@peopl-health/nexus 5.44.0-dev.6870 → 5.44.0-dev.6882
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/README.md +0 -32
- package/lib/clinical/services/assistantService.js +3 -0
- package/lib/clinical/services/clinicalAirtableService.js +18 -0
- package/lib/core/NexusMessaging.js +0 -5
- package/lib/core/workflowRunner.js +35 -0
- package/lib/index.d.ts +4 -6
- package/lib/index.js +2 -0
- package/lib/services/airtableService.js +0 -7
- package/package.json +1 -1
- package/lib/core/WorkflowRunner.js +0 -265
- package/lib/models/deferredWorkModel.js +0 -24
package/README.md
CHANGED
|
@@ -116,38 +116,6 @@ const bus = nexus.getMessaging().getEventBus();
|
|
|
116
116
|
bus.on('message:received', (m) => console.log('rx', m.id));
|
|
117
117
|
```
|
|
118
118
|
|
|
119
|
-
## Workflows (Optional)
|
|
120
|
-
|
|
121
|
-
Register a named unit of work, deduplicated by a key you derive from the trigger.
|
|
122
|
-
|
|
123
|
-
```js
|
|
124
|
-
const runner = nexus.getMessaging().getWorkflowRunner();
|
|
125
|
-
|
|
126
|
-
await runner.register({
|
|
127
|
-
kind: 'triage-projection',
|
|
128
|
-
dedupeKey: (trigger) => String(trigger.submissionId),
|
|
129
|
-
prepare: async (trigger, checkpoints, save) => { /* ... */ },
|
|
130
|
-
retrySchedule: [5, 10, 20, 40], // minutes; omit for fire-and-forget
|
|
131
|
-
});
|
|
132
|
-
|
|
133
|
-
await runner.enqueue('triage-projection', { submissionId });
|
|
134
|
-
```
|
|
135
|
-
|
|
136
|
-
With `retrySchedule` the trigger and a checkpoint journal are stored in Mongo and the schedule runs
|
|
137
|
-
on an in-process timer — deferred workflows never touch the queue adapter, so they add no Redis load.
|
|
138
|
-
`prepare` resumes from `checkpoints`; call `save(patch)` before each external call so a retry does
|
|
139
|
-
not repeat completed work. Throw an error carrying `permanent: true` to give up immediately instead
|
|
140
|
-
of walking the schedule; the Airtable helpers tag their own permanent 4xx errors that way, in
|
|
141
|
-
`withAirtableRetry`, which all five of them route through. They rethrow on failure too, so an
|
|
142
|
-
Airtable outage reaches `prepare` as an exception and walks the retry schedule on its own rather
|
|
143
|
-
than reading as success.
|
|
144
|
-
|
|
145
|
-
Recovery is automatic: registering sweeps for work left behind by a previous process, and the runner
|
|
146
|
-
keeps sweeping so work whose worker died mid-`prepare` is reclaimed once its claim goes stale.
|
|
147
|
-
`enqueue` on an abandoned key revives it, keeping its checkpoints. Work is retained for 24 hours
|
|
148
|
-
after its last update and then expires — the dedupe key expires with it, so idempotency has the
|
|
149
|
-
same horizon as retention.
|
|
150
|
-
|
|
151
119
|
## Assistants (Optional)
|
|
152
120
|
|
|
153
121
|
Register assistant classes and (optionally) a custom resolver. OpenAI is supported via `llm: 'openai'`.
|
|
@@ -15,6 +15,7 @@ const { cleanupFiles } = require('../../helpers/filesHelper.js');
|
|
|
15
15
|
const { getMessages, countMessages } = require('../../services/messageService');
|
|
16
16
|
|
|
17
17
|
const { getAssistantById } = require('./assistantResolver');
|
|
18
|
+
const { recordPausedThreadMessage } = require('./clinicalAirtableService');
|
|
18
19
|
|
|
19
20
|
const MEDIA_BACKLOG_LIMIT = 10;
|
|
20
21
|
|
|
@@ -111,6 +112,8 @@ const preProcessMessagesCore = async (code, message_ = null, thread) => {
|
|
|
111
112
|
return { shouldProcess: false, messages: null, timings, pendingMediaUpdates: [] };
|
|
112
113
|
}
|
|
113
114
|
|
|
115
|
+
if (thread.stopped) await recordPausedThreadMessage({ code, text: lastMessage[0].plainBody || lastMessage[0].body });
|
|
116
|
+
|
|
114
117
|
const backlog = await fetchMediaBacklog(code, lastMessage[0]);
|
|
115
118
|
const replies = [...backlog, ...lastMessage];
|
|
116
119
|
|
|
@@ -12,6 +12,8 @@ const recentIntakeCache = new MapCache({ maxSize: 1000, ttl: 15 * 60 * 1000 });
|
|
|
12
12
|
const EMERGENCY_MERGE_WINDOW_MS = 5 * 60 * 1000;
|
|
13
13
|
const EMERGENCY_URGENCY_RANK = { routine: 0, urgent: 1, asap: 2, stat: 3 };
|
|
14
14
|
const recentEmergencyCache = new MapCache({ maxSize: 500, ttl: EMERGENCY_MERGE_WINDOW_MS });
|
|
15
|
+
const PAUSED_ALERT_WINDOW_MS = 4 * 60 * 60 * 1000;
|
|
16
|
+
const recentPausedAlertCache = new MapCache({ maxSize: 1000, ttl: PAUSED_ALERT_WINDOW_MS });
|
|
15
17
|
const RECORD_ID_RX = /^rec[A-Za-z0-9]+$/;
|
|
16
18
|
|
|
17
19
|
const MEDICAL_URGENCY_TO_EMERGENCY_LEVEL = {
|
|
@@ -97,6 +99,20 @@ async function recordUnresolvedRequest({ code, trace, question, details, resolut
|
|
|
97
99
|
});
|
|
98
100
|
}
|
|
99
101
|
|
|
102
|
+
async function recordPausedThreadMessage({ code, text }) {
|
|
103
|
+
if (recentPausedAlertCache.get(code)) return false;
|
|
104
|
+
const isWritten = await recordUnresolvedRequest({
|
|
105
|
+
code,
|
|
106
|
+
trace: null,
|
|
107
|
+
question: text || 'El paciente envió un archivo sin texto.',
|
|
108
|
+
details: `Bot en pausa: el paciente ${code} escribió y nadie ha respondido. Revisar la conversación.`,
|
|
109
|
+
category: 'contact_team_member',
|
|
110
|
+
urgency: 'soon',
|
|
111
|
+
});
|
|
112
|
+
if (isWritten) recentPausedAlertCache.set(code, true);
|
|
113
|
+
return isWritten;
|
|
114
|
+
}
|
|
115
|
+
|
|
100
116
|
async function readEmergencyRow(recordId, code) {
|
|
101
117
|
try {
|
|
102
118
|
const open = await airtableService.getRecordByFilter(Monitoreo_ID, 'emergencies', `RECORD_ID() = '${recordId}'`, null);
|
|
@@ -161,6 +177,7 @@ async function recordEmergency({ code, trace, symptom, suspectedCondition = '',
|
|
|
161
177
|
function _clearRecentIntakeCache() {
|
|
162
178
|
recentIntakeCache.clear();
|
|
163
179
|
recentEmergencyCache.clear();
|
|
180
|
+
recentPausedAlertCache.clear();
|
|
164
181
|
}
|
|
165
182
|
|
|
166
183
|
async function recordClinicalIntake({ code, trace, messageRaw, source = 'self', isLive = null }) {
|
|
@@ -379,6 +396,7 @@ async function recordClusterRecommendationsMirror({ code, triageRecordId, step1
|
|
|
379
396
|
|
|
380
397
|
module.exports = {
|
|
381
398
|
recordUnresolvedRequest,
|
|
399
|
+
recordPausedThreadMessage,
|
|
382
400
|
recordEmergency,
|
|
383
401
|
recordClinicalIntake,
|
|
384
402
|
recordReminder,
|
|
@@ -33,7 +33,6 @@ const { isLlmMonitorEnabled } = require('../clinical/flags/llmMonitorConfig');
|
|
|
33
33
|
const { createQueueAdapter } = require('../queue');
|
|
34
34
|
const { ScheduledMessageJob } = require('../jobs/ScheduledMessageJob');
|
|
35
35
|
const { TemplateApprovalJob } = require('../jobs/TemplateApprovalJob');
|
|
36
|
-
const { WorkflowRunner } = require('./WorkflowRunner');
|
|
37
36
|
|
|
38
37
|
const { PhiProcessor } = require('./PhiProcessor');
|
|
39
38
|
|
|
@@ -118,8 +117,6 @@ class NexusMessaging {
|
|
|
118
117
|
queueAdapter: this.queueAdapter,
|
|
119
118
|
});
|
|
120
119
|
|
|
121
|
-
this.workflowRunner = new WorkflowRunner({ queueAdapter: this.queueAdapter });
|
|
122
|
-
|
|
123
120
|
this.phiProcessor = new PhiProcessor({
|
|
124
121
|
encode: config.phi?.encode || false,
|
|
125
122
|
ner: config.phi?.ner || null,
|
|
@@ -303,7 +300,6 @@ class NexusMessaging {
|
|
|
303
300
|
getAssistantProcessor() { return this.assistantProcessor; }
|
|
304
301
|
getLlmMonitor() { return this.llmMonitor; }
|
|
305
302
|
getQueueAdapter() { return this.queueAdapter; }
|
|
306
|
-
getWorkflowRunner() { return this.workflowRunner; }
|
|
307
303
|
getPhiProcessor() { return this.phiProcessor; }
|
|
308
304
|
isConnected() { return this.provider?.getConnectionStatus() ?? false; }
|
|
309
305
|
isProcessing(chatId) { return this.batchingManager.isProcessing(chatId); }
|
|
@@ -744,7 +740,6 @@ class NexusMessaging {
|
|
|
744
740
|
async disconnect() {
|
|
745
741
|
this.queueReconciliation?.stop();
|
|
746
742
|
if (this.provider) await this.provider.disconnect();
|
|
747
|
-
if (this.workflowRunner) this.workflowRunner.stop();
|
|
748
743
|
if (this.queueAdapter) await this.queueAdapter.shutdown();
|
|
749
744
|
this.events.removeAllListeners();
|
|
750
745
|
}
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
function createWorkflowRunner({ queueAdapter } = {}) {
|
|
2
|
+
if (!queueAdapter) throw new Error('createWorkflowRunner requires a queueAdapter');
|
|
3
|
+
const workflows = new Map();
|
|
4
|
+
|
|
5
|
+
async function register(workflow = {}) {
|
|
6
|
+
if (!workflow.kind || typeof workflow.kind !== 'string') throw new Error('workflow requires a kind');
|
|
7
|
+
if (workflow.kind.includes('__')) throw new Error(`workflow '${workflow.kind}' must not contain '__' (reserved as the jobId delimiter)`);
|
|
8
|
+
if (typeof workflow.dedupeKey !== 'function') throw new Error(`workflow '${workflow.kind}' requires a dedupeKey function`);
|
|
9
|
+
if (typeof workflow.prepare !== 'function') throw new Error(`workflow '${workflow.kind}' requires a prepare function`);
|
|
10
|
+
if (workflows.has(workflow.kind)) throw new Error(`workflow '${workflow.kind}' is already registered`);
|
|
11
|
+
workflows.set(workflow.kind, workflow);
|
|
12
|
+
try {
|
|
13
|
+
await queueAdapter.process(workflow.kind, (trigger) => workflow.prepare(trigger));
|
|
14
|
+
} catch (error) {
|
|
15
|
+
workflows.delete(workflow.kind);
|
|
16
|
+
throw error;
|
|
17
|
+
}
|
|
18
|
+
return workflow;
|
|
19
|
+
}
|
|
20
|
+
|
|
21
|
+
function enqueue(kind, trigger, options = {}) {
|
|
22
|
+
const workflow = workflows.get(kind);
|
|
23
|
+
if (!workflow) throw new Error(`no workflow registered for '${kind}'`);
|
|
24
|
+
const dedupeKey = workflow.dedupeKey(trigger);
|
|
25
|
+
if (dedupeKey == null || (typeof dedupeKey === 'string' && !dedupeKey.trim())) {
|
|
26
|
+
throw new Error(`workflow '${kind}' produced an empty dedupe key`);
|
|
27
|
+
}
|
|
28
|
+
if (typeof dedupeKey !== 'string') throw new Error(`workflow '${kind}' produced a non-string dedupe key`);
|
|
29
|
+
return queueAdapter.enqueue(kind, trigger, { ...options, jobId: `${kind}__${dedupeKey}` });
|
|
30
|
+
}
|
|
31
|
+
|
|
32
|
+
return { register, enqueue };
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
module.exports = { createWorkflowRunner };
|
package/lib/index.d.ts
CHANGED
|
@@ -243,7 +243,6 @@ declare module '@peopl-health/nexus' {
|
|
|
243
243
|
getAssistantProcessor(): AssistantProcessor;
|
|
244
244
|
getLlmMonitor(): { start(options?: { scheduleDaily?: boolean }): Promise<{ enabled: boolean }> } | null;
|
|
245
245
|
initializeLlmMonitor(): Promise<{ enabled: boolean; error?: string } | undefined>;
|
|
246
|
-
getWorkflowRunner(): WorkflowRunner;
|
|
247
246
|
processInstruction(code: string, instruction: string, role?: string, options?: { triggeredBy?: string }): Promise<void>;
|
|
248
247
|
processSystemMessage(code: string, messages: string | string[], role?: string, options?: { triggeredBy?: string; reply?: boolean }): Promise<void>;
|
|
249
248
|
isConnected(): boolean;
|
|
@@ -456,17 +455,16 @@ declare module '@peopl-health/nexus' {
|
|
|
456
455
|
export interface Workflow {
|
|
457
456
|
kind: string;
|
|
458
457
|
dedupeKey: (trigger: any) => string;
|
|
459
|
-
prepare: (trigger: any
|
|
460
|
-
retrySchedule?: number[];
|
|
458
|
+
prepare: (trigger: any) => Promise<any>;
|
|
461
459
|
}
|
|
462
460
|
|
|
463
461
|
export interface WorkflowRunner {
|
|
464
462
|
register(workflow: Workflow): Promise<Workflow>;
|
|
465
|
-
enqueue(kind: string, trigger: any, options?: QueueJobOptions): Promise<string
|
|
466
|
-
sweep(only?: string[]): Promise<string[]>;
|
|
467
|
-
stop(): void;
|
|
463
|
+
enqueue(kind: string, trigger: any, options?: QueueJobOptions): Promise<string>;
|
|
468
464
|
}
|
|
469
465
|
|
|
466
|
+
export function createWorkflowRunner(options: { queueAdapter: QueueAdapter }): WorkflowRunner;
|
|
467
|
+
|
|
470
468
|
// Memory System
|
|
471
469
|
export interface PatientMemoryDocument {
|
|
472
470
|
_id: any;
|
package/lib/index.js
CHANGED
|
@@ -27,6 +27,7 @@ const { BaileysProvider } = require('./adapters/BaileysProvider');
|
|
|
27
27
|
const { setPreprocessingHandler, hasPreprocessingHandler, invokePreprocessingHandler } = require('./services/preprocessingService');
|
|
28
28
|
const { requestIdMiddleware, getRequestId } = require('./middleware/requestId');
|
|
29
29
|
const { QueueAdapter, LocalQueueAdapter, RedisQueueAdapter, createQueueAdapter, registerQueueAdapter } = require('./queue');
|
|
30
|
+
const { createWorkflowRunner } = require('./core/workflowRunner');
|
|
30
31
|
const routes = require('./routes');
|
|
31
32
|
const { resetAll } = require('./config/lifecycle');
|
|
32
33
|
const { EvalProvider } = require('./eval/EvalProvider');
|
|
@@ -220,6 +221,7 @@ class Nexus {
|
|
|
220
221
|
}
|
|
221
222
|
|
|
222
223
|
module.exports = {
|
|
224
|
+
createWorkflowRunner,
|
|
223
225
|
Nexus,
|
|
224
226
|
TwilioProvider,
|
|
225
227
|
BaileysProvider,
|
|
@@ -15,11 +15,6 @@ const AIRTABLE_API_URL = 'https://api.airtable.com/v0';
|
|
|
15
15
|
const AXIOS_TIMEOUT_CODE = 'ECONNABORTED';
|
|
16
16
|
const RETRY_BASE_DELAY_MS = 1000;
|
|
17
17
|
|
|
18
|
-
function isPermanentAirtableError(error) {
|
|
19
|
-
const status = error?.statusCode;
|
|
20
|
-
return typeof status === 'number' && status >= 400 && status < 500 && status !== 429;
|
|
21
|
-
}
|
|
22
|
-
|
|
23
18
|
let isEvalMode = false;
|
|
24
19
|
|
|
25
20
|
function setEvalMode(enabled) {
|
|
@@ -47,7 +42,6 @@ async function withAirtableRetry(operation, label, { retryNetworkErrors = true }
|
|
|
47
42
|
const isTransient = TRANSIENT_STATUS_CODES.includes(error.statusCode)
|
|
48
43
|
|| (retryNetworkErrors && isTransientNetworkError(error));
|
|
49
44
|
if (!isTransient || attempt === MAX_ATTEMPTS) {
|
|
50
|
-
if (isPermanentAirtableError(error)) error.permanent = true;
|
|
51
45
|
throw error;
|
|
52
46
|
}
|
|
53
47
|
const baseDelay = error.statusCode === RATE_LIMITED_STATUS ? RATE_LIMIT_DELAY_MS : RETRY_BASE_DELAY_MS;
|
|
@@ -235,7 +229,6 @@ async function upsertRecord(baseID, tableName, fields, { mergeOn } = {}, context
|
|
|
235
229
|
|
|
236
230
|
module.exports = {
|
|
237
231
|
setEvalMode,
|
|
238
|
-
isPermanentAirtableError,
|
|
239
232
|
addRecord,
|
|
240
233
|
getRecords,
|
|
241
234
|
getRecordByFilter,
|
package/package.json
CHANGED
|
@@ -1,265 +0,0 @@
|
|
|
1
|
-
const { describeThrown } = require('../utils/errorUtils');
|
|
2
|
-
const { logger } = require('../utils/logger');
|
|
3
|
-
|
|
4
|
-
const { DeferredWork, RETENTION_MS } = require('../models/deferredWorkModel');
|
|
5
|
-
|
|
6
|
-
const STALE_CLAIM_MS = 5 * 60 * 1000;
|
|
7
|
-
const MINUTE_MS = 60 * 1000;
|
|
8
|
-
const INDEX_TIMEOUT_MS = 1000;
|
|
9
|
-
const SWEEP_INTERVAL_MS = 60 * 1000;
|
|
10
|
-
|
|
11
|
-
let indexesReady = null;
|
|
12
|
-
|
|
13
|
-
function ensureIndexes() {
|
|
14
|
-
if (!indexesReady) {
|
|
15
|
-
indexesReady = DeferredWork.init().catch((error) => {
|
|
16
|
-
indexesReady = null;
|
|
17
|
-
logger.error('[WorkflowRunner] Index build failed, dedupe and retention are not guaranteed', { error: error.message });
|
|
18
|
-
});
|
|
19
|
-
}
|
|
20
|
-
return indexesReady;
|
|
21
|
-
}
|
|
22
|
-
|
|
23
|
-
function withIndexTimeout(promise) {
|
|
24
|
-
let timer;
|
|
25
|
-
const timeout = new Promise((resolve) => {
|
|
26
|
-
timer = setTimeout(resolve, INDEX_TIMEOUT_MS);
|
|
27
|
-
timer.unref();
|
|
28
|
-
});
|
|
29
|
-
return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
|
|
30
|
-
}
|
|
31
|
-
|
|
32
|
-
class WorkflowRunner {
|
|
33
|
-
constructor({ queueAdapter, sweepIntervalMs = SWEEP_INTERVAL_MS } = {}) {
|
|
34
|
-
if (!queueAdapter) throw new Error('WorkflowRunner requires a queueAdapter');
|
|
35
|
-
this.queueAdapter = queueAdapter;
|
|
36
|
-
this.sweepIntervalMs = sweepIntervalMs;
|
|
37
|
-
this.sweepTimer = null;
|
|
38
|
-
this.timers = new Map();
|
|
39
|
-
this.workflows = new Map();
|
|
40
|
-
this.stopped = false;
|
|
41
|
-
}
|
|
42
|
-
|
|
43
|
-
async register(workflow = {}) {
|
|
44
|
-
if (!workflow.kind || typeof workflow.kind !== 'string') throw new Error('workflow requires a kind');
|
|
45
|
-
if (workflow.kind.includes('__')) throw new Error(`workflow '${workflow.kind}' must not contain '__' (reserved as the jobId delimiter)`);
|
|
46
|
-
if (typeof workflow.dedupeKey !== 'function') throw new Error(`workflow '${workflow.kind}' requires a dedupeKey function`);
|
|
47
|
-
if (typeof workflow.prepare !== 'function') throw new Error(`workflow '${workflow.kind}' requires a prepare function`);
|
|
48
|
-
if (this.workflows.has(workflow.kind)) throw new Error(`workflow '${workflow.kind}' is already registered`);
|
|
49
|
-
if (workflow.retrySchedule !== undefined) {
|
|
50
|
-
if (!Array.isArray(workflow.retrySchedule) || !workflow.retrySchedule.length) {
|
|
51
|
-
throw new Error(`workflow '${workflow.kind}' requires a non-empty retrySchedule array`);
|
|
52
|
-
}
|
|
53
|
-
if (workflow.retrySchedule.some((minutes) => !Number.isFinite(minutes) || minutes <= 0)) {
|
|
54
|
-
throw new Error(`workflow '${workflow.kind}' requires retrySchedule entries to be positive minutes`);
|
|
55
|
-
}
|
|
56
|
-
if (workflow.retrySchedule.some((minutes) => minutes * MINUTE_MS >= RETENTION_MS)) {
|
|
57
|
-
throw new Error(`workflow '${workflow.kind}' requires retrySchedule entries under the ${RETENTION_MS / MINUTE_MS}m retention, or the record expires before the attempt fires`);
|
|
58
|
-
}
|
|
59
|
-
}
|
|
60
|
-
|
|
61
|
-
this.workflows.set(workflow.kind, workflow);
|
|
62
|
-
if (workflow.retrySchedule) {
|
|
63
|
-
await this.sweep([workflow.kind]).catch((error) => {
|
|
64
|
-
logger.error('[WorkflowRunner] Sweep on register failed', { kind: workflow.kind, error: error.message });
|
|
65
|
-
});
|
|
66
|
-
this._startSweeping();
|
|
67
|
-
return workflow;
|
|
68
|
-
}
|
|
69
|
-
|
|
70
|
-
try {
|
|
71
|
-
await this.queueAdapter.process(workflow.kind, (trigger) => workflow.prepare(trigger));
|
|
72
|
-
} catch (error) {
|
|
73
|
-
this.workflows.delete(workflow.kind);
|
|
74
|
-
throw error;
|
|
75
|
-
}
|
|
76
|
-
return workflow;
|
|
77
|
-
}
|
|
78
|
-
|
|
79
|
-
enqueue(kind, trigger, options = {}) {
|
|
80
|
-
const workflow = this.workflows.get(kind);
|
|
81
|
-
if (!workflow) throw new Error(`no workflow registered for '${kind}'`);
|
|
82
|
-
const dedupeKey = workflow.dedupeKey(trigger);
|
|
83
|
-
if (dedupeKey == null || (typeof dedupeKey === 'string' && !dedupeKey.trim())) {
|
|
84
|
-
throw new Error(`workflow '${kind}' produced an empty dedupe key`);
|
|
85
|
-
}
|
|
86
|
-
if (typeof dedupeKey !== 'string') throw new Error(`workflow '${kind}' produced a non-string dedupe key`);
|
|
87
|
-
if (!workflow.retrySchedule) {
|
|
88
|
-
return this.queueAdapter.enqueue(kind, trigger, { ...options, jobId: `${kind}__${dedupeKey}` });
|
|
89
|
-
}
|
|
90
|
-
return this._armDeferred(workflow, trigger, dedupeKey);
|
|
91
|
-
}
|
|
92
|
-
|
|
93
|
-
async sweep(only = null) {
|
|
94
|
-
const deferredKinds = [...this.workflows.values()].filter((w) => w.retrySchedule).map((w) => w.kind);
|
|
95
|
-
const kinds = only ? deferredKinds.filter((kind) => only.includes(kind)) : deferredKinds;
|
|
96
|
-
if (!kinds.length) return [];
|
|
97
|
-
|
|
98
|
-
const orphaned = await DeferredWork.find({
|
|
99
|
-
kind: { $in: kinds },
|
|
100
|
-
$or: [
|
|
101
|
-
{ status: 'pending' },
|
|
102
|
-
{ status: 'processing', claimedAt: { $lt: new Date(Date.now() - STALE_CLAIM_MS) } }
|
|
103
|
-
]
|
|
104
|
-
});
|
|
105
|
-
|
|
106
|
-
const rearmed = orphaned.map((work) => this._arm(
|
|
107
|
-
this.workflows.get(work.kind),
|
|
108
|
-
work._id,
|
|
109
|
-
this._delayUntilDue(work),
|
|
110
|
-
));
|
|
111
|
-
if (rearmed.length) logger.info('[WorkflowRunner] Swept', { kinds, rearmed: rearmed.length });
|
|
112
|
-
return rearmed;
|
|
113
|
-
}
|
|
114
|
-
|
|
115
|
-
_startSweeping() {
|
|
116
|
-
if (this.stopped || this.sweepTimer || !this.sweepIntervalMs) return;
|
|
117
|
-
this.sweepTimer = setInterval(() => {
|
|
118
|
-
this.sweep().catch((error) => logger.error('[WorkflowRunner] Periodic sweep failed', { error: error.message }));
|
|
119
|
-
}, this.sweepIntervalMs);
|
|
120
|
-
this.sweepTimer.unref();
|
|
121
|
-
}
|
|
122
|
-
|
|
123
|
-
stop() {
|
|
124
|
-
this.stopped = true;
|
|
125
|
-
if (this.sweepTimer) clearInterval(this.sweepTimer);
|
|
126
|
-
this.sweepTimer = null;
|
|
127
|
-
for (const timer of this.timers.values()) clearTimeout(timer);
|
|
128
|
-
this.timers.clear();
|
|
129
|
-
}
|
|
130
|
-
|
|
131
|
-
async _armDeferred(workflow, trigger, dedupeKey) {
|
|
132
|
-
const { kind } = workflow;
|
|
133
|
-
await withIndexTimeout(ensureIndexes());
|
|
134
|
-
const work = await DeferredWork.findOne({ kind, dedupeKey })
|
|
135
|
-
|| await DeferredWork.create({ kind, dedupeKey, trigger }).catch(async (error) => {
|
|
136
|
-
if (error.code !== 11000) throw error;
|
|
137
|
-
return await DeferredWork.findOne({ kind, dedupeKey });
|
|
138
|
-
});
|
|
139
|
-
if (work.status === 'abandoned') {
|
|
140
|
-
const { modifiedCount } = await DeferredWork.updateOne(
|
|
141
|
-
{ _id: work._id, status: 'abandoned' },
|
|
142
|
-
{ $set: { status: 'pending', attempt: 0, nextAttemptAt: new Date(), lastError: null, claimedAt: null } }
|
|
143
|
-
);
|
|
144
|
-
if (!modifiedCount) return null;
|
|
145
|
-
logger.info('[WorkflowRunner] Re-arming abandoned work', { kind, dedupeKey });
|
|
146
|
-
return this._arm(workflow, work._id, 0);
|
|
147
|
-
}
|
|
148
|
-
if (work.status !== 'pending') {
|
|
149
|
-
logger.info('[WorkflowRunner] Skipping, work is in flight or complete', { kind, dedupeKey, status: work.status });
|
|
150
|
-
return null;
|
|
151
|
-
}
|
|
152
|
-
return this._arm(workflow, work._id, this._delayUntilDue(work));
|
|
153
|
-
}
|
|
154
|
-
|
|
155
|
-
_delayUntilDue(work) {
|
|
156
|
-
return Math.max(0, new Date(work.nextAttemptAt).getTime() - Date.now());
|
|
157
|
-
}
|
|
158
|
-
|
|
159
|
-
_arm(workflow, workId, delayMs) {
|
|
160
|
-
const id = String(workId);
|
|
161
|
-
// An attempt still in flight when stop() lands must not re-arm after shutdown. The
|
|
162
|
-
// record stays pending in Mongo, so the next process recovers it on its sweep.
|
|
163
|
-
if (this.stopped) return id;
|
|
164
|
-
clearTimeout(this.timers.get(id));
|
|
165
|
-
const timer = setTimeout(() => {
|
|
166
|
-
this.timers.delete(id);
|
|
167
|
-
this._runDeferred(workflow, id).catch((error) => {
|
|
168
|
-
logger.error('[WorkflowRunner] Deferred run failed', { kind: workflow.kind, deferredWorkId: id, error: error.message });
|
|
169
|
-
});
|
|
170
|
-
}, delayMs);
|
|
171
|
-
timer.unref();
|
|
172
|
-
this.timers.set(id, timer);
|
|
173
|
-
return id;
|
|
174
|
-
}
|
|
175
|
-
|
|
176
|
-
async _runDeferred(workflow, deferredWorkId) {
|
|
177
|
-
const claimedAt = new Date();
|
|
178
|
-
const prior = await DeferredWork.findOneAndUpdate(
|
|
179
|
-
{
|
|
180
|
-
_id: deferredWorkId,
|
|
181
|
-
$or: [
|
|
182
|
-
{ status: 'pending', nextAttemptAt: { $lte: claimedAt } },
|
|
183
|
-
{ status: 'processing', claimedAt: { $lt: new Date(claimedAt.getTime() - STALE_CLAIM_MS) } }
|
|
184
|
-
]
|
|
185
|
-
},
|
|
186
|
-
{ $set: { status: 'processing', claimedAt } },
|
|
187
|
-
{ new: false }
|
|
188
|
-
);
|
|
189
|
-
if (!prior) {
|
|
190
|
-
logger.info('[WorkflowRunner] Claimed elsewhere or not due', { kind: workflow.kind, deferredWorkId });
|
|
191
|
-
return { claimed: false };
|
|
192
|
-
}
|
|
193
|
-
|
|
194
|
-
const reclaimed = prior.status === 'processing';
|
|
195
|
-
const work = { ...prior.toObject(), attempt: prior.attempt + (reclaimed ? 1 : 0) };
|
|
196
|
-
if (reclaimed) await this._write(work, claimedAt, { attempt: work.attempt });
|
|
197
|
-
|
|
198
|
-
const checkpoints = { ...(work.checkpoints || {}) };
|
|
199
|
-
const save = async (patch) => {
|
|
200
|
-
Object.assign(checkpoints, patch);
|
|
201
|
-
const $set = {};
|
|
202
|
-
for (const [key, value] of Object.entries(patch || {})) $set[`checkpoints.${key}`] = value;
|
|
203
|
-
if (Object.keys($set).length) await this._write(work, claimedAt, $set);
|
|
204
|
-
};
|
|
205
|
-
|
|
206
|
-
try {
|
|
207
|
-
await workflow.prepare(work.trigger, checkpoints, save);
|
|
208
|
-
} catch (error) {
|
|
209
|
-
return await this._rescheduleOrAbandon(workflow, work, claimedAt, error);
|
|
210
|
-
}
|
|
211
|
-
await this._write(work, claimedAt, { status: 'done', claimedAt: null });
|
|
212
|
-
return { status: 'done' };
|
|
213
|
-
}
|
|
214
|
-
|
|
215
|
-
// matchedCount, not modifiedCount: the fence is "do we still hold the claim", and
|
|
216
|
-
// modifiedCount is also 0 for a write that matched but set an identical value - a
|
|
217
|
-
// save() rewriting an unchanged checkpoint is not a lost claim.
|
|
218
|
-
async _write(work, claimedAt, $set) {
|
|
219
|
-
const { matchedCount } = await DeferredWork.updateOne({ _id: work._id, claimedAt }, { $set });
|
|
220
|
-
if (!matchedCount) logger.warn('[WorkflowRunner] Claim lost, write discarded', { deferredWorkId: String(work._id) });
|
|
221
|
-
return matchedCount > 0;
|
|
222
|
-
}
|
|
223
|
-
|
|
224
|
-
async _rescheduleOrAbandon(workflow, work, claimedAt, error) {
|
|
225
|
-
// prepare() is consumer code and can throw anything, including null. Dereferencing
|
|
226
|
-
// it here used to raise a TypeError out of this handler, so neither branch ran and
|
|
227
|
-
// the record stayed `processing` until stale recovery hit the same failure again.
|
|
228
|
-
const lastError = describeThrown(error);
|
|
229
|
-
const delayMinutes = workflow.retrySchedule[work.attempt];
|
|
230
|
-
if (error?.permanent === true || delayMinutes === undefined) {
|
|
231
|
-
await this._write(work, claimedAt, { status: 'abandoned', lastError, claimedAt: null });
|
|
232
|
-
logger.error('[WorkflowRunner] Abandoned', {
|
|
233
|
-
kind: workflow.kind,
|
|
234
|
-
dedupeKey: work.dedupeKey,
|
|
235
|
-
attempt: work.attempt,
|
|
236
|
-
permanent: error?.permanent === true,
|
|
237
|
-
error: lastError
|
|
238
|
-
});
|
|
239
|
-
return { status: 'abandoned' };
|
|
240
|
-
}
|
|
241
|
-
|
|
242
|
-
const attempt = work.attempt + 1;
|
|
243
|
-
const delayMs = delayMinutes * MINUTE_MS;
|
|
244
|
-
const kept = await this._write(work, claimedAt, {
|
|
245
|
-
status: 'pending',
|
|
246
|
-
attempt,
|
|
247
|
-
nextAttemptAt: new Date(Date.now() + delayMs),
|
|
248
|
-
lastError,
|
|
249
|
-
claimedAt: null
|
|
250
|
-
});
|
|
251
|
-
if (!kept) return { claimed: false };
|
|
252
|
-
|
|
253
|
-
logger.warn('[WorkflowRunner] Rescheduled', {
|
|
254
|
-
kind: workflow.kind,
|
|
255
|
-
dedupeKey: work.dedupeKey,
|
|
256
|
-
attempt,
|
|
257
|
-
delayMinutes,
|
|
258
|
-
error: lastError
|
|
259
|
-
});
|
|
260
|
-
this._arm(workflow, work._id, delayMs);
|
|
261
|
-
return { status: 'pending', attempt };
|
|
262
|
-
}
|
|
263
|
-
}
|
|
264
|
-
|
|
265
|
-
module.exports = { WorkflowRunner };
|
|
@@ -1,24 +0,0 @@
|
|
|
1
|
-
const mongoose = require('mongoose');
|
|
2
|
-
|
|
3
|
-
const STATUS_VALUES = ['pending', 'processing', 'done', 'abandoned'];
|
|
4
|
-
const RETENTION_MS = 24 * 60 * 60 * 1000;
|
|
5
|
-
|
|
6
|
-
const deferredWorkSchema = new mongoose.Schema({
|
|
7
|
-
kind: { type: String, required: true },
|
|
8
|
-
dedupeKey: { type: String, required: true },
|
|
9
|
-
trigger: { type: mongoose.Schema.Types.Mixed, default: null },
|
|
10
|
-
checkpoints: { type: mongoose.Schema.Types.Mixed, default: () => ({}) },
|
|
11
|
-
status: { type: String, enum: STATUS_VALUES, default: 'pending' },
|
|
12
|
-
attempt: { type: Number, default: 0 },
|
|
13
|
-
nextAttemptAt: { type: Date, default: Date.now },
|
|
14
|
-
claimedAt: { type: Date, default: null },
|
|
15
|
-
lastError: { type: String, default: null }
|
|
16
|
-
}, { timestamps: true });
|
|
17
|
-
|
|
18
|
-
deferredWorkSchema.index({ kind: 1, dedupeKey: 1 }, { unique: true, name: 'dedupe_idx' });
|
|
19
|
-
deferredWorkSchema.index({ status: 1, nextAttemptAt: 1 }, { name: 'due_idx' });
|
|
20
|
-
deferredWorkSchema.index({ updatedAt: 1 }, { expireAfterSeconds: RETENTION_MS / 1000, name: 'ttl_idx' });
|
|
21
|
-
|
|
22
|
-
const DeferredWork = mongoose.model('DeferredWork', deferredWorkSchema);
|
|
23
|
-
|
|
24
|
-
module.exports = { DeferredWork, RETENTION_MS };
|