 +18390,20 @@ async function claimPostgresSchedulerOutboxPrefix(
18313
18390
  -- that the replay included a new event. A terminal which has not
18314
18391
  -- started a batch yet remains an ordinary causal event and can be
18315
18392
  -- claimed with its existing contiguous prefix.
18393
+ ${
18394
+ findSearchBlockersOnce
18395
+ ? `-- Each cutoff reads this run's blockers once, even when stale
18396
+ -- statistics make a broad pending scan look cheaper than a seek.
18316
18397
  AND NOT EXISTS (
18398
+ SELECT 1 FROM first_started_terminal AS blocker
18399
+ WHERE ${supportsTerminalBatchMarker ? 'outbox.delivery_batch_started_at IS NULL' : 'true'}
18400
+ AND ${runLedgerPositionSql('blocker')} < ${runLedgerPositionSql('outbox')}
18401
+ )
18402
+ AND NOT EXISTS (
18403
+ SELECT 1 FROM first_blocker AS blocker
18404
+ WHERE ${runLedgerPositionSql('blocker')} < ${runLedgerPositionSql('outbox')}
18405
+ )`
18406
+ : ` AND NOT EXISTS (
18317
18407
  SELECT 1
18318
18408
  FROM ${tableName('outbox', options)} AS terminal_predecessor
18319
18409
  WHERE terminal_predecessor.run_id = outbox.run_id
@@ -18341,20 +18431,10 @@ async function claimPostgresSchedulerOutboxPrefix(
18341
18431
  WHERE predecessor.run_id = outbox.run_id
18342
18432
  AND ${pending('predecessor')}
18343
18433
  AND ${runLedgerPrecedesSql('predecessor', 'outbox', false, input.searchOnly === true)}
18344
- AND (
18345
- (predecessor.event_type = ${RUN_SCHEDULER_OUTBOX_EVENT_TYPE_SQL_LITERAL.runEvents}
18346
- AND NOT (${ingressPrefixSql}))
18347
- OR (${supportsTerminalBatchMarker && input.allowIngressPrefix === true ? 'head.delivery_batch_started_at IS NOT NULL AND predecessor.delivery_batch_started_at IS NULL' : 'false'})
18348
- OR predecessor.failed_at IS NOT NULL
18349
- OR coalesce(predecessor.next_attempt_at, predecessor.created_at) > now()
18350
- OR (
18351
- $3::text IS NOT NULL
18352
- AND predecessor.lease_expires_at IS NOT NULL
18353
- AND predecessor.lease_expires_at > now()
18354
- AND predecessor.lease_owner IS DISTINCT FROM $3::text
18355
- )
18356
- )
18434
+ AND ${blockingPredecessorSql('predecessor')}
18357
18435
  )
18436
+ `
18437
+ }
18358
18438
  ORDER BY ${runLedgerOrderSql('outbox')}
18359
18439
  FOR UPDATE OF outbox
18360
18440
  LIMIT $1