@ewhauser/world-celld 0.3.0 → 0.3.2

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.
package/dist/testing.js CHANGED
@@ -1,4 +1,4 @@
1
- import { a as cleanupTombstone, c as globalRunIndexKey, d as workflowRunIndexKey, f as compact, h as parse, i as TOMBSTONE_KEY, l as hookMarkerKey, m as rpcStringify, n as HOOK_MARKER_PREFIX, o as correlationIndexKey, p as rpcParse, r as INDEX_MARKER_PREFIX, s as expiredRead, t as CLEANUP_RECORD_KEY, u as indexMarkerKey } from "./retention-bLY1vcOu.js";
1
+ import { A as validateStreamReadRequest, E as encodeStreamReadResult, O as encodeStreamWriteResult, S as STREAM_BATCH_CONTENT_TYPE, c as HOOK_CLAIM_LEASE_MS, d as MAX_QUEUE_DELIVERY_TIMEOUT_MS, f as MAX_RUN_INDEX_PUBLICATION_LIFETIME_MS, g as MAX_STREAM_CHUNK_BYTES, h as MAX_STREAM_BATCH_BYTES, i as parse, j as validateStreamWriteChunks, k as normalizeStreamError, l as LIFECYCLE_COMPACTION_RETRY_MS, m as QUEUE_INFLIGHT_GRACE_MS, n as rpcParse, o as createWorkflowIndex, p as QUEUE_FENCE_GRACE_MS, r as rpcStringify, t as compact, v as MAX_STREAM_LONG_POLL_MS, w as decodeStreamWriteBatch, y as MAX_STREAM_READ_BYTES } from "./util-ya3V_FTA.js";
2
2
  import { EventSchema, HookSchema, SPEC_VERSION_CURRENT, StepSchema, WaitSchema, WorkflowRunSchema, applyAttributeChanges, eventIdToSlot, getMaxEventsPerRun, isChildEntityCreationEvent, isTerminalStepStatus, isTerminalWorkflowRunStatus, slotToEventId, validateAttributeChanges } from "@workflow/world";
3
3
  import http from "node:http";
4
4
  //#region src/testing/fake-cell.ts
@@ -33,40 +33,109 @@ var FakeStorage = class {
33
33
  clock;
34
34
  data = /* @__PURE__ */ new Map();
35
35
  alarmAt = null;
36
+ getManyCalls = [];
37
+ operationCounts = {
38
+ get: 0,
39
+ getMany: 0,
40
+ list: 0,
41
+ put: 0,
42
+ putMany: 0,
43
+ delete: 0,
44
+ deleteMany: 0,
45
+ transaction: 0
46
+ };
47
+ transactionTail = Promise.resolve();
48
+ transactionDelayMs = 0;
49
+ /** Deterministic query-shape instrumentation for storage scaling tests. */
50
+ listCalls = [];
51
+ operationCalls = [];
36
52
  mutationFailure;
37
53
  constructor(clock = () => Date.now()) {
38
54
  this.clock = clock;
39
55
  }
40
- async get(key) {
41
- return this.data.get(key);
42
- }
43
- async put(key, value) {
44
- this.maybeFailMutation({
56
+ async get(keyOrKeys) {
57
+ if (Array.isArray(keyOrKeys)) {
58
+ this.operationCounts.getMany += 1;
59
+ this.getManyCalls.push([...keyOrKeys]);
60
+ this.recordStorageCall("get", keyOrKeys, false);
61
+ return new Map(keyOrKeys.filter((key) => this.data.has(key)).map((key) => [key, this.data.get(key)]));
62
+ }
63
+ this.operationCounts.get += 1;
64
+ this.recordStorageCall("get", [keyOrKeys], false);
65
+ return this.data.get(keyOrKeys);
66
+ }
67
+ async put(keyOrEntries, value) {
68
+ if (typeof keyOrEntries === "string") {
69
+ this.operationCounts.put += 1;
70
+ this.maybeFailMutation({
71
+ operation: "put",
72
+ key: keyOrEntries,
73
+ value
74
+ });
75
+ this.recordStorageCall("put", [keyOrEntries], false);
76
+ this.data.set(keyOrEntries, structuredClone(value));
77
+ return;
78
+ }
79
+ this.operationCounts.putMany += 1;
80
+ for (const [key, entryValue] of Object.entries(keyOrEntries)) this.maybeFailMutation({
45
81
  operation: "put",
46
82
  key,
47
- value
83
+ value: entryValue
48
84
  });
49
- this.data.set(key, structuredClone(value));
50
- }
51
- async delete(key) {
52
- this.maybeFailMutation({
85
+ this.recordStorageCall("put", Object.keys(keyOrEntries), false);
86
+ for (const [key, entryValue] of Object.entries(keyOrEntries)) this.data.set(key, structuredClone(entryValue));
87
+ }
88
+ async delete(keyOrKeys) {
89
+ if (typeof keyOrKeys === "string") {
90
+ this.operationCounts.delete += 1;
91
+ this.maybeFailMutation({
92
+ operation: "delete",
93
+ key: keyOrKeys
94
+ });
95
+ this.recordStorageCall("delete", [keyOrKeys], false);
96
+ return this.data.delete(keyOrKeys);
97
+ }
98
+ this.operationCounts.deleteMany += 1;
99
+ for (const key of keyOrKeys) this.maybeFailMutation({
53
100
  operation: "delete",
54
101
  key
55
102
  });
56
- return this.data.delete(key);
103
+ this.recordStorageCall("delete", keyOrKeys, false);
104
+ let deleted = 0;
105
+ for (const key of keyOrKeys) if (this.data.delete(key)) deleted += 1;
106
+ return deleted;
57
107
  }
58
108
  async deleteAll() {
59
109
  this.data.clear();
60
110
  }
61
111
  async list(options = {}) {
112
+ this.operationCounts.list += 1;
62
113
  const keys = applyListOptions(Array.from(this.data.keys()), options);
114
+ this.recordList(options, keys.length, false);
63
115
  return new Map(keys.map((k) => [k, this.data.get(k)]));
64
116
  }
65
117
  async transaction(cb) {
66
- const txn = new FakeTransaction(this);
67
- const result = await cb(txn);
68
- txn.commit();
69
- return result;
118
+ this.operationCounts.transaction += 1;
119
+ let release;
120
+ const previous = this.transactionTail;
121
+ this.transactionTail = new Promise((resolve) => {
122
+ release = resolve;
123
+ });
124
+ await previous;
125
+ try {
126
+ if (this.transactionDelayMs > 0) await new Promise((resolve) => setTimeout(resolve, this.transactionDelayMs));
127
+ const txn = new FakeTransaction(this);
128
+ const result = await cb(txn);
129
+ txn.commit();
130
+ return result;
131
+ } finally {
132
+ release();
133
+ }
134
+ }
135
+ /** Add deterministic per-transaction work for contention benchmarks. */
136
+ setTransactionDelay(ms) {
137
+ if (!Number.isFinite(ms) || ms < 0) throw new Error("transaction delay must be non-negative");
138
+ this.transactionDelayMs = ms;
70
139
  }
71
140
  async getAlarm() {
72
141
  return this.alarmAt;
@@ -80,6 +149,24 @@ var FakeStorage = class {
80
149
  now() {
81
150
  return this.clock();
82
151
  }
152
+ resetOperationCounts() {
153
+ for (const key of Object.keys(this.operationCounts)) this.operationCounts[key] = 0;
154
+ this.getManyCalls.length = 0;
155
+ this.operationCalls.length = 0;
156
+ }
157
+ /** @internal Used by FakeTransaction to count the same platform operations. */
158
+ recordOperation(operation) {
159
+ this.operationCounts[operation] += 1;
160
+ }
161
+ /** @internal Shared with FakeTransaction for deterministic key tracing. */
162
+ recordStorageCall(operation, keys, transactional) {
163
+ this.operationCalls.push({
164
+ operation,
165
+ keys: [...keys],
166
+ transactional,
167
+ synchronous: false
168
+ });
169
+ }
83
170
  /** Inject one matching write failure, including writes inside transactions. */
84
171
  failNextMutation(predicate, error = /* @__PURE__ */ new Error("injected fake storage mutation failure")) {
85
172
  this.mutationFailure = {
@@ -94,6 +181,14 @@ var FakeStorage = class {
94
181
  this.mutationFailure = void 0;
95
182
  throw error;
96
183
  }
184
+ /** @internal Shared with FakeTransaction for deterministic query tracing. */
185
+ recordList(options, resultSize, transactional) {
186
+ this.listCalls.push({
187
+ options: { ...options },
188
+ resultSize,
189
+ transactional
190
+ });
191
+ }
97
192
  };
98
193
  var FakeTransaction = class {
99
194
  storage;
@@ -103,21 +198,64 @@ var FakeTransaction = class {
103
198
  constructor(storage) {
104
199
  this.storage = storage;
105
200
  }
106
- async get(key) {
201
+ async get(keyOrKeys) {
202
+ if (Array.isArray(keyOrKeys)) {
203
+ this.storage.recordOperation("getMany");
204
+ this.storage.getManyCalls.push([...keyOrKeys]);
205
+ this.storage.recordStorageCall("get", keyOrKeys, true);
206
+ const result = /* @__PURE__ */ new Map();
207
+ for (const key of keyOrKeys) {
208
+ if (this.deleted.has(key)) continue;
209
+ if (this.staged.has(key)) result.set(key, this.staged.get(key));
210
+ else if (this.storage.data.has(key)) result.set(key, this.storage.data.get(key));
211
+ }
212
+ return result;
213
+ }
214
+ this.storage.recordOperation("get");
215
+ const key = keyOrKeys;
216
+ this.storage.recordStorageCall("get", [key], true);
107
217
  if (this.deleted.has(key)) return void 0;
108
218
  if (this.staged.has(key)) return this.staged.get(key);
109
219
  return this.storage.data.get(key);
110
220
  }
111
- async put(key, value) {
112
- this.storage.maybeFailMutation({
221
+ async put(keyOrEntries, value) {
222
+ if (typeof keyOrEntries === "string") {
223
+ this.storage.recordOperation("put");
224
+ this.storage.maybeFailMutation({
225
+ operation: "put",
226
+ key: keyOrEntries,
227
+ value
228
+ });
229
+ this.storage.recordStorageCall("put", [keyOrEntries], true);
230
+ this.deleted.delete(keyOrEntries);
231
+ this.staged.set(keyOrEntries, structuredClone(value));
232
+ return;
233
+ }
234
+ this.storage.recordOperation("putMany");
235
+ for (const [key, entryValue] of Object.entries(keyOrEntries)) this.storage.maybeFailMutation({
113
236
  operation: "put",
114
237
  key,
115
- value
238
+ value: entryValue
116
239
  });
117
- this.deleted.delete(key);
118
- this.staged.set(key, structuredClone(value));
240
+ this.storage.recordStorageCall("put", Object.keys(keyOrEntries), true);
241
+ for (const [key, entryValue] of Object.entries(keyOrEntries)) {
242
+ this.deleted.delete(key);
243
+ this.staged.set(key, structuredClone(entryValue));
244
+ }
245
+ }
246
+ async delete(keyOrKeys) {
247
+ if (typeof keyOrKeys === "string") {
248
+ this.storage.recordOperation("delete");
249
+ this.storage.recordStorageCall("delete", [keyOrKeys], true);
250
+ return this.deleteOne(keyOrKeys);
251
+ }
252
+ this.storage.recordOperation("deleteMany");
253
+ this.storage.recordStorageCall("delete", keyOrKeys, true);
254
+ let deleted = 0;
255
+ for (const key of keyOrKeys) if (this.deleteOne(key)) deleted += 1;
256
+ return deleted;
119
257
  }
120
- async delete(key) {
258
+ deleteOne(key) {
121
259
  this.storage.maybeFailMutation({
122
260
  operation: "delete",
123
261
  key
@@ -128,10 +266,12 @@ var FakeTransaction = class {
128
266
  return existed;
129
267
  }
130
268
  async list(options = {}) {
269
+ this.storage.recordOperation("list");
131
270
  const merged = new Map(this.storage.data);
132
271
  for (const key of this.deleted) merged.delete(key);
133
272
  for (const [key, value] of this.staged) merged.set(key, value);
134
273
  const keys = applyListOptions(Array.from(merged.keys()), options);
274
+ this.storage.recordList(options, keys.length, true);
135
275
  return new Map(keys.map((k) => [k, merged.get(k)]));
136
276
  }
137
277
  async getAlarm() {
@@ -175,28 +315,31 @@ var FakeFleet = class {
175
315
  get: (id) => this.cell(bindingKey, id.toString()).instance
176
316
  };
177
317
  }
318
+ instantiate(bindingKey, name, storage, pendingWaits) {
319
+ const CellCtor = this.classes[bindingKey];
320
+ if (!CellCtor) throw new Error(`no cell class registered for binding: ${bindingKey}`);
321
+ return new CellCtor({
322
+ storage,
323
+ id: {
324
+ toString: () => name,
325
+ name
326
+ },
327
+ waitUntil(promise) {
328
+ pendingWaits.push(promise);
329
+ },
330
+ async blockConcurrencyWhile(cb) {
331
+ return cb();
332
+ }
333
+ }, this.cellEnv);
334
+ }
178
335
  cell(bindingKey, name) {
179
336
  const key = `${bindingKey}\0${name}`;
180
337
  let slot = this.cells.get(key);
181
338
  if (!slot) {
182
- const CellCtor = this.classes[bindingKey];
183
- if (!CellCtor) throw new Error(`no cell class registered for binding: ${bindingKey}`);
184
339
  const storage = new FakeStorage(() => this.now);
185
340
  const pendingWaits = [];
186
341
  slot = {
187
- instance: new CellCtor({
188
- storage,
189
- id: {
190
- toString: () => name,
191
- name
192
- },
193
- waitUntil(promise) {
194
- pendingWaits.push(promise);
195
- },
196
- async blockConcurrencyWhile(cb) {
197
- return cb();
198
- }
199
- }, this.cellEnv),
342
+ instance: this.instantiate(bindingKey, name, storage, pendingWaits),
200
343
  storage,
201
344
  pendingWaits,
202
345
  alarmRetryCount: 0,
@@ -206,6 +349,12 @@ var FakeFleet = class {
206
349
  }
207
350
  return slot;
208
351
  }
352
+ /** Recreate one in-memory DO instance while preserving its durable storage. */
353
+ restartCell(bindingKey, name) {
354
+ const slot = this.cell(bindingKey, name);
355
+ slot.instance = this.instantiate(bindingKey, name, slot.storage, slot.pendingWaits);
356
+ return slot.instance;
357
+ }
209
358
  advance(ms) {
210
359
  this.now += ms;
211
360
  }
@@ -292,18 +441,28 @@ async function authenticate(request, secret) {
292
441
  * Routes:
293
442
  * GET /v1/health — liveness + spec version (unauthenticated)
294
443
  * POST /v1/rpc/{binding}/{name}/{method} — DO RPC (bearer auth)
444
+ * POST /v1/index/{domain}/{method} — cohesive index operation
445
+ * GET /v1/streams/{name}/chunks — bounded binary long-poll read
446
+ * POST /v1/streams/{name}/chunks — bounded binary batch append
295
447
  *
296
- * The RPC body is an rpc-codec-encoded JSON array of arguments; the response
297
- * body is the rpc-codec-encoded return value. Only whitelisted methods
298
- * dispatch — everything else (including fetch/alarm) is unreachable.
448
+ * Generic RPC bodies use the tagged JSON codec. Stream chunk bodies use the
449
+ * compact binary stream protocol. Only whitelisted routes/methods dispatch.
299
450
  */
300
451
  const WORLD_NAME = "world-celld";
301
452
  const WORLD_VERSION = "0.1.0";
453
+ function workflowIndex(env) {
454
+ return createWorkflowIndex({
455
+ runCatalog: env.WORKFLOW_RUN_CATALOG,
456
+ hookTokens: env.WORKFLOW_HOOK_TOKENS,
457
+ hookIds: env.WORKFLOW_HOOK_IDS
458
+ });
459
+ }
302
460
  const BINDINGS = {
303
461
  runs: {
304
462
  env: "WORKFLOW_DB",
305
463
  methods: /* @__PURE__ */ new Set([
306
464
  "applyEvent",
465
+ "getLifecycleStatus",
307
466
  "getRun",
308
467
  "getStep",
309
468
  "getEvent",
@@ -313,16 +472,15 @@ const BINDINGS = {
313
472
  "getCleanupStatus",
314
473
  "scheduleCleanup",
315
474
  "cleanupNow",
316
- "rearmCleanup"
475
+ "rearmCleanup",
476
+ "resolveHookTokenClaim"
317
477
  ])
318
478
  },
319
479
  streams: {
320
480
  env: "WORKFLOW_STREAMS",
321
481
  methods: /* @__PURE__ */ new Set([
322
- "writeChunk",
323
482
  "closeStream",
324
- "getChunks",
325
- "getInfo",
483
+ "failStream",
326
484
  "registerStream",
327
485
  "listStreams",
328
486
  "expireRegistry",
@@ -330,20 +488,13 @@ const BINDINGS = {
330
488
  "expireStream"
331
489
  ])
332
490
  },
333
- index: {
334
- env: "WORKFLOW_INDEX",
335
- methods: /* @__PURE__ */ new Set([
336
- "get",
337
- "put",
338
- "delete",
339
- "putOwned",
340
- "expireRun",
341
- "list",
342
- "reserveHookToken",
343
- "finalizeHookIndexes",
344
- "releaseHookToken",
345
- "deleteHookIndexes"
346
- ])
491
+ "hook-tokens": {
492
+ env: "WORKFLOW_HOOK_TOKENS",
493
+ methods: /* @__PURE__ */ new Set(["get"])
494
+ },
495
+ "hook-ids": {
496
+ env: "WORKFLOW_HOOK_IDS",
497
+ methods: /* @__PURE__ */ new Set(["get"])
347
498
  },
348
499
  queue: {
349
500
  env: "WORKFLOW_QUEUE",
@@ -354,12 +505,14 @@ const BINDINGS = {
354
505
  "redriveDeadLetter",
355
506
  "purgeDeadLetters",
356
507
  "rearmAlarm",
357
- "expireRun"
508
+ "expireRun",
509
+ "acknowledgeExpireRun"
358
510
  ])
359
511
  }
360
512
  };
361
513
  /** Request body cap: oversize payloads get a clear 413 instead of an OOM. */
362
514
  const MAX_BODY_BYTES = 33554432;
515
+ const MAX_STREAM_WRITE_BODY_BYTES = MAX_STREAM_BATCH_BYTES + 1024;
363
516
  function errorResponse(status, name, message) {
364
517
  return Response.json({ error: {
365
518
  name,
@@ -388,6 +541,49 @@ async function readBoundedBody(request) {
388
541
  reader.releaseLock();
389
542
  }
390
543
  }
544
+ async function readBoundedBytes(request, maxBytes) {
545
+ if (!request.body) return /* @__PURE__ */ new Uint8Array();
546
+ const reader = request.body.getReader();
547
+ const chunks = [];
548
+ let total = 0;
549
+ try {
550
+ for (;;) {
551
+ const { done, value } = await reader.read();
552
+ if (done) break;
553
+ total += value.byteLength;
554
+ if (total > maxBytes) {
555
+ await reader.cancel("stream body exceeds configured limit").catch(() => void 0);
556
+ return null;
557
+ }
558
+ chunks.push(value);
559
+ }
560
+ } finally {
561
+ reader.releaseLock();
562
+ }
563
+ const body = new Uint8Array(total);
564
+ let offset = 0;
565
+ for (const chunk of chunks) {
566
+ body.set(chunk, offset);
567
+ offset += chunk.byteLength;
568
+ }
569
+ return body;
570
+ }
571
+ function parseBoundedInteger(url, name, minimum, maximum) {
572
+ const raw = url.searchParams.get(name);
573
+ const value = raw === null ? NaN : Number(raw);
574
+ if (!Number.isSafeInteger(value) || value < minimum || value > maximum) {
575
+ const error = /* @__PURE__ */ new Error(`${name} must be an integer between ${minimum} and ${maximum}`);
576
+ error.name = "StreamProtocolError";
577
+ throw error;
578
+ }
579
+ return value;
580
+ }
581
+ function streamResponse(body) {
582
+ return new Response(body, {
583
+ status: 200,
584
+ headers: { "content-type": STREAM_BATCH_CONTENT_TYPE }
585
+ });
586
+ }
391
587
  function createRouter(env) {
392
588
  return async (request) => {
393
589
  const url = new URL(request.url);
@@ -399,9 +595,83 @@ function createRouter(env) {
399
595
  version: WORLD_VERSION,
400
596
  specVersion: SPEC_VERSION_CURRENT
401
597
  });
402
- if (parts[1] !== "rpc") return errorResponse(404, "NotFound", `unknown path: ${url.pathname}`);
403
598
  const auth = await authenticate(request, env.WORLD_SECRET);
404
599
  if (!auth.ok) return auth.response;
600
+ if (parts[1] === "index" && parts.length === 4) {
601
+ if (request.method !== "POST") return errorResponse(405, "MethodNotAllowed", "expected POST");
602
+ if (Number(request.headers.get("content-length") ?? "0") > MAX_BODY_BYTES) return errorResponse(413, "PayloadTooLarge", `body exceeds ${MAX_BODY_BYTES} bytes`);
603
+ let args;
604
+ try {
605
+ const text = await readBoundedBody(request);
606
+ if (text === null) return errorResponse(413, "PayloadTooLarge", `body exceeds ${MAX_BODY_BYTES} bytes`);
607
+ const parsed = rpcParse(text);
608
+ if (!Array.isArray(parsed)) return errorResponse(400, "BadRequest", "body must be an argument array");
609
+ args = parsed;
610
+ } catch {
611
+ return errorResponse(400, "BadRequest", "malformed rpc body");
612
+ }
613
+ try {
614
+ const index = workflowIndex(env);
615
+ let result;
616
+ const operation = `${parts[2]}.${parts[3]}`;
617
+ if (operation === "runs.list") result = await index.listRuns(args[0]);
618
+ else if (operation === "runs.commit") result = await index.commitRun(args[0], args[1], args[2]);
619
+ else if (operation === "runs.expire") result = await index.expireRun(args[0]);
620
+ else if (operation === "hooks.reserve") result = await index.reserveHook(args[0], args[1]);
621
+ else if (operation === "hooks.finalize") result = await index.finalizeHookIndexes(args[0], args[1], args[2], args[3], args[4]);
622
+ else if (operation === "hooks.release-reservation") result = await index.releaseHookReservation(args[0], args[1], args[2]);
623
+ else if (operation === "hooks.release") result = await index.releaseHookIndexes(args[0]);
624
+ else return errorResponse(404, "NotFound", `unknown index operation: ${operation}`);
625
+ return new Response(rpcStringify(result ?? null), {
626
+ status: 200,
627
+ headers: { "content-type": "application/json" }
628
+ });
629
+ } catch (error) {
630
+ const err = error;
631
+ return Response.json({ error: {
632
+ name: err?.name ?? "Error",
633
+ message: err?.message ?? String(error),
634
+ status: typeof err?.status === "number" ? err.status : void 0
635
+ } }, { status: 500 });
636
+ }
637
+ }
638
+ if (parts[1] === "streams" && parts.length === 4 && parts[3] === "chunks") {
639
+ const namespace = env.WORKFLOW_STREAMS;
640
+ if (!namespace) return errorResponse(500, "WorldMisconfigured", "missing binding: WORKFLOW_STREAMS");
641
+ const name = decodeURIComponent(parts[2]);
642
+ const runId = url.searchParams.get("runId");
643
+ if (!runId) return errorResponse(400, "BadRequest", "runId is required");
644
+ const stub = namespace.get(namespace.idFromName(name));
645
+ try {
646
+ if (request.method === "POST") {
647
+ if (!request.headers.get("content-type")?.startsWith("application/vnd.world-celld.stream-batch.v1")) return errorResponse(415, "UnsupportedMediaType", STREAM_BATCH_CONTENT_TYPE);
648
+ if (Number(request.headers.get("content-length") ?? "0") > MAX_STREAM_WRITE_BODY_BYTES) return errorResponse(413, "PayloadTooLarge", `body exceeds ${MAX_STREAM_WRITE_BODY_BYTES} bytes`);
649
+ const body = await readBoundedBytes(request, MAX_STREAM_WRITE_BODY_BYTES);
650
+ if (body === null) return errorResponse(413, "PayloadTooLarge", `body exceeds ${MAX_STREAM_WRITE_BODY_BYTES} bytes`);
651
+ const result = await stub.writeChunks(runId, decodeStreamWriteBatch(body));
652
+ return streamResponse(encodeStreamWriteResult(result));
653
+ }
654
+ if (request.method === "GET") {
655
+ const result = await stub.readChunks({
656
+ runId,
657
+ startIndex: parseBoundedInteger(url, "startIndex", 0, 2147483647),
658
+ maxChunks: parseBoundedInteger(url, "maxChunks", 0, 32),
659
+ maxBytes: parseBoundedInteger(url, "maxBytes", MAX_STREAM_CHUNK_BYTES, MAX_STREAM_READ_BYTES),
660
+ waitMs: parseBoundedInteger(url, "waitMs", 0, MAX_STREAM_LONG_POLL_MS)
661
+ }, request.signal);
662
+ return streamResponse(encodeStreamReadResult(result));
663
+ }
664
+ return errorResponse(405, "MethodNotAllowed", "expected GET or POST");
665
+ } catch (error) {
666
+ const err = error;
667
+ return Response.json({ error: {
668
+ name: err?.name ?? "Error",
669
+ message: err?.message ?? String(error),
670
+ status: typeof err?.status === "number" ? err.status : void 0
671
+ } }, { status: err?.name === "StreamProtocolError" ? 400 : 500 });
672
+ }
673
+ }
674
+ if (parts[1] !== "rpc") return errorResponse(404, "NotFound", `unknown path: ${url.pathname}`);
405
675
  if (request.method !== "POST" || parts.length !== 5) return errorResponse(404, "NotFound", "expected POST /v1/rpc/{binding}/{name}/{method}");
406
676
  const [, , bindingKey, encodedName, method] = parts;
407
677
  const binding = BINDINGS[bindingKey];
@@ -457,162 +727,367 @@ var DurableObject = class {
457
727
  }
458
728
  };
459
729
  //#endregion
460
- //#region src/worker/durable-objects/IndexDO.ts
730
+ //#region src/worker/durable-objects/hook-index-shared.ts
461
731
  function sameOwner(left, right) {
462
732
  return left.runId === right.runId && left.hookId === right.hookId;
463
733
  }
464
- function ownerFromHook(raw) {
734
+ function ownerFromSerializedHook(raw) {
465
735
  const hook = parse(raw);
466
736
  return {
467
737
  runId: hook.runId,
468
738
  hookId: hook.hookId
469
739
  };
470
740
  }
471
- /**
472
- * Global index cell, replacing the Workers KV namespace world-cloudflare
473
- * uses for run listing and hook token/id lookup.
474
- *
475
- * A single cell (named `index`) serializes every read and write. Hook-token
476
- * reservation, publication, and release use storage transactions so
477
- * concurrent runs cannot both acquire the same token.
478
- *
479
- * The RPC surface matches the storage-layer KVNamespace interface bit-for-bit
480
- * (including MockKVNamespace's pagination contract: cursor = last returned
481
- * key, `list_complete` reflects whether further keys exist), so the vendored
482
- * storage.ts works against it unchanged.
483
- */
484
- var IndexDO = class extends DurableObject {
485
- async get(key) {
486
- return await this.ctx.storage.get(key) ?? null;
741
+ function indexKey(prefix, value) {
742
+ return `${prefix}:${encodeURIComponent(value)}`;
743
+ }
744
+ async function runIsActive(env, runId) {
745
+ const namespace = env.WORKFLOW_DB;
746
+ if (!namespace) throw new Error("world-celld hook index missing WORKFLOW_DB binding");
747
+ return await namespace.get(namespace.idFromName(runId)).getLifecycleStatus() === "active";
748
+ }
749
+ //#endregion
750
+ //#region src/worker/durable-objects/HookIdDO.ts
751
+ const MAX_RELEASE_BATCH$1 = 64;
752
+ const CLAIM_DEADLINE_PREFIX$1 = "claim-deadline:";
753
+ function pad$3(ms) {
754
+ return String(Math.max(0, Math.floor(ms))).padStart(13, "0");
755
+ }
756
+ function recordKey$1(hookId) {
757
+ return indexKey("hookid", hookId);
758
+ }
759
+ function claimKey$1(hookId) {
760
+ return indexKey("claim", hookId);
761
+ }
762
+ function claimDeadlineKey$1(hookId, expiresAt) {
763
+ return `${CLAIM_DEADLINE_PREFIX$1}${pad$3(expiresAt)}:${encodeURIComponent(hookId)}`;
764
+ }
765
+ /** Hook-ID ownership and lookup shard; every hook ID has one transaction domain. */
766
+ var HookIdDO = class extends DurableObject {
767
+ now() {
768
+ return typeof this.env.clock === "function" ? this.env.clock() : Date.now();
487
769
  }
488
- async put(key, value) {
489
- await this.ctx.storage.put(key, value);
770
+ async get(hookId) {
771
+ const raw = await this.ctx.storage.get(recordKey$1(hookId));
772
+ if (!raw) return null;
773
+ const owner = ownerFromSerializedHook(raw);
774
+ return await runIsActive(this.env, owner.runId) ? raw : null;
490
775
  }
491
- /** Publish a derived index only while its owning run is not expired. */
492
- async putOwned(runId, key, value) {
776
+ async reserve(hookId, owner, proposedClaimId) {
493
777
  return await this.ctx.storage.transaction(async (txn) => {
494
- if (await txn.get(`expired:${runId}`) !== void 0) return { stored: false };
495
- await txn.put(key, value);
778
+ const record = recordKey$1(hookId);
779
+ const claim = claimKey$1(hookId);
780
+ const values = await txn.get([record, claim]);
781
+ const indexed = values.get(record);
782
+ if (indexed !== void 0) return {
783
+ claimed: false,
784
+ holder: ownerFromSerializedHook(indexed)
785
+ };
786
+ let existing = values.get(claim);
787
+ const now = this.now();
788
+ if (existing && existing.expiresAt <= now) {
789
+ await txn.delete([claim, existing.deadlineKey]);
790
+ existing = void 0;
791
+ }
792
+ if (existing !== void 0 && !sameOwner(existing.owner, owner)) return {
793
+ claimed: false,
794
+ holder: existing.owner
795
+ };
796
+ if (existing === void 0 || existing.claimId !== proposedClaimId) {
797
+ if (existing) await txn.delete(existing.deadlineKey);
798
+ const expiresAt = now + HOOK_CLAIM_LEASE_MS;
799
+ const deadlineKey = claimDeadlineKey$1(hookId, expiresAt);
800
+ await txn.put(claim, {
801
+ owner,
802
+ claimId: proposedClaimId,
803
+ reservedAt: now,
804
+ expiresAt,
805
+ deadlineKey
806
+ });
807
+ await txn.put(deadlineKey, {
808
+ hookId,
809
+ claimId: proposedClaimId,
810
+ expiresAt
811
+ });
812
+ await this.armAtMost(txn, expiresAt);
813
+ return {
814
+ claimed: true,
815
+ claimId: proposedClaimId
816
+ };
817
+ }
818
+ return {
819
+ claimed: true,
820
+ claimId: existing.claimId
821
+ };
822
+ });
823
+ }
824
+ async publish(hookId, serializedHook, owner, claimId) {
825
+ if (!await runIsActive(this.env, owner.runId)) {
826
+ await this.releaseBatch([{
827
+ hookId,
828
+ owner
829
+ }]);
830
+ return { stored: false };
831
+ }
832
+ return await this.ctx.storage.transaction(async (txn) => {
833
+ const record = recordKey$1(hookId);
834
+ const claim = claimKey$1(hookId);
835
+ const values = await txn.get([record, claim]);
836
+ const indexed = values.get(record);
837
+ if (indexed !== void 0 && !sameOwner(ownerFromSerializedHook(indexed), owner)) throw new Error(`Hook id ${hookId} is owned by another hook`);
838
+ const existingClaim = values.get(claim);
839
+ if (existingClaim && existingClaim.expiresAt <= this.now()) {
840
+ await txn.delete([claim, existingClaim.deadlineKey]);
841
+ return { stored: false };
842
+ }
843
+ if (existingClaim !== void 0 && (!sameOwner(existingClaim.owner, owner) || existingClaim.claimId !== claimId)) throw new Error(`Hook id ${hookId} is reserved by another hook`);
844
+ if (indexed === void 0 && existingClaim === void 0) throw new Error(`Hook id ${hookId} has no active reservation`);
845
+ await txn.put(record, serializedHook);
846
+ if (existingClaim !== void 0) await txn.delete([claim, existingClaim.deadlineKey]);
496
847
  return { stored: true };
497
848
  });
498
849
  }
499
- async delete(key) {
500
- await this.ctx.storage.delete(key);
850
+ async releaseClaim(hookId, owner, claimId) {
851
+ await this.ctx.storage.transaction(async (txn) => {
852
+ const key = claimKey$1(hookId);
853
+ const claim = await txn.get(key);
854
+ if (claim && sameOwner(claim.owner, owner) && claim.claimId === claimId) await txn.delete([key, claim.deadlineKey]);
855
+ });
501
856
  }
502
- async list(options) {
503
- const prefix = options?.prefix ?? "";
504
- const limit = options?.limit ?? 1e3;
505
- const entries = await this.ctx.storage.list({
506
- prefix,
507
- ...options?.reverse ? {
508
- reverse: true,
509
- end: options.cursor
510
- } : { startAfter: options?.cursor },
511
- limit: limit + 1
857
+ async releaseBatch(hooks) {
858
+ if (hooks.length > MAX_RELEASE_BATCH$1) throw new Error("Hook id release batch is too large");
859
+ if (hooks.length === 0) return { deleted: 0 };
860
+ return await this.ctx.storage.transaction(async (txn) => {
861
+ const keys = hooks.flatMap(({ hookId }) => [recordKey$1(hookId), claimKey$1(hookId)]);
862
+ const values = await txn.get(keys);
863
+ const deletes = [];
864
+ const deadlineDeletes = [];
865
+ for (const { hookId, owner } of hooks) {
866
+ const record = recordKey$1(hookId);
867
+ const claim = claimKey$1(hookId);
868
+ const indexed = values.get(record);
869
+ const existingClaim = values.get(claim);
870
+ if (indexed !== void 0 && sameOwner(ownerFromSerializedHook(indexed), owner)) deletes.push(record);
871
+ if (existingClaim !== void 0 && sameOwner(existingClaim.owner, owner)) {
872
+ deletes.push(claim);
873
+ deadlineDeletes.push(existingClaim.deadlineKey);
874
+ }
875
+ }
876
+ const deleted = deletes.length === 0 ? 0 : await txn.delete(deletes);
877
+ if (deadlineDeletes.length > 0) await txn.delete(deadlineDeletes);
878
+ return { deleted };
512
879
  });
513
- const names = Array.from(entries.keys());
514
- const page = names.slice(0, limit);
515
- const listComplete = names.length <= limit;
516
- return {
517
- keys: page.map((name) => ({ name })),
518
- list_complete: listComplete,
519
- cursor: listComplete ? void 0 : page.at(-1)
520
- };
521
880
  }
522
- async reserveHookToken(token, owner) {
523
- return this.ctx.storage.transaction(async (txn) => {
524
- const indexedHook = await txn.get(`hook:${token}`);
525
- if (indexedHook !== void 0) return {
881
+ async alarm() {
882
+ try {
883
+ await this.compactClaims();
884
+ } catch {
885
+ await this.ctx.storage.setAlarm(this.now() + LIFECYCLE_COMPACTION_RETRY_MS);
886
+ }
887
+ }
888
+ async armAtMost(storage, at) {
889
+ const current = await storage.getAlarm();
890
+ if (current === null || at < current) await storage.setAlarm(at);
891
+ }
892
+ async compactClaims() {
893
+ const now = this.now();
894
+ await this.ctx.storage.transaction(async (txn) => {
895
+ const due = await txn.list({
896
+ prefix: CLAIM_DEADLINE_PREFIX$1,
897
+ end: `${CLAIM_DEADLINE_PREFIX$1}${pad$3(now + 1)}`,
898
+ limit: 129
899
+ });
900
+ const page = Array.from(due).slice(0, 128);
901
+ const claims = await txn.get(page.map(([, item]) => claimKey$1(item.hookId)));
902
+ const deletes = page.flatMap(([deadline, item]) => {
903
+ const key = claimKey$1(item.hookId);
904
+ const claim = claims.get(key);
905
+ return claim?.claimId === item.claimId && claim.expiresAt <= now ? [deadline, key] : [deadline];
906
+ });
907
+ for (let offset = 0; offset < deletes.length; offset += 128) await txn.delete(deletes.slice(offset, offset + 128));
908
+ if (due.size > 128) {
909
+ await txn.setAlarm(now + 1);
910
+ return;
911
+ }
912
+ const nextAt = (await txn.list({
913
+ prefix: CLAIM_DEADLINE_PREFIX$1,
914
+ limit: 1
915
+ })).values().next().value?.expiresAt;
916
+ if (nextAt === void 0) await txn.deleteAlarm();
917
+ else await txn.setAlarm(Math.max(now + 1, nextAt));
918
+ });
919
+ }
920
+ };
921
+ //#endregion
922
+ //#region src/worker/durable-objects/HookTokenDO.ts
923
+ const MAX_RELEASE_BATCH = 64;
924
+ const CLAIM_DEADLINE_PREFIX = "claim-deadline:";
925
+ function pad$2(ms) {
926
+ return String(Math.max(0, Math.floor(ms))).padStart(13, "0");
927
+ }
928
+ function recordKey(token) {
929
+ return indexKey("hook", token);
930
+ }
931
+ function claimKey(token) {
932
+ return indexKey("claim", token);
933
+ }
934
+ function claimDeadlineKey(token, expiresAt) {
935
+ return `${CLAIM_DEADLINE_PREFIX}${pad$2(expiresAt)}:${encodeURIComponent(token)}`;
936
+ }
937
+ /** Token-ownership shard. A token hashes to exactly one shard and transaction domain. */
938
+ var HookTokenDO = class extends DurableObject {
939
+ now() {
940
+ return typeof this.env.clock === "function" ? this.env.clock() : Date.now();
941
+ }
942
+ async get(token) {
943
+ const raw = await this.ctx.storage.get(recordKey(token));
944
+ if (!raw) return null;
945
+ const owner = ownerFromSerializedHook(raw);
946
+ return await runIsActive(this.env, owner.runId) ? raw : null;
947
+ }
948
+ async reserve(token, owner, proposedClaimId) {
949
+ return await this.ctx.storage.transaction(async (txn) => {
950
+ const record = recordKey(token);
951
+ const claim = claimKey(token);
952
+ const values = await txn.get([record, claim]);
953
+ const indexed = values.get(record);
954
+ if (indexed !== void 0) return {
526
955
  claimed: false,
527
- holder: ownerFromHook(indexedHook)
956
+ holder: ownerFromSerializedHook(indexed)
528
957
  };
529
- const claimKey = `hookclaim:${token}`;
530
- const claim = await txn.get(claimKey);
531
- if (claim !== void 0 && !sameOwner(claim.owner, owner)) return {
958
+ let existing = values.get(claim);
959
+ const now = this.now();
960
+ if (existing && existing.expiresAt <= now) {
961
+ await txn.delete([claim, existing.deadlineKey]);
962
+ existing = void 0;
963
+ }
964
+ if (existing !== void 0 && !sameOwner(existing.owner, owner)) return {
532
965
  claimed: false,
533
- holder: claim.owner
966
+ holder: existing.owner
967
+ };
968
+ if (existing === void 0) {
969
+ const expiresAt = now + HOOK_CLAIM_LEASE_MS;
970
+ const deadlineKey = claimDeadlineKey(token, expiresAt);
971
+ await txn.put(claim, {
972
+ owner,
973
+ claimId: proposedClaimId,
974
+ reservedAt: now,
975
+ expiresAt,
976
+ deadlineKey
977
+ });
978
+ await txn.put(deadlineKey, {
979
+ token,
980
+ claimId: proposedClaimId,
981
+ expiresAt
982
+ });
983
+ await this.armAtMost(txn, expiresAt);
984
+ return {
985
+ claimed: true,
986
+ claimId: proposedClaimId
987
+ };
988
+ }
989
+ return {
990
+ claimed: true,
991
+ claimId: existing.claimId
534
992
  };
535
- await txn.put(claimKey, {
536
- owner,
537
- reservedAt: Date.now()
538
- });
539
- return { claimed: true };
540
993
  });
541
994
  }
542
- async finalizeHookIndexes(token, hookId, serializedHook, owner) {
543
- await this.ctx.storage.transaction(async (txn) => {
544
- if (await txn.get(`expired:${owner.runId}`) !== void 0) {
545
- const claim = await txn.get(`hookclaim:${token}`);
546
- if (claim !== void 0 && sameOwner(claim.owner, owner)) await txn.delete(`hookclaim:${token}`);
547
- return;
995
+ async finalize(token, serializedHook, owner, claimId) {
996
+ if (!await runIsActive(this.env, owner.runId)) {
997
+ await this.releaseBatch([{
998
+ token,
999
+ owner
1000
+ }]);
1001
+ return { stored: false };
1002
+ }
1003
+ return await this.ctx.storage.transaction(async (txn) => {
1004
+ const record = recordKey(token);
1005
+ const claim = claimKey(token);
1006
+ const values = await txn.get([record, claim]);
1007
+ const indexed = values.get(record);
1008
+ if (indexed !== void 0 && !sameOwner(ownerFromSerializedHook(indexed), owner)) throw new Error(`Hook token ${token} is owned by another hook`);
1009
+ const existingClaim = values.get(claim);
1010
+ if (existingClaim && existingClaim.expiresAt <= this.now()) {
1011
+ await txn.delete([claim, existingClaim.deadlineKey]);
1012
+ return { stored: false };
548
1013
  }
549
- const indexedHook = await txn.get(`hook:${token}`);
550
- if (indexedHook !== void 0 && !sameOwner(ownerFromHook(indexedHook), owner)) throw new Error(`Hook token ${token} is owned by another hook`);
551
- const claim = await txn.get(`hookclaim:${token}`);
552
- if (claim !== void 0 && !sameOwner(claim.owner, owner)) throw new Error(`Hook token ${token} is reserved by another hook`);
553
- await txn.put(`hook:${token}`, serializedHook);
554
- await txn.put(`hookid:${hookId}`, serializedHook);
555
- if (claim !== void 0) await txn.delete(`hookclaim:${token}`);
1014
+ if (existingClaim !== void 0 && (!sameOwner(existingClaim.owner, owner) || existingClaim.claimId !== claimId)) throw new Error(`Hook token ${token} is reserved by another hook`);
1015
+ if (indexed === void 0 && existingClaim === void 0) throw new Error(`Hook token ${token} has no active reservation`);
1016
+ await txn.put(record, serializedHook);
1017
+ if (existingClaim !== void 0) await txn.delete([claim, existingClaim.deadlineKey]);
1018
+ return { stored: true };
556
1019
  });
557
1020
  }
558
- async releaseHookToken(token, owner) {
1021
+ async releaseClaim(token, owner, claimId) {
559
1022
  await this.ctx.storage.transaction(async (txn) => {
560
- const key = `hookclaim:${token}`;
1023
+ const key = claimKey(token);
561
1024
  const claim = await txn.get(key);
562
- if (claim !== void 0 && sameOwner(claim.owner, owner)) await txn.delete(key);
1025
+ if (claim && sameOwner(claim.owner, owner) && claim.claimId === claimId) await txn.delete([key, claim.deadlineKey]);
563
1026
  });
564
1027
  }
565
- async deleteHookIndexes(token, hookId, owner) {
566
- await this.ctx.storage.transaction(async (txn) => {
567
- const tokenKey = `hook:${token}`;
568
- const idKey = `hookid:${hookId}`;
569
- const claimKey = `hookclaim:${token}`;
570
- const [byToken, byId, claim] = await Promise.all([
571
- txn.get(tokenKey),
572
- txn.get(idKey),
573
- txn.get(claimKey)
574
- ]);
575
- if (byToken !== void 0 && sameOwner(ownerFromHook(byToken), owner)) await txn.delete(tokenKey);
576
- if (byId !== void 0 && sameOwner(ownerFromHook(byId), owner)) await txn.delete(idKey);
577
- if (claim !== void 0 && sameOwner(claim.owner, owner)) await txn.delete(claimKey);
578
- });
579
- }
580
- /**
581
- * Fence a run and remove all of its known derived indexes atomically.
582
- * Once the fence exists, putOwned() and hook finalization cannot resurrect
583
- * entries from a delayed request.
584
- */
585
- async expireRun(request) {
1028
+ async releaseBatch(hooks) {
1029
+ if (hooks.length > MAX_RELEASE_BATCH) throw new Error("Hook token release batch is too large");
1030
+ if (hooks.length === 0) return { deleted: 0 };
586
1031
  return await this.ctx.storage.transaction(async (txn) => {
587
- await txn.put(`expired:${request.runId}`, request.expiredAt);
588
- let deleted = 0;
589
- for (const key of new Set(request.keys)) if (await txn.delete(key)) deleted++;
590
- for (const hook of request.hooks) {
591
- const owner = {
592
- runId: request.runId,
593
- hookId: hook.hookId
594
- };
595
- const tokenKey = `hook:${hook.token}`;
596
- const idKey = `hookid:${hook.hookId}`;
597
- const claimKey = `hookclaim:${hook.token}`;
598
- const [byToken, byId, claim] = await Promise.all([
599
- txn.get(tokenKey),
600
- txn.get(idKey),
601
- txn.get(claimKey)
602
- ]);
603
- if (byToken !== void 0 && sameOwner(ownerFromHook(byToken), owner)) {
604
- if (await txn.delete(tokenKey)) deleted++;
605
- }
606
- if (byId !== void 0 && sameOwner(ownerFromHook(byId), owner)) {
607
- if (await txn.delete(idKey)) deleted++;
608
- }
609
- if (claim !== void 0 && sameOwner(claim.owner, owner)) {
610
- if (await txn.delete(claimKey)) deleted++;
1032
+ const keys = hooks.flatMap(({ token }) => [recordKey(token), claimKey(token)]);
1033
+ const values = await txn.get(keys);
1034
+ const deletes = [];
1035
+ const deadlineDeletes = [];
1036
+ for (const { token, owner } of hooks) {
1037
+ const record = recordKey(token);
1038
+ const claim = claimKey(token);
1039
+ const indexed = values.get(record);
1040
+ const existingClaim = values.get(claim);
1041
+ if (indexed !== void 0 && sameOwner(ownerFromSerializedHook(indexed), owner)) deletes.push(record);
1042
+ if (existingClaim !== void 0 && sameOwner(existingClaim.owner, owner)) {
1043
+ deletes.push(claim);
1044
+ deadlineDeletes.push(existingClaim.deadlineKey);
611
1045
  }
612
1046
  }
1047
+ const deleted = deletes.length === 0 ? 0 : await txn.delete(deletes);
1048
+ if (deadlineDeletes.length > 0) await txn.delete(deadlineDeletes);
613
1049
  return { deleted };
614
1050
  });
615
1051
  }
1052
+ async alarm() {
1053
+ try {
1054
+ await this.compactClaims();
1055
+ } catch {
1056
+ await this.ctx.storage.setAlarm(this.now() + LIFECYCLE_COMPACTION_RETRY_MS);
1057
+ }
1058
+ }
1059
+ async armAtMost(storage, at) {
1060
+ const current = await storage.getAlarm();
1061
+ if (current === null || at < current) await storage.setAlarm(at);
1062
+ }
1063
+ async compactClaims() {
1064
+ const now = this.now();
1065
+ await this.ctx.storage.transaction(async (txn) => {
1066
+ const due = await txn.list({
1067
+ prefix: CLAIM_DEADLINE_PREFIX,
1068
+ end: `${CLAIM_DEADLINE_PREFIX}${pad$2(now + 1)}`,
1069
+ limit: 129
1070
+ });
1071
+ const page = Array.from(due).slice(0, 128);
1072
+ const claims = await txn.get(page.map(([, item]) => claimKey(item.token)));
1073
+ const deletes = page.flatMap(([deadline, item]) => {
1074
+ const key = claimKey(item.token);
1075
+ const claim = claims.get(key);
1076
+ return claim?.claimId === item.claimId && claim.expiresAt <= now ? [deadline, key] : [deadline];
1077
+ });
1078
+ for (let offset = 0; offset < deletes.length; offset += 128) await txn.delete(deletes.slice(offset, offset + 128));
1079
+ if (due.size > 128) {
1080
+ await txn.setAlarm(now + 1);
1081
+ return;
1082
+ }
1083
+ const nextAt = (await txn.list({
1084
+ prefix: CLAIM_DEADLINE_PREFIX,
1085
+ limit: 1
1086
+ })).values().next().value?.expiresAt;
1087
+ if (nextAt === void 0) await txn.deleteAlarm();
1088
+ else await txn.setAlarm(Math.max(now + 1, nextAt));
1089
+ });
1090
+ }
616
1091
  };
617
1092
  //#endregion
618
1093
  //#region src/worker/durable-objects/QueueDO.ts
@@ -625,15 +1100,26 @@ const PERMANENT_STATUSES = /* @__PURE__ */ new Set([
625
1100
  const DEFAULT_MAX_ATTEMPTS = 5;
626
1101
  const DEFAULT_MAX_INFLIGHT = 5;
627
1102
  const DEFAULT_DELIVERY_TIMEOUT_MS = 3e5;
628
- /** Grace period past the delivery timeout before a claim counts as lost. */
629
- const INFLIGHT_GRACE_MS = 3e4;
630
1103
  /** Move overdue alarms to a new timestamp so celld observes a fresh edge. */
631
1104
  const MIN_ALARM_DELAY_MS = 1;
632
- function pad(ms) {
1105
+ /** Bound expiry recovery so one alarm cannot monopolize the cell. */
1106
+ const EXPIRED_INFLIGHT_BATCH = 128;
1107
+ const STORAGE_BATCH_SIZE$2 = 128;
1108
+ const EXPIRE_RUN_REFERENCE_BATCH = 64;
1109
+ const PURGE_DEAD_LETTER_BATCH = 128;
1110
+ const INFLIGHT_DEADLINE_PREFIX = "inflight-deadline:";
1111
+ const EXPIRED_RUN_GC_PREFIX = "expired-run-gc:";
1112
+ function pad$1(ms) {
633
1113
  return String(Math.max(0, Math.floor(ms))).padStart(13, "0");
634
1114
  }
635
1115
  function dueKey(atMs, messageId) {
636
- return `due:${pad(atMs)}:${messageId}`;
1116
+ return `due:${pad$1(atMs)}:${messageId}`;
1117
+ }
1118
+ function inflightKey(deadlineMs, messageId) {
1119
+ return `${INFLIGHT_DEADLINE_PREFIX}${pad$1(deadlineMs)}:${messageId}`;
1120
+ }
1121
+ function deadlineFromInflightKey(key) {
1122
+ return Number.parseInt(key.slice(18, 31), 10);
637
1123
  }
638
1124
  function runReferenceKey(runId, messageId) {
639
1125
  return `run:${runId}:${messageId}`;
@@ -641,9 +1127,40 @@ function runReferenceKey(runId, messageId) {
641
1127
  function expiredRunKey(runId) {
642
1128
  return `expired-run:${runId}`;
643
1129
  }
1130
+ function expiredRunGcKey(runId, compactAt) {
1131
+ return `${EXPIRED_RUN_GC_PREFIX}${pad$1(compactAt)}:${encodeURIComponent(runId)}`;
1132
+ }
1133
+ function boundedLimit$1(value, fallback, maximum) {
1134
+ if (value === void 0 || !Number.isFinite(value)) return fallback;
1135
+ return Math.max(1, Math.min(maximum, Math.floor(value)));
1136
+ }
1137
+ async function getMany$1(storage, keys) {
1138
+ const result = /* @__PURE__ */ new Map();
1139
+ for (let offset = 0; offset < keys.length; offset += STORAGE_BATCH_SIZE$2) {
1140
+ const page = await storage.get(keys.slice(offset, offset + STORAGE_BATCH_SIZE$2));
1141
+ for (const [key, value] of page) result.set(key, value);
1142
+ }
1143
+ return result;
1144
+ }
1145
+ async function putMany(storage, entries) {
1146
+ const all = Array.from(entries);
1147
+ for (let offset = 0; offset < all.length; offset += STORAGE_BATCH_SIZE$2) await storage.put(Object.fromEntries(all.slice(offset, offset + STORAGE_BATCH_SIZE$2)));
1148
+ }
1149
+ async function deleteMany$1(storage, keys) {
1150
+ let deleted = 0;
1151
+ const unique = Array.from(new Set(keys));
1152
+ for (let offset = 0; offset < unique.length; offset += STORAGE_BATCH_SIZE$2) deleted += await storage.delete(unique.slice(offset, offset + STORAGE_BATCH_SIZE$2));
1153
+ return deleted;
1154
+ }
644
1155
  function backoffSeconds(attempt) {
645
1156
  return Math.min(60, 2 ** attempt);
646
1157
  }
1158
+ async function cancelUnusedResponseBody(response) {
1159
+ try {
1160
+ await response.body?.cancel();
1161
+ } catch {}
1162
+ }
1163
+ const MAX_QUEUE_INFLIGHT = 128;
647
1164
  var QueueDO = class extends DurableObject {
648
1165
  #now() {
649
1166
  const clock = this.env?.clock;
@@ -658,15 +1175,34 @@ var QueueDO = class extends DurableObject {
658
1175
  const parsed = raw ? Number.parseInt(raw, 10) : NaN;
659
1176
  return Number.isFinite(parsed) && parsed > 0 ? parsed : fallback;
660
1177
  }
1178
+ #maxInflight() {
1179
+ const configured = this.#intVar("QUEUE_MAX_INFLIGHT", DEFAULT_MAX_INFLIGHT);
1180
+ if (configured > MAX_QUEUE_INFLIGHT) throw new Error(`QUEUE_MAX_INFLIGHT must be at most ${MAX_QUEUE_INFLIGHT}`);
1181
+ return configured;
1182
+ }
1183
+ #deliveryTimeoutMs() {
1184
+ const configured = this.#intVar("QUEUE_DELIVERY_TIMEOUT_MS", DEFAULT_DELIVERY_TIMEOUT_MS);
1185
+ if (configured > 3e5) throw new Error(`QUEUE_DELIVERY_TIMEOUT_MS must be at most ${MAX_QUEUE_DELIVERY_TIMEOUT_MS}`);
1186
+ return configured;
1187
+ }
1188
+ async #authoritativeRunExpired(runId) {
1189
+ const namespace = this.env?.WORKFLOW_DB;
1190
+ if (!namespace) throw new Error("world-celld queue missing WORKFLOW_DB binding");
1191
+ return await namespace.get(namespace.idFromName(runId)).getLifecycleStatus() === "expired";
1192
+ }
1193
+ async #withRunMutation(runId, onExpired, onAllowed) {
1194
+ const expired = await this.#authoritativeRunExpired(runId);
1195
+ return await this.ctx.storage.transaction(async (txn) => {
1196
+ if (await txn.get(expiredRunKey(runId)) || expired) return await onExpired(txn);
1197
+ return await onAllowed(txn);
1198
+ });
1199
+ }
661
1200
  async enqueue(request) {
1201
+ this.#maxInflight();
1202
+ this.#deliveryTimeoutMs();
662
1203
  const storage = this.ctx.storage;
663
1204
  const now = this.#now();
664
- return await storage.transaction(async (txn) => {
665
- if (request.runId && await txn.get(expiredRunKey(request.runId)) !== void 0) return {
666
- ok: false,
667
- code: "RUN_EXPIRED",
668
- message: `Workflow run "${request.runId}" has expired`
669
- };
1205
+ const persist = async (txn) => {
670
1206
  const pinned = await txn.get("cfg");
671
1207
  if (!pinned) await txn.put("cfg", {
672
1208
  queueShards: request.config.queueShards,
@@ -718,7 +1254,12 @@ var QueueDO = class extends DurableObject {
718
1254
  messageId: row.messageId,
719
1255
  deduped: false
720
1256
  };
721
- });
1257
+ };
1258
+ return request.runId ? await this.#withRunMutation(request.runId, async () => ({
1259
+ ok: false,
1260
+ code: "RUN_EXPIRED",
1261
+ message: `Workflow run "${request.runId}" has expired`
1262
+ }), persist) : await storage.transaction(persist);
722
1263
  }
723
1264
  async alarm() {
724
1265
  let claimed = [];
@@ -732,8 +1273,11 @@ var QueueDO = class extends DurableObject {
732
1273
  };
733
1274
  if (typeof this.ctx.blockConcurrencyWhile === "function") await this.ctx.blockConcurrencyWhile(attempt);
734
1275
  else await attempt();
735
- if (failure !== null) throw failure;
736
- for (const row of claimed) this.ctx.waitUntil(this.#deliver(row));
1276
+ if (failure !== null) {
1277
+ await this.ctx.storage.setAlarm(this.#now() + LIFECYCLE_COMPACTION_RETRY_MS);
1278
+ throw failure;
1279
+ }
1280
+ for (const claim of claimed) this.ctx.waitUntil(this.#deliver(claim));
737
1281
  }
738
1282
  /**
739
1283
  * Storage-only: recover lost inflight claims, claim due messages up to the
@@ -742,49 +1286,83 @@ var QueueDO = class extends DurableObject {
742
1286
  async #claimPhase() {
743
1287
  const storage = this.ctx.storage;
744
1288
  const now = this.#now();
745
- const maxInflight = this.#intVar("QUEUE_MAX_INFLIGHT", DEFAULT_MAX_INFLIGHT);
1289
+ const maxInflight = this.#maxInflight();
746
1290
  return await storage.transaction(async (txn) => {
747
- const inflight = await txn.list({ prefix: "inflight:" });
748
- let inflightCount = 0;
749
- for (const [key, deadline] of inflight) if (deadline <= now) {
750
- const messageId = key.slice(9);
751
- await txn.delete(key);
752
- const row = await txn.get(`msg:${messageId}`);
1291
+ if (await this.#compactExpiredRunFences(txn, now)) {
1292
+ await txn.setAlarm(now + MIN_ALARM_DELAY_MS);
1293
+ return [];
1294
+ }
1295
+ const expiredPage = await txn.list({
1296
+ prefix: INFLIGHT_DEADLINE_PREFIX,
1297
+ end: `${INFLIGHT_DEADLINE_PREFIX}${pad$1(now + 1)}`,
1298
+ limit: 129
1299
+ });
1300
+ const expired = Array.from(expiredPage.entries()).slice(0, EXPIRED_INFLIGHT_BATCH);
1301
+ const expiredRows = await getMany$1(txn, expired.map(([, messageId]) => `msg:${messageId}`));
1302
+ await deleteMany$1(txn, expired.map(([key]) => key));
1303
+ const recoveryWrites = [];
1304
+ for (const [, messageId] of expired) {
1305
+ const row = expiredRows.get(`msg:${messageId}`);
753
1306
  if (row) {
754
1307
  const scheduleKey = dueKey(now, messageId);
755
- await txn.put(scheduleKey, messageId);
756
- if (row.runId) await txn.put(runReferenceKey(row.runId, row.messageId), {
1308
+ recoveryWrites.push([scheduleKey, messageId]);
1309
+ if (row.runId) recoveryWrites.push([runReferenceKey(row.runId, row.messageId), {
757
1310
  messageId: row.messageId,
758
1311
  dueKey: scheduleKey,
759
1312
  idempotencyKey: row.idempotencyKey
760
- });
1313
+ }]);
761
1314
  }
762
- } else inflightCount++;
1315
+ }
1316
+ await putMany(txn, recoveryWrites);
1317
+ if (expiredPage.size > EXPIRED_INFLIGHT_BATCH) {
1318
+ await txn.setAlarm(now + MIN_ALARM_DELAY_MS);
1319
+ return [];
1320
+ }
1321
+ const inflightCount = (await txn.list({
1322
+ prefix: INFLIGHT_DEADLINE_PREFIX,
1323
+ limit: maxInflight
1324
+ })).size;
763
1325
  const rows = [];
764
1326
  if (inflightCount < maxInflight) {
765
1327
  const due = await txn.list({
766
1328
  prefix: "due:",
767
- end: `due:${pad(now + 1)}`,
1329
+ end: `due:${pad$1(now + 1)}`,
768
1330
  limit: maxInflight - inflightCount
769
1331
  });
770
- const timeoutMs = this.#intVar("QUEUE_DELIVERY_TIMEOUT_MS", DEFAULT_DELIVERY_TIMEOUT_MS);
771
- for (const [key, messageId] of due) {
772
- await txn.delete(key);
773
- const row = await txn.get(`msg:${messageId}`);
1332
+ const timeoutMs = this.#deliveryTimeoutMs();
1333
+ const dueRows = await getMany$1(txn, Array.from(due.values(), (messageId) => `msg:${messageId}`));
1334
+ await deleteMany$1(txn, Array.from(due.keys()));
1335
+ const claimWrites = [];
1336
+ for (const [, messageId] of due) {
1337
+ const row = dueRows.get(`msg:${messageId}`);
774
1338
  if (!row) continue;
775
- await txn.put(`inflight:${messageId}`, now + timeoutMs + INFLIGHT_GRACE_MS);
776
- rows.push(row);
1339
+ const claimKey = inflightKey(now + timeoutMs + QUEUE_INFLIGHT_GRACE_MS, messageId);
1340
+ claimWrites.push([claimKey, messageId]);
1341
+ if (row.runId) claimWrites.push([runReferenceKey(row.runId, row.messageId), {
1342
+ messageId: row.messageId,
1343
+ inflightKey: claimKey,
1344
+ idempotencyKey: row.idempotencyKey
1345
+ }]);
1346
+ rows.push({
1347
+ row,
1348
+ inflightKey: claimKey
1349
+ });
777
1350
  }
1351
+ await putMany(txn, claimWrites);
778
1352
  }
779
1353
  await this.#scheduleNextAlarm(txn, now);
780
1354
  return rows;
781
1355
  });
782
1356
  }
783
- async #deliver(row) {
1357
+ async #deliver({ row, inflightKey: claimKey }) {
784
1358
  const storage = this.ctx.storage;
785
- const timeoutMs = this.#intVar("QUEUE_DELIVERY_TIMEOUT_MS", DEFAULT_DELIVERY_TIMEOUT_MS);
1359
+ const timeoutMs = this.#deliveryTimeoutMs();
786
1360
  const callbackSecret = this.env?.WORKFLOW_CALLBACK_SECRET;
787
1361
  const url = `${row.targetBaseUrl.replace(/\/$/, "")}/.well-known/workflow/v1/${row.pathname}`;
1362
+ if (row.runId && await this.#authoritativeRunExpired(row.runId)) {
1363
+ await storage.transaction((txn) => this.#dropExpiredMessage(txn, row, claimKey, this.#now()));
1364
+ return;
1365
+ }
788
1366
  let response = null;
789
1367
  let transportError = null;
790
1368
  try {
@@ -804,47 +1382,44 @@ var QueueDO = class extends DurableObject {
804
1382
  transportError = error;
805
1383
  }
806
1384
  const now = this.#now();
807
- if (response?.ok) await this.#ack(row);
808
- else if (response && response.status === 503) {
1385
+ if (response?.ok) {
1386
+ await cancelUnusedResponseBody(response);
1387
+ await this.#ack(row, claimKey);
1388
+ } else if (response && response.status === 503) {
809
1389
  let timeoutSeconds;
810
1390
  try {
811
1391
  timeoutSeconds = (await response.json())?.timeoutSeconds;
812
1392
  } catch {
813
1393
  timeoutSeconds = void 0;
814
1394
  }
815
- if (typeof timeoutSeconds === "number") await storage.transaction(async (txn) => {
816
- if (row.runId && await txn.get(expiredRunKey(row.runId)) !== void 0) {
817
- await txn.delete(`msg:${row.messageId}`);
818
- await txn.delete(`inflight:${row.messageId}`);
819
- await txn.delete(runReferenceKey(row.runId, row.messageId));
820
- if (row.idempotencyKey) {
821
- const dedupKey = `key:${row.idempotencyKey}`;
822
- if (await txn.get(dedupKey) === row.messageId) await txn.delete(dedupKey);
823
- }
1395
+ if (typeof timeoutSeconds === "number") {
1396
+ const reschedule = async (txn) => {
1397
+ if (!await txn.delete(claimKey)) return;
1398
+ const scheduleKey = dueKey(now + timeoutSeconds * 1e3, row.messageId);
1399
+ await txn.put(scheduleKey, row.messageId);
1400
+ if (row.runId) await txn.put(runReferenceKey(row.runId, row.messageId), {
1401
+ messageId: row.messageId,
1402
+ dueKey: scheduleKey,
1403
+ idempotencyKey: row.idempotencyKey
1404
+ });
824
1405
  await this.#scheduleNextAlarm(txn, now);
825
- return;
826
- }
827
- await txn.delete(`inflight:${row.messageId}`);
828
- const scheduleKey = dueKey(now + timeoutSeconds * 1e3, row.messageId);
829
- await txn.put(scheduleKey, row.messageId);
830
- if (row.runId) await txn.put(runReferenceKey(row.runId, row.messageId), {
831
- messageId: row.messageId,
832
- dueKey: scheduleKey,
833
- idempotencyKey: row.idempotencyKey
834
- });
835
- await this.#scheduleNextAlarm(txn, now);
836
- });
837
- else await this.#retry(row, "HTTP 503");
838
- } else if (response && PERMANENT_STATUSES.has(response.status)) await this.#ack(row);
839
- else {
1406
+ };
1407
+ if (row.runId) await this.#withRunMutation(row.runId, (txn) => this.#dropExpiredMessage(txn, row, claimKey, now), reschedule);
1408
+ else await storage.transaction(reschedule);
1409
+ } else await this.#retry(row, claimKey, "HTTP 503");
1410
+ } else if (response && PERMANENT_STATUSES.has(response.status)) {
1411
+ await cancelUnusedResponseBody(response);
1412
+ await this.#ack(row, claimKey);
1413
+ } else {
840
1414
  const reason = response ? `HTTP ${response.status}` : `transport error: ${String(transportError)}`;
841
- await this.#retry(row, reason);
1415
+ if (response) await cancelUnusedResponseBody(response);
1416
+ await this.#retry(row, claimKey, reason);
842
1417
  }
843
1418
  }
844
- async #ack(row) {
1419
+ async #ack(row, claimKey) {
845
1420
  await this.ctx.storage.transaction(async (txn) => {
1421
+ if (!await txn.delete(claimKey)) return;
846
1422
  await txn.delete(`msg:${row.messageId}`);
847
- await txn.delete(`inflight:${row.messageId}`);
848
1423
  if (row.idempotencyKey) {
849
1424
  if (await txn.get(`key:${row.idempotencyKey}`) === row.messageId) await txn.delete(`key:${row.idempotencyKey}`);
850
1425
  }
@@ -852,24 +1427,23 @@ var QueueDO = class extends DurableObject {
852
1427
  await this.#scheduleNextAlarm(txn, this.#now());
853
1428
  });
854
1429
  }
855
- async #retry(row, reason) {
1430
+ async #dropExpiredMessage(txn, row, claimKey, now) {
1431
+ if (!await txn.delete(claimKey)) return;
1432
+ await txn.delete(`msg:${row.messageId}`);
1433
+ if (row.runId) await txn.delete(runReferenceKey(row.runId, row.messageId));
1434
+ if (row.idempotencyKey) {
1435
+ const dedupKey = `key:${row.idempotencyKey}`;
1436
+ if (await txn.get(dedupKey) === row.messageId) await txn.delete(dedupKey);
1437
+ }
1438
+ await this.#scheduleNextAlarm(txn, now);
1439
+ }
1440
+ async #retry(row, claimKey, reason) {
856
1441
  const storage = this.ctx.storage;
857
1442
  const now = this.#now();
858
1443
  const maxAttempts = this.#intVar("QUEUE_MAX_ATTEMPTS", DEFAULT_MAX_ATTEMPTS);
859
1444
  const attempt = row.attempt + 1;
860
- await storage.transaction(async (txn) => {
861
- if (row.runId && await txn.get(expiredRunKey(row.runId)) !== void 0) {
862
- await txn.delete(`msg:${row.messageId}`);
863
- await txn.delete(`inflight:${row.messageId}`);
864
- await txn.delete(runReferenceKey(row.runId, row.messageId));
865
- if (row.idempotencyKey) {
866
- const dedupKey = `key:${row.idempotencyKey}`;
867
- if (await txn.get(dedupKey) === row.messageId) await txn.delete(dedupKey);
868
- }
869
- await this.#scheduleNextAlarm(txn, now);
870
- return;
871
- }
872
- await txn.delete(`inflight:${row.messageId}`);
1445
+ const persistRetry = async (txn) => {
1446
+ if (!await txn.delete(claimKey)) return;
873
1447
  if (attempt >= maxAttempts) {
874
1448
  const dead = {
875
1449
  ...row,
@@ -877,7 +1451,7 @@ var QueueDO = class extends DurableObject {
877
1451
  lastError: reason,
878
1452
  failedAt: now
879
1453
  };
880
- const deadKey = `dlq:${pad(now)}:${row.messageId}`;
1454
+ const deadKey = `dlq:${pad$1(now)}:${row.messageId}`;
881
1455
  await txn.put(deadKey, dead);
882
1456
  await txn.delete(`msg:${row.messageId}`);
883
1457
  if (row.idempotencyKey) {
@@ -905,7 +1479,32 @@ var QueueDO = class extends DurableObject {
905
1479
  idempotencyKey: row.idempotencyKey
906
1480
  });
907
1481
  await this.#scheduleNextAlarm(txn, now);
1482
+ };
1483
+ if (row.runId) await this.#withRunMutation(row.runId, (txn) => this.#dropExpiredMessage(txn, row, claimKey, now), persistRetry);
1484
+ else await storage.transaction(persistRetry);
1485
+ }
1486
+ /** Return true when another due compaction page must run immediately. */
1487
+ async #compactExpiredRunFences(txn, now) {
1488
+ const due = await txn.list({
1489
+ prefix: EXPIRED_RUN_GC_PREFIX,
1490
+ end: `${EXPIRED_RUN_GC_PREFIX}${pad$1(now + 1)}`,
1491
+ limit: 129
908
1492
  });
1493
+ const page = Array.from(due).slice(0, 128);
1494
+ if (page.length === 0) return false;
1495
+ const markerKeys = page.map(([, item]) => expiredRunKey(item.runId));
1496
+ const markers = await getMany$1(txn, markerKeys);
1497
+ const deletes = [];
1498
+ for (let index = 0; index < page.length; index++) {
1499
+ const [gcKey, item] = page[index];
1500
+ const markerKey = markerKeys[index];
1501
+ const marker = markers.get(markerKey);
1502
+ deletes.push(gcKey);
1503
+ if (marker?.compactAt !== item.compactAt) continue;
1504
+ deletes.push(markerKey);
1505
+ }
1506
+ await deleteMany$1(txn, deletes);
1507
+ return due.size > 128;
909
1508
  }
910
1509
  /** Arm the alarm no later than `atMs` (durable backoff never throws). */
911
1510
  async #armAlarmAtMost(storage, atMs, now) {
@@ -929,8 +1528,19 @@ var QueueDO = class extends DurableObject {
929
1528
  limit: 1
930
1529
  });
931
1530
  for (const key of due.keys()) next = Number.parseInt(key.slice(4, 17), 10);
932
- const inflight = await storage.list({ prefix: "inflight:" });
933
- for (const deadline of inflight.values()) if (next === null || deadline < next) next = deadline;
1531
+ const inflight = await storage.list({
1532
+ prefix: INFLIGHT_DEADLINE_PREFIX,
1533
+ limit: 1
1534
+ });
1535
+ for (const key of inflight.keys()) {
1536
+ const deadline = deadlineFromInflightKey(key);
1537
+ if (next === null || deadline < next) next = deadline;
1538
+ }
1539
+ const expiryGc = await storage.list({
1540
+ prefix: EXPIRED_RUN_GC_PREFIX,
1541
+ limit: 1
1542
+ });
1543
+ for (const item of expiryGc.values()) if (next === null || item.compactAt < next) next = item.compactAt;
934
1544
  if (next !== null) await storage.setAlarm(next <= now ? now + MIN_ALARM_DELAY_MS : next);
935
1545
  else await storage.deleteAlarm();
936
1546
  }
@@ -938,7 +1548,7 @@ var QueueDO = class extends DurableObject {
938
1548
  const storage = this.ctx.storage;
939
1549
  const [due, inflight, dlq, alarmAt] = await Promise.all([
940
1550
  storage.list({ prefix: "due:" }),
941
- storage.list({ prefix: "inflight:" }),
1551
+ storage.list({ prefix: INFLIGHT_DEADLINE_PREFIX }),
942
1552
  storage.list({ prefix: "dlq:" }),
943
1553
  storage.getAlarm()
944
1554
  ]);
@@ -969,47 +1579,55 @@ var QueueDO = class extends DurableObject {
969
1579
  async redriveDeadLetter(messageId) {
970
1580
  const storage = this.ctx.storage;
971
1581
  const now = this.#now();
972
- return await storage.transaction(async (txn) => {
973
- const dlq = await txn.list({ prefix: "dlq:" });
974
- for (const [key, dead] of dlq) {
975
- if (dead.messageId !== messageId) continue;
976
- const { failedAt: _failedAt, ...rest } = dead;
977
- const row = {
978
- ...rest,
979
- attempt: 0,
980
- lastError: void 0
981
- };
982
- if (row.idempotencyKey) {
983
- const dedupKey = `key:${row.idempotencyKey}`;
984
- const holder = await txn.get(dedupKey);
985
- if (holder && holder !== row.messageId && await txn.get(`msg:${holder}`)) return { ok: false };
986
- if (holder && holder !== row.messageId) await txn.delete(dedupKey);
987
- await txn.put(dedupKey, row.messageId);
988
- }
989
- await txn.put(`msg:${row.messageId}`, row);
990
- const scheduleKey = dueKey(now, row.messageId);
991
- await txn.put(scheduleKey, row.messageId);
992
- await txn.delete(key);
993
- if (row.runId) await txn.put(runReferenceKey(row.runId, row.messageId), {
994
- messageId: row.messageId,
995
- dueKey: scheduleKey,
996
- idempotencyKey: row.idempotencyKey
997
- });
998
- await this.#armAlarmAtMost(txn, now, now);
999
- return { ok: true };
1582
+ const found = Array.from(await storage.list({ prefix: "dlq:" })).find(([, dead]) => dead.messageId === messageId);
1583
+ if (!found) return { ok: false };
1584
+ const [deadKey, candidate] = found;
1585
+ const redrive = async (txn) => {
1586
+ const dead = await txn.get(deadKey);
1587
+ if (!dead || dead.messageId !== messageId) return { ok: false };
1588
+ const { failedAt: _failedAt, ...rest } = dead;
1589
+ const row = {
1590
+ ...rest,
1591
+ attempt: 0,
1592
+ lastError: void 0
1593
+ };
1594
+ if (row.idempotencyKey) {
1595
+ const dedupKey = `key:${row.idempotencyKey}`;
1596
+ const holder = await txn.get(dedupKey);
1597
+ if (holder && holder !== row.messageId && await txn.get(`msg:${holder}`)) return { ok: false };
1598
+ if (holder && holder !== row.messageId) await txn.delete(dedupKey);
1599
+ await txn.put(dedupKey, row.messageId);
1000
1600
  }
1601
+ await txn.put(`msg:${row.messageId}`, row);
1602
+ const scheduleKey = dueKey(now, row.messageId);
1603
+ await txn.put(scheduleKey, row.messageId);
1604
+ await txn.delete(deadKey);
1605
+ if (row.runId) await txn.put(runReferenceKey(row.runId, row.messageId), {
1606
+ messageId: row.messageId,
1607
+ dueKey: scheduleKey,
1608
+ idempotencyKey: row.idempotencyKey
1609
+ });
1610
+ await this.#armAlarmAtMost(txn, now, now);
1611
+ return { ok: true };
1612
+ };
1613
+ const rejectExpired = async (txn) => {
1614
+ if ((await txn.get(deadKey))?.messageId === messageId) await txn.delete([deadKey, runReferenceKey(candidate.runId, messageId)]);
1001
1615
  return { ok: false };
1002
- });
1616
+ };
1617
+ return candidate.runId ? await this.#withRunMutation(candidate.runId, rejectExpired, redrive) : await storage.transaction(redrive);
1003
1618
  }
1004
1619
  async purgeDeadLetters() {
1005
1620
  return await this.ctx.storage.transaction(async (txn) => {
1006
- const dlq = await txn.list({ prefix: "dlq:" });
1007
- for (const [key, value] of dlq) {
1008
- await txn.delete(key);
1009
- const dead = value;
1010
- if (dead.runId) await txn.delete(runReferenceKey(dead.runId, dead.messageId));
1011
- }
1012
- return { purged: dlq.size };
1621
+ const entries = await txn.list({
1622
+ prefix: "dlq:",
1623
+ limit: 129
1624
+ });
1625
+ const page = Array.from(entries).slice(0, PURGE_DEAD_LETTER_BATCH);
1626
+ await deleteMany$1(txn, page.flatMap(([key, dead]) => [key, ...dead.runId ? [runReferenceKey(dead.runId, dead.messageId)] : []]));
1627
+ return {
1628
+ purged: page.length,
1629
+ hasMore: entries.size > PURGE_DEAD_LETTER_BATCH
1630
+ };
1013
1631
  });
1014
1632
  }
1015
1633
  /**
@@ -1021,140 +1639,450 @@ var QueueDO = class extends DurableObject {
1021
1639
  return { alarmAt: await this.ctx.storage.getAlarm() };
1022
1640
  }
1023
1641
  /** Fence a run and purge every live, inflight, and dead-letter message it owns. */
1024
- async expireRun(runId, expiredAt) {
1025
- return { deleted: await this.ctx.storage.transaction(async (txn) => {
1642
+ async expireRun(runId, expiredAt, options) {
1643
+ const storage = this.ctx.storage;
1644
+ const limit = boundedLimit$1(options?.limit, EXPIRE_RUN_REFERENCE_BATCH, EXPIRE_RUN_REFERENCE_BATCH);
1645
+ return await storage.transaction(async (txn) => {
1026
1646
  const existing = await txn.get(expiredRunKey(runId));
1027
- const references = await txn.list({ prefix: `run:${runId}:` });
1647
+ const references = await txn.list({
1648
+ prefix: `run:${runId}:`,
1649
+ limit: limit + 1
1650
+ });
1651
+ const page = Array.from(references).slice(0, limit);
1652
+ const dedupHolders = await getMany$1(txn, page.flatMap(([, reference]) => reference.idempotencyKey ? [`key:${reference.idempotencyKey}`] : []));
1653
+ const keys = [];
1028
1654
  let count = existing?.deleted ?? 0;
1029
- for (const [key, reference] of references) {
1030
- await txn.delete(`msg:${reference.messageId}`);
1031
- await txn.delete(`inflight:${reference.messageId}`);
1032
- if (reference.dueKey) await txn.delete(reference.dueKey);
1033
- if (reference.dlqKey) await txn.delete(reference.dlqKey);
1655
+ for (const [key, reference] of page) {
1656
+ keys.push(`msg:${reference.messageId}`, key);
1657
+ if (reference.inflightKey) keys.push(reference.inflightKey);
1658
+ if (reference.dueKey) keys.push(reference.dueKey);
1659
+ if (reference.dlqKey) keys.push(reference.dlqKey);
1034
1660
  if (reference.idempotencyKey) {
1035
1661
  const dedupKey = `key:${reference.idempotencyKey}`;
1036
- if (await txn.get(dedupKey) === reference.messageId) await txn.delete(dedupKey);
1662
+ if (dedupHolders.get(dedupKey) === reference.messageId) keys.push(dedupKey);
1037
1663
  }
1038
- await txn.delete(key);
1039
1664
  count++;
1040
1665
  }
1666
+ await deleteMany$1(txn, keys);
1667
+ const done = references.size <= limit;
1668
+ const markerExpiredAt = existing?.expiredAt ?? expiredAt;
1041
1669
  await txn.put(expiredRunKey(runId), {
1042
- expiredAt,
1043
- deleted: count
1670
+ expiredAt: markerExpiredAt,
1671
+ deleted: count,
1672
+ compactAt: existing?.compactAt ?? null
1044
1673
  });
1045
1674
  await this.#scheduleNextAlarm(txn, this.#now());
1046
- return count;
1047
- }) };
1675
+ return done ? {
1676
+ deleted: count,
1677
+ done: true,
1678
+ receipt: {
1679
+ expiredAt: markerExpiredAt,
1680
+ deleted: count
1681
+ }
1682
+ } : {
1683
+ deleted: count,
1684
+ done: false
1685
+ };
1686
+ });
1687
+ }
1688
+ /** Acknowledge a final cleanup receipt before making its marker compactable. */
1689
+ async acknowledgeExpireRun(runId, receipt) {
1690
+ return await this.ctx.storage.transaction(async (txn) => {
1691
+ const key = expiredRunKey(runId);
1692
+ const marker = await txn.get(key);
1693
+ if (!marker) return { acknowledged: true };
1694
+ if (marker.expiredAt !== receipt.expiredAt || marker.deleted !== receipt.deleted) return { acknowledged: false };
1695
+ if (marker.compactAt === null) {
1696
+ const compactAt = this.#now() + QUEUE_FENCE_GRACE_MS;
1697
+ await txn.put(key, {
1698
+ ...marker,
1699
+ compactAt
1700
+ });
1701
+ await txn.put(expiredRunGcKey(runId, compactAt), {
1702
+ runId,
1703
+ compactAt
1704
+ });
1705
+ }
1706
+ await this.#scheduleNextAlarm(txn, this.#now());
1707
+ return { acknowledged: true };
1708
+ });
1709
+ }
1710
+ };
1711
+ //#endregion
1712
+ //#region src/worker/durable-objects/RunCatalogDO.ts
1713
+ const MAX_INDEX_LIST_SIZE = 1001;
1714
+ const EXPIRY_GC_PREFIX = "expiry-gc:";
1715
+ function pad(ms) {
1716
+ return String(Math.max(0, Math.floor(ms))).padStart(13, "0");
1717
+ }
1718
+ function expiredKey(runId) {
1719
+ return `expired:${runId}`;
1720
+ }
1721
+ function expiryGcKey(runId, compactAt) {
1722
+ return `${EXPIRY_GC_PREFIX}${pad(compactAt)}:${encodeURIComponent(runId)}`;
1723
+ }
1724
+ /** One of a fixed set of run-catalog shards, routed by the stable run ID hash. */
1725
+ var RunCatalogDO = class extends DurableObject {
1726
+ now() {
1727
+ const clock = this.env?.clock;
1728
+ return typeof clock === "function" ? clock() : Date.now();
1729
+ }
1730
+ async upsertRun(runId, keys, serializedMetadata, publicationExpiresAt) {
1731
+ if (keys.length !== 2) throw new Error("Run catalog commit requires exactly two keys");
1732
+ return await this.ctx.storage.transaction(async (txn) => {
1733
+ if (await txn.get(expiredKey(runId)) !== void 0) return { stored: false };
1734
+ const now = this.now();
1735
+ if (!Number.isSafeInteger(publicationExpiresAt) || publicationExpiresAt <= now || publicationExpiresAt > now + 1200900) return { stored: false };
1736
+ await txn.put(Object.fromEntries(keys.map((key) => [key, serializedMetadata])));
1737
+ return { stored: true };
1738
+ });
1739
+ }
1740
+ async list(options = {}) {
1741
+ const suppliedLimit = options.limit;
1742
+ const limit = Math.min(MAX_INDEX_LIST_SIZE, Math.max(1, typeof suppliedLimit === "number" && Number.isFinite(suppliedLimit) ? Math.floor(suppliedLimit) : MAX_INDEX_LIST_SIZE));
1743
+ const entries = await this.ctx.storage.list({
1744
+ prefix: options.prefix ?? "",
1745
+ ...options.reverse ? {
1746
+ reverse: true,
1747
+ end: options.cursor
1748
+ } : { startAfter: options.cursor },
1749
+ limit: limit + 1
1750
+ });
1751
+ const page = Array.from(entries).slice(0, limit);
1752
+ const listComplete = entries.size <= limit;
1753
+ return {
1754
+ keys: page.map(([name, value]) => ({
1755
+ name,
1756
+ value
1757
+ })),
1758
+ list_complete: listComplete,
1759
+ cursor: listComplete ? void 0 : page.at(-1)?.[0]
1760
+ };
1761
+ }
1762
+ async expireRun(runId, keys, expiredAt) {
1763
+ if (keys.length !== 2) throw new Error("Run catalog expiry requires exactly two keys");
1764
+ return await this.ctx.storage.transaction(async (txn) => {
1765
+ const marker = expiredKey(runId);
1766
+ const compactAt = (await txn.get(marker))?.compactAt ?? this.now() + 1200900;
1767
+ await txn.put(marker, {
1768
+ expiredAt,
1769
+ compactAt
1770
+ });
1771
+ await txn.put(expiryGcKey(runId, compactAt), {
1772
+ runId,
1773
+ compactAt
1774
+ });
1775
+ await this.armAtMost(txn, compactAt);
1776
+ return { deleted: await txn.delete(keys) };
1777
+ });
1778
+ }
1779
+ async alarm() {
1780
+ try {
1781
+ await this.compactExpiryFences();
1782
+ } catch {
1783
+ await this.ctx.storage.setAlarm(this.now() + LIFECYCLE_COMPACTION_RETRY_MS);
1784
+ }
1785
+ }
1786
+ async armAtMost(storage, at) {
1787
+ const current = await storage.getAlarm();
1788
+ if (current === null || at < current) await storage.setAlarm(at);
1789
+ }
1790
+ async compactExpiryFences() {
1791
+ const now = this.now();
1792
+ await this.ctx.storage.transaction(async (txn) => {
1793
+ const due = await txn.list({
1794
+ prefix: EXPIRY_GC_PREFIX,
1795
+ end: `${EXPIRY_GC_PREFIX}${pad(now + 1)}`,
1796
+ limit: 129
1797
+ });
1798
+ const page = Array.from(due).slice(0, 128);
1799
+ const markerKeys = page.map(([, item]) => expiredKey(item.runId));
1800
+ const markers = await txn.get(markerKeys);
1801
+ const deletes = page.flatMap(([key, item], index) => {
1802
+ return markers.get(markerKeys[index])?.compactAt === item.compactAt ? [key, markerKeys[index]] : [key];
1803
+ });
1804
+ for (let offset = 0; offset < deletes.length; offset += 128) await txn.delete(deletes.slice(offset, offset + 128));
1805
+ if (due.size > 128) {
1806
+ await txn.setAlarm(now + 1);
1807
+ return;
1808
+ }
1809
+ const nextAt = (await txn.list({
1810
+ prefix: EXPIRY_GC_PREFIX,
1811
+ limit: 1
1812
+ })).values().next().value?.compactAt;
1813
+ if (nextAt === void 0) await txn.deleteAlarm();
1814
+ else await txn.setAlarm(Math.max(now + 1, nextAt));
1815
+ });
1048
1816
  }
1049
1817
  };
1050
1818
  //#endregion
1051
1819
  //#region src/worker/durable-objects/StreamDO.ts
1052
1820
  const META_KEY = "meta";
1053
1821
  const CHUNK_KEY_PREFIX = "chunk:";
1822
+ const CHUNK_SIZE_KEY_PREFIX = "chunk-size:";
1054
1823
  /** Registry keys used when this DO instance acts as a per-run stream index. */
1055
1824
  const STREAM_REGISTRY_PREFIX = "stream:";
1056
1825
  const REGISTRY_OWNER_KEY = "registry:owner";
1057
1826
  const REGISTRY_EXPIRED_KEY = "registry:expired";
1058
- /**
1059
- * Zero-pad chunk indexes so `storage.list({ prefix })` returns chunks in
1060
- * write order. 12 digits comfortably exceeds any realistic chunk count.
1061
- */
1827
+ /** Each chunk occupies a payload key and a size key; one delete accepts at most 128 keys. */
1828
+ const DEFAULT_EXPIRE_CHUNK_LIMIT = 64;
1829
+ const MAX_EXPIRE_CHUNK_LIMIT = 64;
1830
+ /** Bound storage deletion work as well as item count; one oversized chunk still makes progress. */
1831
+ const DEFAULT_EXPIRE_BYTE_LIMIT = 16777216;
1832
+ const MAX_EXPIRE_BYTE_LIMIT = DEFAULT_EXPIRE_BYTE_LIMIT;
1833
+ function boundedLimit(value, fallback, maximum) {
1834
+ if (value === void 0 || !Number.isFinite(value)) return fallback;
1835
+ return Math.max(1, Math.min(maximum, Math.floor(value)));
1836
+ }
1837
+ function emptyMeta() {
1838
+ return {
1839
+ count: 0,
1840
+ state: "open"
1841
+ };
1842
+ }
1843
+ function validateMeta(meta) {
1844
+ if (!Number.isSafeInteger(meta.count) || meta.count < 0 || meta.count > 2147483647 || ![
1845
+ "open",
1846
+ "closed",
1847
+ "errored",
1848
+ "expired"
1849
+ ].includes(meta.state)) throw new Error("Invalid persisted stream metadata");
1850
+ if (meta.state === "errored" && !meta.error) throw new Error("Invalid persisted stream error metadata");
1851
+ return meta;
1852
+ }
1853
+ /** Zero-padding keeps storage.list() results in stream offset order. */
1062
1854
  function chunkKey(index) {
1063
1855
  return `${CHUNK_KEY_PREFIX}${index.toString().padStart(12, "0")}`;
1064
1856
  }
1857
+ function chunkSizeKey(index) {
1858
+ return `${CHUNK_SIZE_KEY_PREFIX}${index.toString().padStart(12, "0")}`;
1859
+ }
1860
+ function indexFromChunkSizeKey(key) {
1861
+ const suffix = key.slice(11);
1862
+ const index = Number(suffix);
1863
+ if (!/^\d{12}$/.test(suffix) || !Number.isSafeInteger(index) || index < 0) throw new Error(`Invalid persisted stream chunk size key "${key}"`);
1864
+ return index;
1865
+ }
1866
+ function validateChunkSize(size, index) {
1867
+ if (typeof size !== "number" || !Number.isSafeInteger(size) || size < 0 || size > 1048576) throw new Error(`Invalid persisted stream chunk size at index ${index}`);
1868
+ return size;
1869
+ }
1870
+ function compactChunk(chunk) {
1871
+ return chunk.byteOffset === 0 && chunk.byteLength === chunk.buffer.byteLength ? chunk : new Uint8Array(chunk);
1872
+ }
1873
+ function abortReason(signal) {
1874
+ return signal.reason ?? new DOMException("The operation was aborted", "AbortError");
1875
+ }
1065
1876
  /**
1066
1877
  * Durable Object backing workflow streams.
1067
1878
  *
1068
- * Two roles, selected by the DO name:
1069
- * - `stream:<name>`: chunk storage for a single stream. A monotonic
1070
- * per-stream counter (maintained transactionally inside the DO) assigns
1071
- * each chunk its index; one chunk per storage key.
1072
- * - `run-streams:<runId>`: registry of stream names owned by a run,
1073
- * powering `listStreamsByRunId`.
1879
+ * Stream cells persist binary chunks under monotonic offset keys. Public
1880
+ * reads are bounded long polls, and writes append a bounded ordered batch in
1881
+ * one storage transaction. Per-run registry cells retain their existing role.
1074
1882
  */
1075
1883
  var StreamDO = class extends DurableObject {
1884
+ meta;
1885
+ metaLoad;
1886
+ mutationTail = Promise.resolve();
1887
+ changeVersion = 0;
1888
+ waiters = /* @__PURE__ */ new Set();
1076
1889
  async getMeta() {
1077
- return await this.ctx.storage.get(META_KEY) ?? {
1078
- count: 0,
1079
- closed: false
1080
- };
1890
+ if (this.meta) return this.meta;
1891
+ this.metaLoad ??= this.ctx.storage.get(META_KEY).then((meta) => {
1892
+ const loaded = meta ? validateMeta(meta) : emptyMeta();
1893
+ this.meta = loaded;
1894
+ return loaded;
1895
+ });
1896
+ try {
1897
+ return await this.metaLoad;
1898
+ } catch (error) {
1899
+ this.metaLoad = void 0;
1900
+ throw error;
1901
+ }
1081
1902
  }
1082
1903
  assertOwner(meta, runId) {
1083
- if (meta.expiredAt !== void 0) throw new Error(`Workflow run "${runId}" has expired`);
1904
+ if (!runId) throw new Error("Stream runId is required");
1084
1905
  if (meta.ownerRunId !== void 0 && meta.ownerRunId !== runId) throw new Error(`Stream is owned by workflow run "${meta.ownerRunId}"`);
1085
1906
  }
1086
- /**
1087
- * Append a chunk. The index is allocated inside the transaction, so
1088
- * concurrent writers can never collide or skip.
1089
- */
1090
- async writeChunk(runId, data) {
1091
- return await this.ctx.storage.transaction(async (txn) => {
1092
- const meta = await txn.get(META_KEY) ?? {
1093
- count: 0,
1094
- closed: false
1095
- };
1096
- this.assertOwner(meta, runId);
1097
- if (meta.closed) throw new Error("Cannot write to a closed stream");
1098
- const index = meta.count;
1099
- await txn.put(chunkKey(index), data);
1100
- await txn.put(META_KEY, {
1101
- count: index + 1,
1102
- closed: false,
1103
- ownerRunId: runId
1907
+ assertWritable(meta, runId) {
1908
+ this.assertOwner(meta, runId);
1909
+ if (meta.state === "expired") throw new Error(`Workflow run "${runId}" has expired`);
1910
+ if (meta.state === "closed") throw new Error("Cannot write to a closed stream");
1911
+ if (meta.state === "errored") throw new Error("Cannot write to an errored stream");
1912
+ }
1913
+ runMutation(operation) {
1914
+ const result = this.mutationTail.then(operation, operation);
1915
+ this.mutationTail = result.then(() => void 0, () => void 0);
1916
+ return result;
1917
+ }
1918
+ wakeReaders() {
1919
+ this.changeVersion += 1;
1920
+ for (const waiter of Array.from(this.waiters)) waiter.resolve("change");
1921
+ }
1922
+ waitForChange(observedVersion, waitMs, signal) {
1923
+ if (signal?.aborted) return Promise.reject(abortReason(signal));
1924
+ return new Promise((resolve, reject) => {
1925
+ let settled = false;
1926
+ const waiter = {
1927
+ timer: setTimeout(() => finish("timeout"), waitMs),
1928
+ signal,
1929
+ resolve: (reason) => finish(reason)
1930
+ };
1931
+ const cleanup = () => {
1932
+ clearTimeout(waiter.timer);
1933
+ this.waiters.delete(waiter);
1934
+ if (waiter.signal && waiter.onAbort) waiter.signal.removeEventListener("abort", waiter.onAbort);
1935
+ };
1936
+ const finish = (reason) => {
1937
+ if (settled) return;
1938
+ settled = true;
1939
+ cleanup();
1940
+ resolve(reason);
1941
+ };
1942
+ waiter.onAbort = () => {
1943
+ if (settled) return;
1944
+ settled = true;
1945
+ cleanup();
1946
+ reject(abortReason(signal));
1947
+ };
1948
+ this.waiters.add(waiter);
1949
+ signal?.addEventListener("abort", waiter.onAbort, { once: true });
1950
+ if (this.changeVersion !== observedVersion) finish("change");
1951
+ });
1952
+ }
1953
+ async readSnapshot(request, timedOut) {
1954
+ const meta = await this.getMeta();
1955
+ this.assertOwner(meta, request.runId);
1956
+ const available = meta.state === "expired" ? 0 : Math.max(0, Math.min(request.maxChunks, meta.count - request.startIndex));
1957
+ const chunks = [];
1958
+ if (available > 0) {
1959
+ const indexes = Array.from({ length: available }, (_, offset) => request.startIndex + offset);
1960
+ const sizeKeys = indexes.map(chunkSizeKey);
1961
+ const storedSizes = await this.ctx.storage.get(sizeKeys);
1962
+ const selectedIndexes = [];
1963
+ const selectedSizes = [];
1964
+ let bytes = 0;
1965
+ for (const index of indexes) {
1966
+ const size = validateChunkSize(storedSizes.get(chunkSizeKey(index)), index);
1967
+ if (bytes + size > request.maxBytes) break;
1968
+ selectedIndexes.push(index);
1969
+ selectedSizes.push(size);
1970
+ bytes += size;
1971
+ }
1972
+ const payloadKeys = selectedIndexes.map(chunkKey);
1973
+ const storedChunks = await this.ctx.storage.get(payloadKeys);
1974
+ for (let offset = 0; offset < payloadKeys.length; offset++) {
1975
+ const chunk = storedChunks.get(payloadKeys[offset]);
1976
+ if (!(chunk instanceof Uint8Array) || chunk.byteLength !== selectedSizes[offset]) throw new Error(`Invalid persisted stream chunk at index ${selectedIndexes[offset]}`);
1977
+ chunks.push(chunk);
1978
+ }
1979
+ }
1980
+ return {
1981
+ startIndex: request.startIndex,
1982
+ tailIndex: meta.count - 1,
1983
+ chunks,
1984
+ state: meta.state,
1985
+ timedOut,
1986
+ error: meta.error
1987
+ };
1988
+ }
1989
+ /** Append one bounded binary batch with contiguous indexes. */
1990
+ async writeChunks(runId, chunks) {
1991
+ validateStreamWriteChunks(chunks);
1992
+ const storedChunks = chunks.map(compactChunk);
1993
+ return await this.runMutation(async () => {
1994
+ const committed = await this.ctx.storage.transaction(async (txn) => {
1995
+ const stored = await txn.get(META_KEY);
1996
+ const meta = stored ? validateMeta(stored) : emptyMeta();
1997
+ this.assertWritable(meta, runId);
1998
+ const startIndex = meta.count;
1999
+ if (startIndex + chunks.length > 2147483647) throw new Error("Stream offset limit exceeded");
2000
+ const nextMeta = {
2001
+ count: startIndex + chunks.length,
2002
+ state: "open",
2003
+ ownerRunId: runId
2004
+ };
2005
+ const entries = { [META_KEY]: nextMeta };
2006
+ for (let offset = 0; offset < storedChunks.length; offset++) {
2007
+ const index = startIndex + offset;
2008
+ entries[chunkKey(index)] = storedChunks[offset];
2009
+ entries[chunkSizeKey(index)] = storedChunks[offset].byteLength;
2010
+ }
2011
+ await txn.put(entries);
2012
+ return {
2013
+ meta: nextMeta,
2014
+ result: {
2015
+ startIndex,
2016
+ count: chunks.length,
2017
+ tailIndex: nextMeta.count - 1
2018
+ }
2019
+ };
1104
2020
  });
1105
- return index;
2021
+ this.meta = committed.meta;
2022
+ this.metaLoad = Promise.resolve(committed.meta);
2023
+ this.wakeReaders();
2024
+ return committed.result;
1106
2025
  });
1107
2026
  }
1108
2027
  /**
1109
- * Close the stream (idempotent). Readers observe `done: true` once all
1110
- * previously written chunks have been consumed.
2028
+ * Return available binary chunks and terminal metadata in one bounded
2029
+ * operation, waiting at most waitMs when the requested offset is idle.
1111
2030
  */
2031
+ async readChunks(request, signal) {
2032
+ validateStreamReadRequest(request);
2033
+ const observedVersion = this.changeVersion;
2034
+ const immediate = await this.readSnapshot(request, false);
2035
+ if (immediate.chunks.length > 0 || immediate.state !== "open" || request.waitMs === 0 || request.maxChunks === 0) return immediate;
2036
+ const reason = await this.waitForChange(observedVersion, request.waitMs, signal);
2037
+ return await this.readSnapshot(request, reason === "timeout");
2038
+ }
2039
+ /** Close idempotently; the first terminal state wins. */
1112
2040
  async closeStream(runId) {
1113
- await this.ctx.storage.transaction(async (txn) => {
1114
- const meta = await txn.get(META_KEY) ?? {
1115
- count: 0,
1116
- closed: false
1117
- };
1118
- this.assertOwner(meta, runId);
1119
- await txn.put(META_KEY, {
1120
- ...meta,
1121
- ownerRunId: runId,
1122
- closed: true
2041
+ await this.runMutation(async () => {
2042
+ const nextMeta = await this.ctx.storage.transaction(async (txn) => {
2043
+ const stored = await txn.get(META_KEY);
2044
+ const meta = stored ? validateMeta(stored) : emptyMeta();
2045
+ this.assertOwner(meta, runId);
2046
+ if (meta.state === "expired") throw new Error(`Workflow run "${runId}" has expired`);
2047
+ if (meta.state !== "open") return meta;
2048
+ const closed = {
2049
+ ...meta,
2050
+ ownerRunId: runId,
2051
+ state: "closed"
2052
+ };
2053
+ await txn.put(META_KEY, closed);
2054
+ return closed;
1123
2055
  });
2056
+ this.meta = nextMeta;
2057
+ this.metaLoad = Promise.resolve(nextMeta);
2058
+ this.wakeReaders();
1124
2059
  });
1125
2060
  }
1126
- /**
1127
- * Read up to `limit` chunks starting at `startIndex` (0-based, inclusive).
1128
- */
1129
- async getChunks(params) {
1130
- const meta = await this.getMeta();
1131
- if (meta.expiredAt !== void 0) throw new Error("Stream has expired");
1132
- const start = Math.max(0, params.startIndex);
1133
- const entries = await this.ctx.storage.list({
1134
- prefix: CHUNK_KEY_PREFIX,
1135
- start: chunkKey(start),
1136
- limit: params.limit
2061
+ /** Persist a terminal stream failure and wake every pending reader. */
2062
+ async failStream(runId, error) {
2063
+ const normalized = normalizeStreamError(error);
2064
+ await this.runMutation(async () => {
2065
+ const nextMeta = await this.ctx.storage.transaction(async (txn) => {
2066
+ const stored = await txn.get(META_KEY);
2067
+ const meta = stored ? validateMeta(stored) : emptyMeta();
2068
+ this.assertOwner(meta, runId);
2069
+ if (meta.state === "expired") throw new Error(`Workflow run "${runId}" has expired`);
2070
+ if (meta.state !== "open") return meta;
2071
+ const failed = {
2072
+ ...meta,
2073
+ ownerRunId: runId,
2074
+ state: "errored",
2075
+ error: normalized
2076
+ };
2077
+ await txn.put(META_KEY, failed);
2078
+ return failed;
2079
+ });
2080
+ this.meta = nextMeta;
2081
+ this.metaLoad = Promise.resolve(nextMeta);
2082
+ this.wakeReaders();
1137
2083
  });
1138
- return {
1139
- chunks: Array.from(entries.values()),
1140
- done: meta.closed,
1141
- tailIndex: meta.count - 1
1142
- };
1143
- }
1144
- /**
1145
- * Lightweight stream metadata: last chunk index and completion flag.
1146
- */
1147
- async getInfo() {
1148
- const meta = await this.getMeta();
1149
- if (meta.expiredAt !== void 0) throw new Error("Stream has expired");
1150
- return {
1151
- tailIndex: meta.count - 1,
1152
- done: meta.closed
1153
- };
1154
2084
  }
1155
- /**
1156
- * Register a stream name against this per-run registry instance.
1157
- */
2085
+ /** Register a stream name against this per-run registry instance. */
1158
2086
  async registerStream(runId, name) {
1159
2087
  await this.ctx.storage.transaction(async (txn) => {
1160
2088
  if (await txn.get(REGISTRY_EXPIRED_KEY) !== void 0) throw new Error(`Workflow run "${runId}" has expired`);
@@ -1164,57 +2092,124 @@ var StreamDO = class extends DurableObject {
1164
2092
  await txn.put(`${STREAM_REGISTRY_PREFIX}${name}`, true);
1165
2093
  });
1166
2094
  }
1167
- /**
1168
- * List stream names registered against this per-run registry instance.
1169
- */
1170
2095
  async listStreams() {
1171
2096
  if (await this.ctx.storage.get(REGISTRY_EXPIRED_KEY) !== void 0) throw new Error("Workflow run streams have expired");
1172
2097
  const entries = await this.ctx.storage.list({ prefix: STREAM_REGISTRY_PREFIX });
1173
2098
  return Array.from(entries.keys()).map((key) => key.slice(7));
1174
2099
  }
1175
2100
  /** Fence a run's registry before any stream cells are removed. */
1176
- async expireRegistry(runId, expiredAt) {
2101
+ async expireRegistry(runId, expiredAt, options) {
2102
+ const limit = boundedLimit(options?.limit, 16, 128);
1177
2103
  return await this.ctx.storage.transaction(async (txn) => {
1178
- const owner = await txn.get(REGISTRY_OWNER_KEY);
2104
+ const state = await txn.get([REGISTRY_OWNER_KEY, REGISTRY_EXPIRED_KEY]);
2105
+ const owner = state.get(REGISTRY_OWNER_KEY);
1179
2106
  if (owner !== void 0 && owner !== runId) throw new Error(`Stream registry is owned by workflow run "${owner}"`);
1180
- await txn.put(REGISTRY_OWNER_KEY, runId);
1181
- await txn.put(REGISTRY_EXPIRED_KEY, expiredAt);
1182
- const entries = await txn.list({ prefix: STREAM_REGISTRY_PREFIX });
2107
+ const expiry = state.get(REGISTRY_EXPIRED_KEY);
2108
+ await txn.put({
2109
+ [REGISTRY_OWNER_KEY]: runId,
2110
+ [REGISTRY_EXPIRED_KEY]: expiry ?? {
2111
+ expiredAt,
2112
+ deleted: 0
2113
+ }
2114
+ });
2115
+ const entries = await txn.list({
2116
+ prefix: STREAM_REGISTRY_PREFIX,
2117
+ limit
2118
+ });
1183
2119
  return { streams: Array.from(entries.keys()).map((key) => key.slice(7)) };
1184
2120
  });
1185
2121
  }
1186
- /** Remove the registry payload after every referenced stream was fenced. */
1187
- async finalizeRegistry(runId) {
1188
- await this.ctx.storage.transaction(async (txn) => {
1189
- const owner = await txn.get(REGISTRY_OWNER_KEY);
1190
- const expiredAt = await txn.get(REGISTRY_EXPIRED_KEY);
1191
- if (owner !== runId || expiredAt === void 0) throw new Error(`Stream registry for workflow run "${runId}" is not expired`);
1192
- const entries = await txn.list({ prefix: STREAM_REGISTRY_PREFIX });
1193
- for (const key of entries.keys()) await txn.delete(key);
1194
- });
1195
- }
1196
- /** Delete stream payload while leaving a small ownership tombstone. */
1197
- async expireStream(runId, expiredAt) {
2122
+ /** Remove completed registry entries in a bounded, retry-safe batch. */
2123
+ async finalizeRegistry(runId, streams) {
1198
2124
  return await this.ctx.storage.transaction(async (txn) => {
1199
- const meta = await txn.get(META_KEY) ?? {
1200
- count: 0,
1201
- closed: false
1202
- };
1203
- if (meta.ownerRunId !== void 0 && meta.ownerRunId !== runId) throw new Error(`Stream is owned by workflow run "${meta.ownerRunId}"`);
1204
- const chunks = await txn.list({ prefix: CHUNK_KEY_PREFIX });
1205
- for (const key of chunks.keys()) await txn.delete(key);
1206
- await txn.put(META_KEY, {
1207
- count: 0,
1208
- closed: true,
1209
- ownerRunId: runId,
1210
- expiredAt
2125
+ const state = await txn.get([REGISTRY_OWNER_KEY, REGISTRY_EXPIRED_KEY]);
2126
+ const owner = state.get(REGISTRY_OWNER_KEY);
2127
+ const expiry = state.get(REGISTRY_EXPIRED_KEY);
2128
+ if (owner !== runId || expiry === void 0) throw new Error(`Stream registry for workflow run "${runId}" is not expired`);
2129
+ const keys = Array.from(new Set(streams.map((name) => `${STREAM_REGISTRY_PREFIX}${name}`)));
2130
+ const removed = keys.length === 0 ? 0 : await txn.delete(keys);
2131
+ const deleted = expiry.deleted + removed;
2132
+ await txn.put(REGISTRY_EXPIRED_KEY, {
2133
+ ...expiry,
2134
+ deleted
1211
2135
  });
1212
2136
  return {
1213
- deleted: true,
1214
- chunks: chunks.size
2137
+ deleted,
2138
+ done: (await txn.list({
2139
+ prefix: STREAM_REGISTRY_PREFIX,
2140
+ limit: 1
2141
+ })).size === 0
1215
2142
  };
1216
2143
  });
1217
2144
  }
2145
+ /**
2146
+ * Fence the stream and delete one bounded page. Listing size keys avoids
2147
+ * loading payloads, and the transaction makes a crash retry resume safely.
2148
+ */
2149
+ async expireStream(runId, expiredAt, options) {
2150
+ const limit = boundedLimit(options?.limit, DEFAULT_EXPIRE_CHUNK_LIMIT, MAX_EXPIRE_CHUNK_LIMIT);
2151
+ const byteLimit = boundedLimit(options?.byteLimit, DEFAULT_EXPIRE_BYTE_LIMIT, MAX_EXPIRE_BYTE_LIMIT);
2152
+ return await this.runMutation(async () => {
2153
+ const result = await this.ctx.storage.transaction(async (txn) => {
2154
+ const stored = await txn.get(META_KEY);
2155
+ const meta = stored ? validateMeta(stored) : emptyMeta();
2156
+ this.assertOwner(meta, runId);
2157
+ const firstFence = meta.state !== "expired";
2158
+ if (meta.state === "expired" && meta.payloadDeleted) return {
2159
+ meta,
2160
+ result: {
2161
+ deleted: false,
2162
+ chunks: 0,
2163
+ bytes: 0,
2164
+ done: true
2165
+ }
2166
+ };
2167
+ const candidates = await txn.list({
2168
+ prefix: CHUNK_SIZE_KEY_PREFIX,
2169
+ limit: limit + 1
2170
+ });
2171
+ const page = [];
2172
+ let bytes = 0;
2173
+ for (const [key, storedSize] of Array.from(candidates).slice(0, limit)) {
2174
+ const index = indexFromChunkSizeKey(key);
2175
+ const size = validateChunkSize(storedSize, index);
2176
+ if (page.length > 0 && bytes + size > byteLimit) break;
2177
+ page.push({
2178
+ index,
2179
+ size
2180
+ });
2181
+ bytes += size;
2182
+ }
2183
+ const keys = page.flatMap(({ index }) => [chunkKey(index), chunkSizeKey(index)]);
2184
+ if (keys.length > 0) await txn.delete(keys);
2185
+ const done = candidates.size === page.length;
2186
+ const nextMeta = {
2187
+ ...meta,
2188
+ count: done ? 0 : meta.count,
2189
+ ownerRunId: runId,
2190
+ state: "expired",
2191
+ error: void 0,
2192
+ expiredAt: meta.expiredAt ?? expiredAt,
2193
+ expiredChunkCount: meta.expiredChunkCount ?? meta.count,
2194
+ payloadDeleted: done
2195
+ };
2196
+ await txn.put(META_KEY, nextMeta);
2197
+ return {
2198
+ meta: nextMeta,
2199
+ result: {
2200
+ deleted: firstFence,
2201
+ chunks: page.length,
2202
+ bytes,
2203
+ done
2204
+ }
2205
+ };
2206
+ });
2207
+ this.meta = result.meta;
2208
+ this.metaLoad = Promise.resolve(result.meta);
2209
+ this.wakeReaders();
2210
+ return result.result;
2211
+ });
2212
+ }
1218
2213
  };
1219
2214
  //#endregion
1220
2215
  //#region src/apply-event.ts
@@ -1234,10 +2229,36 @@ const HOOK_CREATED_KEY_PREFIX = "hookcreated:";
1234
2229
  * rejected as a duplicate instead of resurrecting the hook.
1235
2230
  */
1236
2231
  const HOOK_EVENT_MARKER_PREFIX = "hookevent:";
2232
+ /** Durable repair record for idempotent post-commit hook-index release. */
2233
+ const HOOK_RELEASE_MARKER_PREFIX = "hookreleased:";
2234
+ /** Exact ambiguous reservations canceled by a serialized RunDO resolution. */
2235
+ const HOOK_CLAIM_CANCELLATION_PREFIX = "hookclaimcancelled:";
1237
2236
  /** Workflow-authored attr_set correlation claims. */
1238
2237
  const ATTRIBUTE_EVENT_MARKER_PREFIX = "attrevent:";
1239
2238
  /** Durable lazy-hook resume claims, scoped by the containing run DO. */
1240
2239
  const HOOK_RESUME_CLAIM_PREFIX = "hookresume:";
2240
+ const STORAGE_BATCH_SIZE$1 = 128;
2241
+ const MAX_ENTITY_PAGE_SIZE = 1e3;
2242
+ async function getMany(store, keys) {
2243
+ if (store.getMany) {
2244
+ const result = /* @__PURE__ */ new Map();
2245
+ for (let offset = 0; offset < keys.length; offset += STORAGE_BATCH_SIZE$1) {
2246
+ const page = await store.getMany(keys.slice(offset, offset + STORAGE_BATCH_SIZE$1));
2247
+ for (const entry of page) result.set(...entry);
2248
+ }
2249
+ return result;
2250
+ }
2251
+ const values = await Promise.all(keys.map((key) => store.get(key)));
2252
+ return new Map(keys.flatMap((key, index) => values[index] === void 0 ? [] : [[key, values[index]]]));
2253
+ }
2254
+ async function deleteMany(store, keys) {
2255
+ if (store.deleteMany) {
2256
+ let deleted = 0;
2257
+ for (let offset = 0; offset < keys.length; offset += STORAGE_BATCH_SIZE$1) deleted += await store.deleteMany(keys.slice(offset, offset + STORAGE_BATCH_SIZE$1));
2258
+ return deleted;
2259
+ }
2260
+ return (await Promise.all(keys.map((key) => store.delete(key)))).filter(Boolean).length;
2261
+ }
1241
2262
  function failure(code, message, extra) {
1242
2263
  return {
1243
2264
  ok: false,
@@ -1246,6 +2267,9 @@ function failure(code, message, extra) {
1246
2267
  ...extra
1247
2268
  };
1248
2269
  }
2270
+ function hookClaimCancellationKey(claimId) {
2271
+ return `${HOOK_CLAIM_CANCELLATION_PREFIX}${encodeURIComponent(claimId)}`;
2272
+ }
1249
2273
  function readStringProp(value, key) {
1250
2274
  if (value !== null && typeof value === "object") {
1251
2275
  const prop = value[key];
@@ -1301,7 +2325,9 @@ function creationIndexKey(prefix, createdAt, id) {
1301
2325
  * requested ordering.
1302
2326
  */
1303
2327
  async function listByCreationTime(store, indexPrefix, entityPrefix, params) {
1304
- const { limit, cursor, sortOrder = "asc" } = params;
2328
+ const requestedLimit = Number.isFinite(params.limit) ? Math.floor(params.limit) : 1;
2329
+ const limit = Math.min(MAX_ENTITY_PAGE_SIZE, Math.max(1, requestedLimit));
2330
+ const { cursor, sortOrder = "asc" } = params;
1305
2331
  const entries = await store.list(sortOrder === "desc" ? {
1306
2332
  prefix: indexPrefix,
1307
2333
  reverse: true,
@@ -1313,11 +2339,12 @@ async function listByCreationTime(store, indexPrefix, entityPrefix, params) {
1313
2339
  startAfter: cursor ? `${indexPrefix}${cursor}` : void 0
1314
2340
  });
1315
2341
  const page = Array.from(entries.entries()).slice(0, limit);
1316
- const data = [];
1317
- for (const [, id] of page) {
1318
- const item = await store.get(`${entityPrefix}${id}`);
1319
- if (item !== void 0) data.push(item);
1320
- }
2342
+ const entityKeys = page.map(([, id]) => `${entityPrefix}${id}`);
2343
+ const entities = await getMany(store, entityKeys);
2344
+ const data = entityKeys.flatMap((key) => {
2345
+ const item = entities.get(key);
2346
+ return item === void 0 ? [] : [item];
2347
+ });
1321
2348
  const hasMore = entries.size > limit;
1322
2349
  const lastKey = page.at(-1)?.[0];
1323
2350
  return {
@@ -1328,22 +2355,26 @@ async function listByCreationTime(store, indexPrefix, entityPrefix, params) {
1328
2355
  }
1329
2356
  /** Delete all hook entities for the run, returning the released tokens. */
1330
2357
  async function releaseAllHooks(store) {
2358
+ if (store.deferTerminalCleanup) return [];
1331
2359
  const hooks = await store.list({ prefix: HOOK_KEY_PREFIX });
1332
2360
  const released = [];
2361
+ const keys = [];
1333
2362
  for (const hook of hooks.values()) {
1334
- await store.delete(`${HOOK_KEY_PREFIX}${hook.hookId}`);
1335
- await store.delete(creationIndexKey(HOOK_CREATED_KEY_PREFIX, hook.createdAt, hook.hookId));
2363
+ keys.push(`${HOOK_KEY_PREFIX}${hook.hookId}`);
2364
+ keys.push(creationIndexKey(HOOK_CREATED_KEY_PREFIX, hook.createdAt, hook.hookId));
1336
2365
  released.push({
1337
2366
  hookId: hook.hookId,
1338
2367
  token: hook.token
1339
2368
  });
1340
2369
  }
2370
+ await deleteMany(store, keys);
1341
2371
  return released;
1342
2372
  }
1343
2373
  /** Delete all persisted waits once their owning run becomes terminal. */
1344
2374
  async function releaseAllWaits(store) {
2375
+ if (store.deferTerminalCleanup) return;
1345
2376
  const waits = await store.list({ prefix: WAIT_KEY_PREFIX });
1346
- for (const key of waits.keys()) await store.delete(key);
2377
+ await deleteMany(store, Array.from(waits.keys()));
1347
2378
  }
1348
2379
  async function applyEvent(store, ctx) {
1349
2380
  const { runId, data, now } = ctx;
@@ -1463,7 +2494,16 @@ async function applyEvent(store, ctx) {
1463
2494
  }
1464
2495
  }
1465
2496
  if ((data.eventType === "hook_disposed" || data.eventType === "hook_received") && data.correlationId) {
1466
- if (!await store.get(`hook:${data.correlationId}`)) return failure("HOOK_NOT_FOUND", `Hook "${data.correlationId}" not found`);
2497
+ if (!await store.get(`hook:${data.correlationId}`)) {
2498
+ if (data.eventType === "hook_disposed") {
2499
+ const released = await store.get(`${HOOK_RELEASE_MARKER_PREFIX}${data.correlationId}`);
2500
+ if (released) return {
2501
+ ok: true,
2502
+ releasedHooks: [released]
2503
+ };
2504
+ }
2505
+ return failure("HOOK_NOT_FOUND", `Hook "${data.correlationId}" not found`);
2506
+ }
1467
2507
  }
1468
2508
  if (!skipRunValidation && data.eventType !== "run_started" && !currentRun) return failure("RUN_NOT_FOUND", `Workflow run "${runId}" not found`);
1469
2509
  switch (data.eventType) {
@@ -1799,6 +2839,7 @@ async function applyEvent(store, ctx) {
1799
2839
  const holder = ctx.tokenHolder ?? null;
1800
2840
  const markerKey = `${HOOK_EVENT_MARKER_PREFIX}${hookId}`;
1801
2841
  const hookKey = `${HOOK_KEY_PREFIX}${hookId}`;
2842
+ if (ctx.hookClaimId && await store.get(hookClaimCancellationKey(ctx.hookClaimId)) !== void 0) return failure("HOOK_CLAIM_CANCELLED", `Hook reservation for "${hookId}" was canceled before commit`);
1802
2843
  if (holder && (holder.runId !== runId || holder.hookId !== hookId)) {
1803
2844
  const event = buildEvent({
1804
2845
  eventType: "hook_conflict",
@@ -1858,6 +2899,10 @@ async function applyEvent(store, ctx) {
1858
2899
  if (!hook) return failure("HOOK_NOT_FOUND", `Hook "${data.correlationId}" not found`);
1859
2900
  await store.delete(hookKey);
1860
2901
  await store.delete(creationIndexKey(HOOK_CREATED_KEY_PREFIX, hook.createdAt, hook.hookId));
2902
+ await store.put(`${HOOK_RELEASE_MARKER_PREFIX}${hook.hookId}`, {
2903
+ hookId: hook.hookId,
2904
+ token: hook.token
2905
+ });
1861
2906
  const event = buildEvent({ ...data });
1862
2907
  await putEvent(event);
1863
2908
  return {
@@ -1972,10 +3017,63 @@ async function finalizeEventPage(store, outcome, params) {
1972
3017
  };
1973
3018
  }
1974
3019
  //#endregion
3020
+ //#region src/retention.ts
3021
+ const CLEANUP_RECORD_KEY = "retention:cleanup";
3022
+ const TOMBSTONE_KEY = "retention:tombstone";
3023
+ const HOOK_MARKER_PREFIX = "retention:hook:";
3024
+ const TERMINAL_CLEANUP_KEY = "terminal:cleanup";
3025
+ function sortableTimestamp(date) {
3026
+ return date.getTime().toString().padStart(13, "0");
3027
+ }
3028
+ function workflowRunIndexKey(run) {
3029
+ return `run:${run.workflowName}:${sortableTimestamp(run.createdAt)}:${run.runId}`;
3030
+ }
3031
+ function globalRunIndexKey(run) {
3032
+ return `runall:${sortableTimestamp(run.createdAt)}:${run.runId}`;
3033
+ }
3034
+ function hookMarkerKey(reference) {
3035
+ return `${HOOK_MARKER_PREFIX}${encodeURIComponent(reference.hookId)}:${encodeURIComponent(reference.token)}`;
3036
+ }
3037
+ function cleanupTombstone(record, now) {
3038
+ const cleanup = {
3039
+ ...record,
3040
+ phase: "tombstoned",
3041
+ tombstonedAt: now
3042
+ };
3043
+ return {
3044
+ version: 1,
3045
+ runId: record.runId,
3046
+ terminalStatus: record.terminalStatus,
3047
+ completedAt: record.completedAt,
3048
+ expiredAt: record.dueAt,
3049
+ tombstonedAt: now,
3050
+ cleanup
3051
+ };
3052
+ }
3053
+ function cleanupFromTombstone(tombstone) {
3054
+ return tombstone.cleanup;
3055
+ }
3056
+ function expiredRead(tombstone) {
3057
+ return {
3058
+ ok: false,
3059
+ code: "RUN_EXPIRED",
3060
+ message: `Workflow run "${tombstone.runId}" expired at ${tombstone.expiredAt.toISOString()}`,
3061
+ tombstone
3062
+ };
3063
+ }
3064
+ //#endregion
1975
3065
  //#region src/worker/durable-objects/WorkflowRunDO.ts
1976
- const PAYLOAD_DELETE_BATCH = 256;
1977
- const STREAM_DELETE_CONCURRENCY = 16;
3066
+ const STORAGE_BATCH_SIZE = 128;
3067
+ const PAYLOAD_DELETE_BATCH = STORAGE_BATCH_SIZE;
3068
+ const HOOK_MARKER_PAGE_SIZE = 64;
3069
+ const TERMINAL_HOOK_PAGE_SIZE = 64;
3070
+ const TERMINAL_WAIT_PAGE_SIZE = STORAGE_BATCH_SIZE;
3071
+ const STREAMS_PER_CLEANUP_PAGE = 16;
3072
+ const STREAM_CHUNKS_PER_CLEANUP_PAGE = STORAGE_BATCH_SIZE;
3073
+ const STREAM_BYTES_PER_CLEANUP_PAGE = 16777216;
3074
+ const QUEUE_REFERENCES_PER_CLEANUP_PAGE = 64;
1978
3075
  const CLEANUP_RETRY_MAX_MS = 36e5;
3076
+ const CLEANUP_PROGRESS_KEY = "retention:progress";
1979
3077
  /**
1980
3078
  * Adapt DO storage (or a transaction) to the {@link EventStore} interface
1981
3079
  * shared with the test mocks. No casts: the methods line up structurally,
@@ -1984,11 +3082,21 @@ const CLEANUP_RETRY_MAX_MS = 36e5;
1984
3082
  function storeFrom(storage) {
1985
3083
  return {
1986
3084
  get: (key) => storage.get(key),
3085
+ getMany: (keys) => storage.get(keys),
1987
3086
  put: (key, value) => storage.put(key, value),
1988
3087
  delete: (key) => storage.delete(key),
1989
- list: (options) => storage.list(options)
3088
+ deleteMany: (keys) => storage.delete(keys),
3089
+ list: (options) => storage.list(options),
3090
+ deferTerminalCleanup: true
1990
3091
  };
1991
3092
  }
3093
+ function hookCreationIndexKey(hook) {
3094
+ return `${HOOK_CREATED_KEY_PREFIX}${hook.createdAt.toISOString()}:${hook.hookId}`;
3095
+ }
3096
+ async function putStorageEntries(storage, entries) {
3097
+ const all = Array.from(entries);
3098
+ for (let offset = 0; offset < all.length; offset += STORAGE_BATCH_SIZE) await storage.put(Object.fromEntries(all.slice(offset, offset + STORAGE_BATCH_SIZE)));
3099
+ }
1992
3100
  /**
1993
3101
  * Durable Object for managing a single workflow run's state.
1994
3102
  *
@@ -2004,26 +3112,51 @@ var WorkflowRunDO = class extends DurableObject {
2004
3112
  const clock = this.env?.clock;
2005
3113
  return typeof clock === "function" ? clock() : Date.now();
2006
3114
  }
2007
- async expiredTombstone(storage, now = this.now()) {
2008
- const tombstone = await storage.get(TOMBSTONE_KEY);
3115
+ tombstoneFrom(values, now = this.now()) {
3116
+ const tombstone = values.get(TOMBSTONE_KEY);
2009
3117
  if (tombstone) return tombstone;
2010
- const cleanup = await storage.get(CLEANUP_RECORD_KEY);
3118
+ const cleanup = values.get(CLEANUP_RECORD_KEY);
2011
3119
  if (!cleanup || cleanup.phase === "retained" && cleanup.dueAt.getTime() > now) return null;
2012
3120
  return cleanupTombstone(cleanup, cleanup.tombstonedAt ?? cleanup.dueAt);
2013
3121
  }
2014
- async read(reader) {
2015
- const tombstone = await this.expiredTombstone(this.ctx.storage);
2016
- if (tombstone) return expiredRead(tombstone);
3122
+ async retentionState(storage, now = this.now()) {
3123
+ const values = await storage.get([TOMBSTONE_KEY, CLEANUP_RECORD_KEY]);
2017
3124
  return {
2018
- ok: true,
2019
- value: await reader()
3125
+ cleanup: values.get(CLEANUP_RECORD_KEY),
3126
+ tombstone: this.tombstoneFrom(values, now)
2020
3127
  };
2021
3128
  }
3129
+ async read(reader) {
3130
+ return await this.ctx.storage.transaction(async (txn) => {
3131
+ const { tombstone } = await this.retentionState(txn);
3132
+ if (tombstone) return expiredRead(tombstone);
3133
+ return {
3134
+ ok: true,
3135
+ value: await reader(txn)
3136
+ };
3137
+ });
3138
+ }
3139
+ async readKey(key) {
3140
+ return await this.ctx.storage.transaction(async (txn) => {
3141
+ const values = await txn.get([
3142
+ TOMBSTONE_KEY,
3143
+ CLEANUP_RECORD_KEY,
3144
+ key
3145
+ ]);
3146
+ const tombstone = this.tombstoneFrom(values);
3147
+ if (tombstone) return expiredRead(tombstone);
3148
+ return {
3149
+ ok: true,
3150
+ value: values.get(key) ?? null
3151
+ };
3152
+ });
3153
+ }
2022
3154
  /** Apply an event and capture all retention metadata in the same transaction. */
2023
3155
  async applyEvent(request) {
2024
3156
  return await this.ctx.storage.transaction(async (txn) => {
2025
3157
  const now = new Date(this.now());
2026
- const tombstone = await this.expiredTombstone(txn, now.getTime());
3158
+ const retention = await this.retentionState(txn, now.getTime());
3159
+ const { tombstone } = retention;
2027
3160
  if (tombstone) return {
2028
3161
  ok: false,
2029
3162
  code: "RUN_EXPIRED",
@@ -2037,20 +3170,12 @@ var WorkflowRunDO = class extends DurableObject {
2037
3170
  });
2038
3171
  if (!outcome.ok) return outcome;
2039
3172
  await txn.put("event_sequence", eventSequence);
2040
- if (outcome.event?.correlationId) {
2041
- const key = correlationIndexKey(outcome.event.correlationId, outcome.event.createdAt, outcome.event.eventId, request.runId);
2042
- await txn.put(indexMarkerKey(key), key);
2043
- }
2044
- if (outcome.hookToIndex) {
2045
- const reference = {
2046
- hookId: outcome.hookToIndex.hookId,
2047
- token: outcome.hookToIndex.token
2048
- };
2049
- await txn.put(hookMarkerKey(reference), reference);
2050
- }
2051
- for (const reference of outcome.releasedHooks) await txn.put(hookMarkerKey(reference), reference);
3173
+ await putStorageEntries(txn, (outcome.hookToIndex ? [{
3174
+ hookId: outcome.hookToIndex.hookId,
3175
+ token: outcome.hookToIndex.token
3176
+ }, ...outcome.releasedHooks] : outcome.releasedHooks).map((reference) => [hookMarkerKey(reference), reference]));
2052
3177
  const retentionMs = request.cleanup?.retentionMs ?? 0;
2053
- if (retentionMs > 0 && outcome.run && isTerminalWorkflowRunStatus(outcome.run.status) && !await txn.get("retention:cleanup")) {
3178
+ if (retentionMs > 0 && outcome.run && isTerminalWorkflowRunStatus(outcome.run.status) && !retention.cleanup) {
2054
3179
  const terminalRun = outcome.run;
2055
3180
  const dueAt = new Date(terminalRun.completedAt.getTime() + retentionMs);
2056
3181
  const run = WorkflowRunSchema.parse({
@@ -2067,6 +3192,7 @@ var WorkflowRunDO = class extends DurableObject {
2067
3192
  dueAt,
2068
3193
  queueShards: request.cleanup?.queueShards ?? 1,
2069
3194
  phase: "retained",
3195
+ generation: 0,
2070
3196
  attempts: 0,
2071
3197
  deletedPayloadKeys: 0,
2072
3198
  deletedStreams: 0,
@@ -2077,50 +3203,103 @@ var WorkflowRunDO = class extends DurableObject {
2077
3203
  await txn.setAlarm(dueAt);
2078
3204
  outcome.run = run;
2079
3205
  }
3206
+ if (outcome.run && isTerminalWorkflowRunStatus(outcome.run.status)) {
3207
+ if (!await txn.get("terminal:cleanup")) await txn.put(TERMINAL_CLEANUP_KEY, {
3208
+ version: 1,
3209
+ runId: outcome.run.runId,
3210
+ phase: "hooks",
3211
+ generation: 0,
3212
+ attempts: 0
3213
+ });
3214
+ await txn.setAlarm(this.now() + 1);
3215
+ }
3216
+ if (outcome.run) outcome.indexPublicationExpiresAt = now.getTime() + MAX_RUN_INDEX_PUBLICATION_LIFETIME_MS;
2080
3217
  return finalizeEventPage(storeFrom(txn), outcome, request.params);
2081
3218
  });
2082
3219
  }
3220
+ /**
3221
+ * Resolve an ambiguous hook-creation RPC in the same serialization domain as
3222
+ * applyEvent. If the hook is absent, the exact reservation is durably fenced
3223
+ * before the caller releases its token and hook-ID claims.
3224
+ */
3225
+ async resolveHookTokenClaim(request) {
3226
+ return await this.ctx.storage.transaction(async (txn) => {
3227
+ const { tombstone } = await this.retentionState(txn);
3228
+ if (tombstone) return { committed: false };
3229
+ if ((await txn.get(`hook:${request.hookId}`))?.token === request.token) return { committed: true };
3230
+ await txn.put(hookClaimCancellationKey(request.claimId), {
3231
+ hookId: request.hookId,
3232
+ token: request.token,
3233
+ canceledAt: this.now()
3234
+ });
3235
+ return { committed: false };
3236
+ });
3237
+ }
2083
3238
  async getRun() {
2084
- return this.read(async () => await this.ctx.storage.get("run") ?? null);
3239
+ return this.readKey("run");
3240
+ }
3241
+ /** Single authoritative lifecycle answer used after derivative compaction. */
3242
+ async getLifecycleStatus() {
3243
+ return await this.ctx.storage.transaction(async (txn) => {
3244
+ const values = await txn.get([
3245
+ TOMBSTONE_KEY,
3246
+ CLEANUP_RECORD_KEY,
3247
+ "run"
3248
+ ]);
3249
+ if (this.tombstoneFrom(values)) return "expired";
3250
+ const run = values.get("run");
3251
+ if (!run) return "missing";
3252
+ return isTerminalWorkflowRunStatus(run.status) ? "terminal" : "active";
3253
+ });
2085
3254
  }
2086
3255
  async getStep(stepId) {
2087
- return this.read(async () => await this.ctx.storage.get(`step:${stepId}`) ?? null);
3256
+ return this.readKey(`${STEP_KEY_PREFIX}${stepId}`);
2088
3257
  }
2089
3258
  async listSteps(params) {
2090
- return this.read(() => listByCreationTime(storeFrom(this.ctx.storage), STEP_CREATED_KEY_PREFIX, STEP_KEY_PREFIX, {
3259
+ return this.read((txn) => listByCreationTime(storeFrom(txn), STEP_CREATED_KEY_PREFIX, STEP_KEY_PREFIX, {
2091
3260
  limit: params?.limit ?? 20,
2092
3261
  cursor: params?.cursor,
2093
3262
  sortOrder: params?.sortOrder ?? "asc"
2094
3263
  }));
2095
3264
  }
2096
3265
  async getEvent(eventId) {
2097
- return this.read(async () => await this.ctx.storage.get(`event:${eventId}`) ?? null);
3266
+ return this.readKey(`${EVENT_KEY_PREFIX}${eventId}`);
2098
3267
  }
2099
3268
  async listEvents(params) {
2100
- return this.read(() => listByPrefix(storeFrom(this.ctx.storage), EVENT_KEY_PREFIX, {
3269
+ return this.read((txn) => listByPrefix(storeFrom(txn), EVENT_KEY_PREFIX, {
2101
3270
  limit: params?.limit ?? 100,
2102
3271
  cursor: params?.cursor,
2103
3272
  sortOrder: params?.sortOrder ?? "asc"
2104
3273
  }, (event) => event.eventId));
2105
3274
  }
2106
3275
  async listHooks(params) {
2107
- return this.read(() => listByCreationTime(storeFrom(this.ctx.storage), HOOK_CREATED_KEY_PREFIX, HOOK_KEY_PREFIX, {
3276
+ return this.read((txn) => listByCreationTime(storeFrom(txn), HOOK_CREATED_KEY_PREFIX, HOOK_KEY_PREFIX, {
2108
3277
  limit: params?.limit ?? 100,
2109
3278
  cursor: params?.cursor,
2110
3279
  sortOrder: params?.sortOrder ?? "asc"
2111
3280
  }));
2112
3281
  }
2113
3282
  async getCleanupStatus() {
2114
- return await this.ctx.storage.get("retention:cleanup") ?? null;
3283
+ const values = await this.ctx.storage.get([CLEANUP_RECORD_KEY, TOMBSTONE_KEY]);
3284
+ const cleanup = values.get(CLEANUP_RECORD_KEY);
3285
+ if (cleanup) return cleanup;
3286
+ const tombstone = values.get(TOMBSTONE_KEY);
3287
+ return tombstone ? cleanupFromTombstone(tombstone) : null;
2115
3288
  }
2116
3289
  /** Explicitly schedule an existing terminal run, used for opt-in backfills. */
2117
3290
  async scheduleCleanup(request) {
2118
3291
  if (request.retentionMs <= 0) return this.getCleanupStatus();
2119
3292
  return await this.ctx.storage.transaction(async (txn) => {
2120
- const existing = await txn.get(CLEANUP_RECORD_KEY);
3293
+ const values = await txn.get([
3294
+ CLEANUP_RECORD_KEY,
3295
+ TOMBSTONE_KEY,
3296
+ TERMINAL_CLEANUP_KEY,
3297
+ "run"
3298
+ ]);
3299
+ const existing = values.get(CLEANUP_RECORD_KEY);
2121
3300
  if (existing) return existing;
2122
- if (await txn.get("retention:tombstone")) return null;
2123
- const current = await txn.get("run");
3301
+ if (values.get("retention:tombstone")) return null;
3302
+ const current = values.get("run");
2124
3303
  if (!current || !isTerminalWorkflowRunStatus(current.status)) return null;
2125
3304
  const terminalRun = current;
2126
3305
  const dueAt = new Date(terminalRun.completedAt.getTime() + request.retentionMs);
@@ -2138,6 +3317,7 @@ var WorkflowRunDO = class extends DurableObject {
2138
3317
  dueAt,
2139
3318
  queueShards: request.queueShards,
2140
3319
  phase: "retained",
3320
+ generation: 0,
2141
3321
  attempts: 0,
2142
3322
  deletedPayloadKeys: 0,
2143
3323
  deletedStreams: 0,
@@ -2145,7 +3325,7 @@ var WorkflowRunDO = class extends DurableObject {
2145
3325
  };
2146
3326
  await txn.put("run", run);
2147
3327
  await txn.put(CLEANUP_RECORD_KEY, cleanup);
2148
- await txn.setAlarm(dueAt.getTime() <= this.now() ? this.now() + 1 : dueAt);
3328
+ await txn.setAlarm(values.get("terminal:cleanup") || dueAt.getTime() <= this.now() ? this.now() + 1 : dueAt);
2149
3329
  return cleanup;
2150
3330
  });
2151
3331
  }
@@ -2165,159 +3345,388 @@ var WorkflowRunDO = class extends DurableObject {
2165
3345
  }));
2166
3346
  await txn.put(CLEANUP_RECORD_KEY, {
2167
3347
  ...cleanup,
2168
- dueAt
3348
+ dueAt,
3349
+ generation: cleanup.generation + 1
2169
3350
  });
2170
3351
  await txn.setAlarm(this.now() + 1);
2171
3352
  });
2172
- try {
2173
- await this.executeCleanup();
2174
- } catch (error) {
2175
- await this.recordCleanupFailure(error);
2176
- }
3353
+ await this.executeScheduledCleanupPage();
2177
3354
  return this.getCleanupStatus();
2178
3355
  }
2179
3356
  async rearmCleanup() {
2180
- const cleanup = await this.ctx.storage.get(CLEANUP_RECORD_KEY);
2181
- if (!cleanup) return null;
2182
- if (cleanup.phase === "tombstoned") await this.ctx.storage.deleteAlarm();
3357
+ const values = await this.ctx.storage.get([CLEANUP_RECORD_KEY, TERMINAL_CLEANUP_KEY]);
3358
+ const cleanup = values.get(CLEANUP_RECORD_KEY);
3359
+ if (values.get("terminal:cleanup")) await this.ctx.storage.setAlarm(this.now() + 1);
3360
+ else if (!cleanup || cleanup.phase === "tombstoned") await this.ctx.storage.deleteAlarm();
2183
3361
  else await this.ctx.storage.setAlarm(Math.max(this.now() + 1, cleanup.dueAt.getTime()));
2184
- return cleanup;
3362
+ return cleanup ?? null;
2185
3363
  }
2186
3364
  async alarm() {
2187
- try {
2188
- await this.executeCleanup();
2189
- } catch (error) {
2190
- await this.recordCleanupFailure(error);
2191
- }
3365
+ await this.executeScheduledCleanupPage();
2192
3366
  }
2193
3367
  namespace(name) {
2194
3368
  const namespace = this.env?.[name];
2195
3369
  if (!namespace || typeof namespace === "function") throw new Error(`world-celld retention missing binding ${name}`);
2196
3370
  return namespace;
2197
3371
  }
2198
- async setCleanupPhase(expected, next, updates = {}) {
3372
+ workflowIndex() {
3373
+ return createWorkflowIndex({
3374
+ runCatalog: this.namespace("WORKFLOW_RUN_CATALOG"),
3375
+ hookTokens: this.namespace("WORKFLOW_HOOK_TOKENS"),
3376
+ hookIds: this.namespace("WORKFLOW_HOOK_IDS")
3377
+ });
3378
+ }
3379
+ async setCleanupPhase(expected, next, updates = {}, continueAt) {
2199
3380
  await this.ctx.storage.transaction(async (txn) => {
2200
3381
  const cleanup = await txn.get(CLEANUP_RECORD_KEY);
2201
- if (!cleanup || cleanup.phase !== expected) return;
3382
+ if (!cleanup || cleanup.phase !== expected.phase || cleanup.generation !== expected.generation) return;
2202
3383
  await txn.put(CLEANUP_RECORD_KEY, {
2203
3384
  ...cleanup,
2204
3385
  ...updates,
2205
3386
  phase: next,
3387
+ generation: cleanup.generation + 1,
2206
3388
  attempts: 0,
2207
3389
  lastError: void 0
2208
3390
  });
3391
+ if (continueAt !== void 0) await txn.setAlarm(continueAt);
2209
3392
  });
2210
3393
  }
2211
- async executeCleanup() {
2212
- for (;;) {
2213
- const cleanup = await this.ctx.storage.get(CLEANUP_RECORD_KEY);
2214
- if (!cleanup) {
2215
- await this.ctx.storage.deleteAlarm();
2216
- return;
3394
+ async executeScheduledCleanupPage() {
3395
+ const terminalCleanup = await this.ctx.storage.get(TERMINAL_CLEANUP_KEY);
3396
+ if (terminalCleanup) {
3397
+ try {
3398
+ await this.executeTerminalCleanup(terminalCleanup);
3399
+ } catch (error) {
3400
+ await this.recordTerminalCleanupFailure(error, terminalCleanup);
2217
3401
  }
2218
- if (cleanup.phase === "tombstoned") {
2219
- await this.ctx.storage.deleteAlarm();
3402
+ return;
3403
+ }
3404
+ const cleanup = await this.ctx.storage.get(CLEANUP_RECORD_KEY);
3405
+ if (!cleanup || cleanup.phase === "tombstoned") {
3406
+ await this.ctx.storage.deleteAlarm();
3407
+ return;
3408
+ }
3409
+ try {
3410
+ await this.executeCleanup(cleanup);
3411
+ } catch (error) {
3412
+ await this.recordCleanupFailure(error, cleanup);
3413
+ }
3414
+ }
3415
+ async executeCleanup(cleanup) {
3416
+ if (cleanup.phase === "retained") {
3417
+ if (cleanup.dueAt.getTime() > this.now()) {
3418
+ await this.ctx.storage.setAlarm(cleanup.dueAt);
2220
3419
  return;
2221
3420
  }
2222
- if (cleanup.phase === "retained") {
2223
- if (cleanup.dueAt.getTime() > this.now()) {
2224
- await this.ctx.storage.setAlarm(cleanup.dueAt);
2225
- return;
2226
- }
2227
- await this.setCleanupPhase("retained", "index");
2228
- continue;
2229
- }
2230
- if (cleanup.phase === "index") {
2231
- const [indexMarkers, hookMarkers] = await Promise.all([this.ctx.storage.list({ prefix: INDEX_MARKER_PREFIX }), this.ctx.storage.list({ prefix: HOOK_MARKER_PREFIX })]);
2232
- const indexNamespace = this.namespace("WORKFLOW_INDEX");
2233
- await indexNamespace.get(indexNamespace.idFromName("index")).expireRun({
2234
- runId: cleanup.runId,
2235
- keys: [
2236
- workflowRunIndexKey(cleanup),
2237
- globalRunIndexKey(cleanup),
2238
- ...indexMarkers.values()
2239
- ],
2240
- hooks: Array.from(hookMarkers.values()),
2241
- expiredAt: cleanup.dueAt.getTime()
3421
+ await this.setCleanupPhase(cleanup, "index", {}, this.now() + 1);
3422
+ return;
3423
+ }
3424
+ if (cleanup.phase === "index") {
3425
+ await this.deleteIndexPage(cleanup);
3426
+ return;
3427
+ }
3428
+ if (cleanup.phase === "streams") {
3429
+ await this.deleteStreamPage(cleanup);
3430
+ return;
3431
+ }
3432
+ if (cleanup.phase === "queues") {
3433
+ await this.deleteQueuePage(cleanup);
3434
+ return;
3435
+ }
3436
+ if (cleanup.phase === "payload") await this.deletePayloadBatch();
3437
+ }
3438
+ async executeTerminalCleanup(cleanup) {
3439
+ if (cleanup.phase === "hooks") await this.deleteTerminalHookPage(cleanup);
3440
+ else if (cleanup.phase === "markers") await this.deleteTerminalHookMarkerPage(cleanup);
3441
+ else await this.deleteTerminalWaitPage(cleanup);
3442
+ }
3443
+ async deleteTerminalHookPage(cleanup) {
3444
+ const entries = await this.ctx.storage.list({
3445
+ prefix: HOOK_KEY_PREFIX,
3446
+ limit: 65
3447
+ });
3448
+ const page = Array.from(entries.values()).slice(0, TERMINAL_HOOK_PAGE_SIZE);
3449
+ await this.workflowIndex().releaseHookIndexes({
3450
+ runId: cleanup.runId,
3451
+ hooks: page.map((hook) => ({
3452
+ hookId: hook.hookId,
3453
+ token: hook.token
3454
+ }))
3455
+ });
3456
+ await this.ctx.storage.transaction(async (txn) => {
3457
+ const current = await txn.get(TERMINAL_CLEANUP_KEY);
3458
+ if (!current || current.phase !== cleanup.phase || current.generation !== cleanup.generation) return;
3459
+ const keys = page.flatMap((hook) => [`${HOOK_KEY_PREFIX}${hook.hookId}`, hookCreationIndexKey(hook)]);
3460
+ if (keys.length > 0) await txn.delete(keys);
3461
+ await txn.put(TERMINAL_CLEANUP_KEY, {
3462
+ ...current,
3463
+ phase: entries.size <= TERMINAL_HOOK_PAGE_SIZE ? "markers" : "hooks",
3464
+ generation: current.generation + 1,
3465
+ attempts: 0,
3466
+ lastError: void 0
3467
+ });
3468
+ await txn.setAlarm(this.now() + 1);
3469
+ });
3470
+ }
3471
+ async deleteTerminalHookMarkerPage(cleanup) {
3472
+ const entries = await this.ctx.storage.list({
3473
+ prefix: HOOK_MARKER_PREFIX,
3474
+ limit: 65
3475
+ });
3476
+ const page = Array.from(entries).slice(0, HOOK_MARKER_PAGE_SIZE);
3477
+ await this.workflowIndex().releaseHookIndexes({
3478
+ runId: cleanup.runId,
3479
+ hooks: page.map(([, reference]) => reference)
3480
+ });
3481
+ await this.ctx.storage.transaction(async (txn) => {
3482
+ const current = await txn.get(TERMINAL_CLEANUP_KEY);
3483
+ if (!current || current.phase !== cleanup.phase || current.generation !== cleanup.generation) return;
3484
+ const keys = page.map(([key]) => key);
3485
+ if (keys.length > 0) await txn.delete(keys);
3486
+ await txn.put(TERMINAL_CLEANUP_KEY, {
3487
+ ...current,
3488
+ phase: entries.size <= HOOK_MARKER_PAGE_SIZE ? "waits" : "markers",
3489
+ generation: current.generation + 1,
3490
+ attempts: 0,
3491
+ lastError: void 0
3492
+ });
3493
+ await txn.setAlarm(this.now() + 1);
3494
+ });
3495
+ }
3496
+ async deleteTerminalWaitPage(cleanup) {
3497
+ await this.ctx.storage.transaction(async (txn) => {
3498
+ const current = await txn.get(TERMINAL_CLEANUP_KEY);
3499
+ if (!current || current.phase !== cleanup.phase || current.generation !== cleanup.generation) return;
3500
+ const entries = await txn.list({
3501
+ prefix: WAIT_KEY_PREFIX,
3502
+ limit: 129
3503
+ });
3504
+ const keys = Array.from(entries.keys()).slice(0, TERMINAL_WAIT_PAGE_SIZE);
3505
+ if (keys.length > 0) await txn.delete(keys);
3506
+ if (entries.size > TERMINAL_WAIT_PAGE_SIZE) {
3507
+ await txn.put(TERMINAL_CLEANUP_KEY, {
3508
+ ...current,
3509
+ generation: current.generation + 1,
3510
+ attempts: 0,
3511
+ lastError: void 0
2242
3512
  });
2243
- await this.setCleanupPhase("index", "streams");
2244
- continue;
3513
+ await txn.setAlarm(this.now() + 1);
3514
+ return;
2245
3515
  }
2246
- if (cleanup.phase === "streams") {
2247
- const streamsNamespace = this.namespace("WORKFLOW_STREAMS");
2248
- const registry = streamsNamespace.get(streamsNamespace.idFromName(`run-streams:${cleanup.runId}`));
2249
- const { streams } = await registry.expireRegistry(cleanup.runId, cleanup.dueAt.getTime());
2250
- let deletedStreams = streams.length === 0 ? cleanup.deletedStreams : 0;
2251
- for (let offset = 0; offset < streams.length; offset += STREAM_DELETE_CONCURRENCY) {
2252
- const batch = streams.slice(offset, offset + STREAM_DELETE_CONCURRENCY);
2253
- const results = await Promise.all(batch.map((name) => {
2254
- return streamsNamespace.get(streamsNamespace.idFromName(`stream:${name}`)).expireStream(cleanup.runId, cleanup.dueAt.getTime());
2255
- }));
2256
- deletedStreams += results.filter((result) => result.deleted).length;
2257
- }
2258
- await this.ctx.storage.transaction(async (txn) => {
2259
- const current = await txn.get(CLEANUP_RECORD_KEY);
2260
- if (current?.phase === "streams") await txn.put(CLEANUP_RECORD_KEY, {
2261
- ...current,
2262
- deletedStreams
2263
- });
3516
+ await txn.delete(TERMINAL_CLEANUP_KEY);
3517
+ const retention = await txn.get(CLEANUP_RECORD_KEY);
3518
+ if (retention && retention.phase !== "tombstoned") await txn.setAlarm(Math.max(this.now() + 1, retention.dueAt.getTime()));
3519
+ else await txn.deleteAlarm();
3520
+ });
3521
+ }
3522
+ async deleteIndexPage(cleanup) {
3523
+ const hookEntries = await this.ctx.storage.list({
3524
+ prefix: HOOK_MARKER_PREFIX,
3525
+ limit: 65
3526
+ });
3527
+ const hookMarkers = Array.from(hookEntries).slice(0, HOOK_MARKER_PAGE_SIZE);
3528
+ await this.workflowIndex().expireRun({
3529
+ runId: cleanup.runId,
3530
+ keys: [workflowRunIndexKey(cleanup), globalRunIndexKey(cleanup)],
3531
+ hooks: hookMarkers.map(([, value]) => value),
3532
+ expiredAt: cleanup.dueAt.getTime()
3533
+ });
3534
+ const markerKeys = hookMarkers.map(([key]) => key);
3535
+ await this.ctx.storage.transaction(async (txn) => {
3536
+ const current = await txn.get(CLEANUP_RECORD_KEY);
3537
+ if (current?.phase !== "index" || current.generation !== cleanup.generation) return;
3538
+ if (markerKeys.length > 0) await txn.delete(markerKeys);
3539
+ if (hookEntries.size <= HOOK_MARKER_PAGE_SIZE) await txn.put(CLEANUP_RECORD_KEY, {
3540
+ ...current,
3541
+ phase: "streams",
3542
+ generation: current.generation + 1,
3543
+ attempts: 0,
3544
+ lastError: void 0
3545
+ });
3546
+ else await txn.put(CLEANUP_RECORD_KEY, {
3547
+ ...current,
3548
+ generation: current.generation + 1,
3549
+ attempts: 0,
3550
+ lastError: void 0
3551
+ });
3552
+ await txn.setAlarm(this.now() + 1);
3553
+ });
3554
+ }
3555
+ async deleteStreamPage(cleanup) {
3556
+ const streamsNamespace = this.namespace("WORKFLOW_STREAMS");
3557
+ const registry = streamsNamespace.get(streamsNamespace.idFromName(`run-streams:${cleanup.runId}`));
3558
+ const { streams } = await registry.expireRegistry(cleanup.runId, cleanup.dueAt.getTime(), { limit: STREAMS_PER_CLEANUP_PAGE });
3559
+ const completed = [];
3560
+ let remainingChunks = STREAM_CHUNKS_PER_CLEANUP_PAGE;
3561
+ let remainingBytes = STREAM_BYTES_PER_CLEANUP_PAGE;
3562
+ for (const name of streams) {
3563
+ if (remainingChunks <= 0 || remainingBytes <= 0) break;
3564
+ const result = await streamsNamespace.get(streamsNamespace.idFromName(`stream:${name}`)).expireStream(cleanup.runId, cleanup.dueAt.getTime(), {
3565
+ limit: remainingChunks,
3566
+ byteLimit: remainingBytes
3567
+ });
3568
+ remainingChunks -= result.chunks;
3569
+ remainingBytes -= result.bytes;
3570
+ if (result.done) completed.push(name);
3571
+ }
3572
+ const registryResult = await registry.finalizeRegistry(cleanup.runId, completed);
3573
+ await this.ctx.storage.transaction(async (txn) => {
3574
+ const current = await txn.get(CLEANUP_RECORD_KEY);
3575
+ if (current?.phase !== "streams" || current.generation !== cleanup.generation) return;
3576
+ await txn.put(CLEANUP_RECORD_KEY, {
3577
+ ...current,
3578
+ deletedStreams: registryResult.deleted,
3579
+ phase: registryResult.done ? "queues" : "streams",
3580
+ generation: current.generation + 1,
3581
+ attempts: 0,
3582
+ lastError: void 0
3583
+ });
3584
+ await txn.setAlarm(this.now() + 1);
3585
+ });
3586
+ }
3587
+ async deleteQueuePage(cleanup) {
3588
+ const queues = this.namespace("WORKFLOW_QUEUE");
3589
+ const progress = await this.ctx.storage.get(CLEANUP_PROGRESS_KEY) ?? {
3590
+ queueShard: 0,
3591
+ queueShardDeleted: 0
3592
+ };
3593
+ if (progress.pendingAck) {
3594
+ const { queueShard, receipt } = progress.pendingAck;
3595
+ if (!(await queues.get(queues.idFromName(`q:${queueShard}`)).acknowledgeExpireRun(cleanup.runId, receipt)).acknowledged) throw new Error(`queue shard ${queueShard} rejected expiry receipt for ${cleanup.runId}`);
3596
+ await this.ctx.storage.transaction(async (txn) => {
3597
+ const current = await txn.get(CLEANUP_RECORD_KEY);
3598
+ if (current?.phase !== "queues" || current.generation !== cleanup.generation) return;
3599
+ const persisted = await txn.get(CLEANUP_PROGRESS_KEY);
3600
+ if (persisted?.pendingAck?.queueShard !== queueShard || persisted.pendingAck.receipt.expiredAt !== receipt.expiredAt || persisted.pendingAck.receipt.deleted !== receipt.deleted) return;
3601
+ await txn.put(CLEANUP_RECORD_KEY, {
3602
+ ...current,
3603
+ generation: current.generation + 1,
3604
+ attempts: 0,
3605
+ lastError: void 0
2264
3606
  });
2265
- await registry.finalizeRegistry(cleanup.runId);
2266
- await this.setCleanupPhase("streams", "queues");
2267
- continue;
2268
- }
2269
- if (cleanup.phase === "queues") {
2270
- const queues = this.namespace("WORKFLOW_QUEUE");
2271
- let deletedQueueMessages = 0;
2272
- for (let shard = 0; shard < cleanup.queueShards; shard++) {
2273
- const result = await queues.get(queues.idFromName(`q:${shard}`)).expireRun(cleanup.runId, cleanup.dueAt.getTime());
2274
- deletedQueueMessages += result.deleted;
2275
- }
2276
- await this.setCleanupPhase("queues", "payload", { deletedQueueMessages });
2277
- continue;
2278
- }
2279
- if (cleanup.phase === "payload") {
2280
- if (!await this.deletePayloadBatch()) return;
2281
- }
3607
+ await txn.put(CLEANUP_PROGRESS_KEY, {
3608
+ queueShard: queueShard + 1,
3609
+ queueShardDeleted: 0
3610
+ });
3611
+ await txn.setAlarm(this.now() + 1);
3612
+ });
3613
+ return;
3614
+ }
3615
+ if (progress.queueShard < cleanup.queueShards) {
3616
+ const requestedShard = progress.queueShard;
3617
+ const result = await queues.get(queues.idFromName(`q:${requestedShard}`)).expireRun(cleanup.runId, cleanup.dueAt.getTime(), { limit: QUEUE_REFERENCES_PER_CLEANUP_PAGE });
3618
+ await this.ctx.storage.transaction(async (txn) => {
3619
+ const current = await txn.get(CLEANUP_RECORD_KEY);
3620
+ if (current?.phase !== "queues" || current.generation !== cleanup.generation) return;
3621
+ const persisted = await txn.get(CLEANUP_PROGRESS_KEY) ?? {
3622
+ queueShard: 0,
3623
+ queueShardDeleted: 0
3624
+ };
3625
+ if (persisted.queueShard !== requestedShard || persisted.pendingAck) return;
3626
+ if (result.deleted < persisted.queueShardDeleted) throw new Error(`queue shard ${requestedShard} regressed expiry receipt for ${cleanup.runId}`);
3627
+ const delta = result.deleted - persisted.queueShardDeleted;
3628
+ const next = result.done ? {
3629
+ queueShard: requestedShard,
3630
+ queueShardDeleted: result.deleted,
3631
+ pendingAck: {
3632
+ queueShard: requestedShard,
3633
+ receipt: result.receipt
3634
+ }
3635
+ } : {
3636
+ queueShard: requestedShard,
3637
+ queueShardDeleted: result.deleted
3638
+ };
3639
+ await txn.put(CLEANUP_RECORD_KEY, {
3640
+ ...current,
3641
+ deletedQueueMessages: current.deletedQueueMessages + delta,
3642
+ generation: current.generation + 1,
3643
+ attempts: 0,
3644
+ lastError: void 0
3645
+ });
3646
+ await txn.put(CLEANUP_PROGRESS_KEY, next);
3647
+ await txn.setAlarm(this.now() + 1);
3648
+ });
3649
+ return;
2282
3650
  }
3651
+ await this.ctx.storage.transaction(async (txn) => {
3652
+ const current = await txn.get(CLEANUP_RECORD_KEY);
3653
+ if (current?.phase !== "queues" || current.generation !== cleanup.generation) return;
3654
+ const persisted = await txn.get(CLEANUP_PROGRESS_KEY) ?? {
3655
+ queueShard: 0,
3656
+ queueShardDeleted: 0
3657
+ };
3658
+ if (persisted.queueShard >= current.queueShards && !persisted.pendingAck) {
3659
+ await txn.put(CLEANUP_RECORD_KEY, {
3660
+ ...current,
3661
+ phase: "payload",
3662
+ generation: current.generation + 1
3663
+ });
3664
+ await txn.delete(CLEANUP_PROGRESS_KEY);
3665
+ await txn.setAlarm(this.now() + 1);
3666
+ }
3667
+ });
2283
3668
  }
2284
3669
  async deletePayloadBatch() {
2285
3670
  return await this.ctx.storage.transaction(async (txn) => {
2286
3671
  const cleanup = await txn.get(CLEANUP_RECORD_KEY);
2287
3672
  if (!cleanup || cleanup.phase !== "payload") return true;
2288
- const entries = await txn.list({ limit: 259 });
3673
+ const entries = await txn.list({
3674
+ limit: 131,
3675
+ noCache: true
3676
+ });
2289
3677
  const keys = Array.from(entries.keys()).filter((key) => key !== "retention:cleanup" && key !== "retention:tombstone").slice(0, PAYLOAD_DELETE_BATCH);
2290
3678
  const now = new Date(this.now());
2291
3679
  await txn.put(TOMBSTONE_KEY, cleanupTombstone(cleanup, now));
2292
- for (const key of keys) await txn.delete(key);
3680
+ if (keys.length > 0) await txn.delete(keys);
2293
3681
  if (keys.length > 0) {
2294
3682
  await txn.put(CLEANUP_RECORD_KEY, {
2295
3683
  ...cleanup,
2296
3684
  deletedPayloadKeys: cleanup.deletedPayloadKeys + keys.length,
3685
+ generation: cleanup.generation + 1,
2297
3686
  lastError: void 0
2298
3687
  });
2299
3688
  await txn.setAlarm(this.now() + 1);
2300
3689
  return false;
2301
3690
  }
2302
- await txn.put(CLEANUP_RECORD_KEY, {
3691
+ const completed = {
2303
3692
  ...cleanup,
2304
3693
  phase: "tombstoned",
3694
+ generation: cleanup.generation + 1,
2305
3695
  attempts: 0,
2306
3696
  lastError: void 0,
2307
3697
  tombstonedAt: now
2308
- });
3698
+ };
3699
+ await txn.put(TOMBSTONE_KEY, cleanupTombstone(completed, now));
3700
+ await txn.delete(CLEANUP_RECORD_KEY);
2309
3701
  await txn.deleteAlarm();
2310
3702
  return true;
2311
3703
  });
2312
3704
  }
2313
- async recordCleanupFailure(error) {
3705
+ async recordCleanupFailure(error, expected) {
2314
3706
  await this.ctx.storage.transaction(async (txn) => {
2315
3707
  const cleanup = await txn.get(CLEANUP_RECORD_KEY);
2316
- if (!cleanup || cleanup.phase === "tombstoned") return;
3708
+ if (!cleanup || cleanup.phase !== expected.phase || cleanup.generation !== expected.generation || cleanup.phase === "tombstoned") return;
2317
3709
  const attempts = cleanup.attempts + 1;
2318
3710
  const delay = Math.min(CLEANUP_RETRY_MAX_MS, 1e3 * 2 ** Math.min(attempts - 1, 12));
2319
3711
  await txn.put(CLEANUP_RECORD_KEY, {
2320
3712
  ...cleanup,
3713
+ generation: cleanup.generation + 1,
3714
+ attempts,
3715
+ lastError: error instanceof Error ? error.message : String(error),
3716
+ lastAttemptAt: new Date(this.now())
3717
+ });
3718
+ await txn.setAlarm(this.now() + delay);
3719
+ });
3720
+ }
3721
+ async recordTerminalCleanupFailure(error, expected) {
3722
+ await this.ctx.storage.transaction(async (txn) => {
3723
+ const cleanup = await txn.get(TERMINAL_CLEANUP_KEY);
3724
+ if (!cleanup || cleanup.phase !== expected.phase || cleanup.generation !== expected.generation) return;
3725
+ const attempts = cleanup.attempts + 1;
3726
+ const delay = Math.min(CLEANUP_RETRY_MAX_MS, 1e3 * 2 ** Math.min(attempts - 1, 12));
3727
+ await txn.put(TERMINAL_CLEANUP_KEY, {
3728
+ ...cleanup,
3729
+ generation: cleanup.generation + 1,
2321
3730
  attempts,
2322
3731
  lastError: error instanceof Error ? error.message : String(error),
2323
3732
  lastAttemptAt: new Date(this.now())
@@ -2352,7 +3761,9 @@ async function startHarness(options = {}) {
2352
3761
  const fleet = new FakeFleet({
2353
3762
  runs: WorkflowRunDO,
2354
3763
  streams: StreamDO,
2355
- index: IndexDO,
3764
+ "run-catalog": RunCatalogDO,
3765
+ "hook-tokens": HookTokenDO,
3766
+ "hook-ids": HookIdDO,
2356
3767
  queue: QueueDO,
2357
3768
  ...options.extraClasses
2358
3769
  }, cellEnv);
@@ -2360,17 +3771,30 @@ async function startHarness(options = {}) {
2360
3771
  Object.assign(cellEnv, {
2361
3772
  WORKFLOW_DB: fleet.namespace("runs"),
2362
3773
  WORKFLOW_STREAMS: fleet.namespace("streams"),
2363
- WORKFLOW_INDEX: fleet.namespace("index"),
3774
+ WORKFLOW_RUN_CATALOG: fleet.namespace("run-catalog"),
3775
+ WORKFLOW_HOOK_TOKENS: fleet.namespace("hook-tokens"),
3776
+ WORKFLOW_HOOK_IDS: fleet.namespace("hook-ids"),
2364
3777
  WORKFLOW_QUEUE: fleet.namespace("queue")
2365
3778
  });
2366
3779
  const router = createRouter({
2367
3780
  WORKFLOW_DB: fleet.namespace("runs"),
2368
3781
  WORKFLOW_STREAMS: fleet.namespace("streams"),
2369
- WORKFLOW_INDEX: fleet.namespace("index"),
3782
+ WORKFLOW_RUN_CATALOG: fleet.namespace("run-catalog"),
3783
+ WORKFLOW_HOOK_TOKENS: fleet.namespace("hook-tokens"),
3784
+ WORKFLOW_HOOK_IDS: fleet.namespace("hook-ids"),
2370
3785
  WORKFLOW_QUEUE: fleet.namespace("queue"),
2371
3786
  WORLD_SECRET: options.secret
2372
3787
  });
2373
3788
  const server = http.createServer(async (req, res) => {
3789
+ const requestAbort = new AbortController();
3790
+ const abortRequest = () => {
3791
+ if (!requestAbort.signal.aborted) requestAbort.abort(new DOMException("HTTP client disconnected", "AbortError"));
3792
+ };
3793
+ const abortOnPrematureClose = () => {
3794
+ if (!res.writableEnded) abortRequest();
3795
+ };
3796
+ req.once("aborted", abortRequest);
3797
+ res.once("close", abortOnPrematureClose);
2374
3798
  try {
2375
3799
  const chunks = [];
2376
3800
  for await (const chunk of req) chunks.push(chunk);
@@ -2378,17 +3802,23 @@ async function startHarness(options = {}) {
2378
3802
  const request = new Request(`http://127.0.0.1${req.url ?? "/"}`, {
2379
3803
  method: req.method,
2380
3804
  headers: req.headers,
2381
- body: req.method === "GET" || req.method === "HEAD" || body.length === 0 ? null : body
3805
+ body: req.method === "GET" || req.method === "HEAD" || body.length === 0 ? null : body,
3806
+ signal: requestAbort.signal
2382
3807
  });
2383
3808
  const response = await router(request);
3809
+ if (requestAbort.signal.aborted || res.destroyed) return;
2384
3810
  res.writeHead(response.status, Object.fromEntries(response.headers.entries()));
2385
3811
  res.end(Buffer.from(await response.arrayBuffer()));
2386
3812
  } catch (error) {
3813
+ if (requestAbort.signal.aborted || res.destroyed) return;
2387
3814
  res.writeHead(500, { "content-type": "application/json" });
2388
3815
  res.end(JSON.stringify({ error: {
2389
3816
  name: "HarnessError",
2390
3817
  message: String(error)
2391
3818
  } }));
3819
+ } finally {
3820
+ req.off("aborted", abortRequest);
3821
+ res.off("close", abortOnPrematureClose);
2392
3822
  }
2393
3823
  });
2394
3824
  await new Promise((resolve) => server.listen(0, "127.0.0.1", resolve));