@peopl-health/nexus 5.44.0-dev.5413 → 5.44.0-dev.5415
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/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 -252
- 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; `addRecord` and `addLinkedRecord` tag their own permanent 4xx errors that
|
|
141
|
-
way. The read and update helpers (`getRecords`, `getRecordByFilter`, `updateRecordByFilter`) return
|
|
142
|
-
`undefined` on failure instead of throwing, so `prepare` must check their results and throw itself —
|
|
143
|
-
otherwise an outage reads as success and the work is marked done.
|
|
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'`.
|
|
@@ -32,7 +32,6 @@ const { isLlmMonitorEnabled } = require('../clinical/flags/llmMonitorConfig');
|
|
|
32
32
|
const { createQueueAdapter } = require('../queue');
|
|
33
33
|
const { ScheduledMessageJob } = require('../jobs/ScheduledMessageJob');
|
|
34
34
|
const { TemplateApprovalJob } = require('../jobs/TemplateApprovalJob');
|
|
35
|
-
const { WorkflowRunner } = require('./WorkflowRunner');
|
|
36
35
|
|
|
37
36
|
const { PhiProcessor } = require('./PhiProcessor');
|
|
38
37
|
|
|
@@ -117,8 +116,6 @@ class NexusMessaging {
|
|
|
117
116
|
queueAdapter: this.queueAdapter,
|
|
118
117
|
});
|
|
119
118
|
|
|
120
|
-
this.workflowRunner = new WorkflowRunner({ queueAdapter: this.queueAdapter });
|
|
121
|
-
|
|
122
119
|
this.phiProcessor = new PhiProcessor({
|
|
123
120
|
encode: config.phi?.encode || false,
|
|
124
121
|
ner: config.phi?.ner || null,
|
|
@@ -289,7 +286,6 @@ class NexusMessaging {
|
|
|
289
286
|
getAssistantProcessor() { return this.assistantProcessor; }
|
|
290
287
|
getLlmMonitor() { return this.llmMonitor; }
|
|
291
288
|
getQueueAdapter() { return this.queueAdapter; }
|
|
292
|
-
getWorkflowRunner() { return this.workflowRunner; }
|
|
293
289
|
getPhiProcessor() { return this.phiProcessor; }
|
|
294
290
|
isConnected() { return this.provider?.getConnectionStatus() ?? false; }
|
|
295
291
|
isProcessing(chatId) { return this.batchingManager.isProcessing(chatId); }
|
|
@@ -721,7 +717,6 @@ class NexusMessaging {
|
|
|
721
717
|
async disconnect() {
|
|
722
718
|
this.queueReconciliation?.stop();
|
|
723
719
|
if (this.provider) await this.provider.disconnect();
|
|
724
|
-
if (this.workflowRunner) this.workflowRunner.stop();
|
|
725
720
|
if (this.queueAdapter) await this.queueAdapter.shutdown();
|
|
726
721
|
this.events.removeAllListeners();
|
|
727
722
|
}
|
|
@@ -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
|
@@ -242,7 +242,6 @@ declare module '@peopl-health/nexus' {
|
|
|
242
242
|
getAssistantProcessor(): AssistantProcessor;
|
|
243
243
|
getLlmMonitor(): { start(options?: { scheduleDaily?: boolean }): Promise<{ enabled: boolean }> } | null;
|
|
244
244
|
initializeLlmMonitor(): Promise<{ enabled: boolean; error?: string } | undefined>;
|
|
245
|
-
getWorkflowRunner(): WorkflowRunner;
|
|
246
245
|
processInstruction(code: string, instruction: string, role?: string, options?: { triggeredBy?: string }): Promise<void>;
|
|
247
246
|
processSystemMessage(code: string, messages: string | string[], role?: string, options?: { triggeredBy?: string; reply?: boolean }): Promise<void>;
|
|
248
247
|
isConnected(): boolean;
|
|
@@ -455,17 +454,16 @@ declare module '@peopl-health/nexus' {
|
|
|
455
454
|
export interface Workflow {
|
|
456
455
|
kind: string;
|
|
457
456
|
dedupeKey: (trigger: any) => string;
|
|
458
|
-
prepare: (trigger: any
|
|
459
|
-
retrySchedule?: number[];
|
|
457
|
+
prepare: (trigger: any) => Promise<any>;
|
|
460
458
|
}
|
|
461
459
|
|
|
462
460
|
export interface WorkflowRunner {
|
|
463
461
|
register(workflow: Workflow): Promise<Workflow>;
|
|
464
|
-
enqueue(kind: string, trigger: any, options?: QueueJobOptions): Promise<string
|
|
465
|
-
sweep(only?: string[]): Promise<string[]>;
|
|
466
|
-
stop(): void;
|
|
462
|
+
enqueue(kind: string, trigger: any, options?: QueueJobOptions): Promise<string>;
|
|
467
463
|
}
|
|
468
464
|
|
|
465
|
+
export function createWorkflowRunner(options: { queueAdapter: QueueAdapter }): WorkflowRunner;
|
|
466
|
+
|
|
469
467
|
// Memory System
|
|
470
468
|
export interface PatientMemoryDocument {
|
|
471
469
|
_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,
|
|
@@ -11,11 +11,6 @@ const TRANSIENT_ERROR_CODES = ['ETIMEDOUT', 'ECONNRESET', 'ECONNREFUSED', 'ENOTF
|
|
|
11
11
|
const MAX_ATTEMPTS = 3;
|
|
12
12
|
const RETRY_BASE_DELAY_MS = 1000;
|
|
13
13
|
|
|
14
|
-
function isPermanentAirtableError(error) {
|
|
15
|
-
const status = error?.statusCode;
|
|
16
|
-
return typeof status === 'number' && status >= 400 && status < 500 && status !== 429;
|
|
17
|
-
}
|
|
18
|
-
|
|
19
14
|
let isEvalMode = false;
|
|
20
15
|
|
|
21
16
|
function setEvalMode(enabled) {
|
|
@@ -43,7 +38,6 @@ async function withAirtableRetry(operation, label, { retryNetworkErrors = true }
|
|
|
43
38
|
const isTransient = TRANSIENT_STATUS_CODES.includes(error.statusCode)
|
|
44
39
|
|| (retryNetworkErrors && isTransientNetworkError(error));
|
|
45
40
|
if (!isTransient || attempt === MAX_ATTEMPTS) {
|
|
46
|
-
if (isPermanentAirtableError(error)) error.permanent = true;
|
|
47
41
|
throw error;
|
|
48
42
|
}
|
|
49
43
|
const baseDelay = error.statusCode === RATE_LIMITED_STATUS ? RATE_LIMIT_DELAY_MS : RETRY_BASE_DELAY_MS;
|
|
@@ -175,7 +169,6 @@ async function addLinkedRecord(baseID, targetTable, fields, linkConfig, context
|
|
|
175
169
|
|
|
176
170
|
module.exports = {
|
|
177
171
|
setEvalMode,
|
|
178
|
-
isPermanentAirtableError,
|
|
179
172
|
addRecord,
|
|
180
173
|
getRecords,
|
|
181
174
|
getRecordByFilter,
|
package/package.json
CHANGED
|
@@ -1,252 +0,0 @@
|
|
|
1
|
-
const { logger } = require('../utils/logger');
|
|
2
|
-
|
|
3
|
-
const { DeferredWork, RETENTION_MS } = require('../models/deferredWorkModel');
|
|
4
|
-
|
|
5
|
-
const STALE_CLAIM_MS = 5 * 60 * 1000;
|
|
6
|
-
const MINUTE_MS = 60 * 1000;
|
|
7
|
-
const INDEX_TIMEOUT_MS = 1000;
|
|
8
|
-
const SWEEP_INTERVAL_MS = 60 * 1000;
|
|
9
|
-
|
|
10
|
-
let indexesReady = null;
|
|
11
|
-
|
|
12
|
-
function ensureIndexes() {
|
|
13
|
-
if (!indexesReady) {
|
|
14
|
-
indexesReady = DeferredWork.init().catch((error) => {
|
|
15
|
-
indexesReady = null;
|
|
16
|
-
logger.error('[WorkflowRunner] Index build failed, dedupe and retention are not guaranteed', { error: error.message });
|
|
17
|
-
});
|
|
18
|
-
}
|
|
19
|
-
return indexesReady;
|
|
20
|
-
}
|
|
21
|
-
|
|
22
|
-
function withIndexTimeout(promise) {
|
|
23
|
-
let timer;
|
|
24
|
-
const timeout = new Promise((resolve) => {
|
|
25
|
-
timer = setTimeout(resolve, INDEX_TIMEOUT_MS);
|
|
26
|
-
timer.unref();
|
|
27
|
-
});
|
|
28
|
-
return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
|
|
29
|
-
}
|
|
30
|
-
|
|
31
|
-
class WorkflowRunner {
|
|
32
|
-
constructor({ queueAdapter, sweepIntervalMs = SWEEP_INTERVAL_MS } = {}) {
|
|
33
|
-
if (!queueAdapter) throw new Error('WorkflowRunner requires a queueAdapter');
|
|
34
|
-
this.queueAdapter = queueAdapter;
|
|
35
|
-
this.sweepIntervalMs = sweepIntervalMs;
|
|
36
|
-
this.sweepTimer = null;
|
|
37
|
-
this.timers = new Map();
|
|
38
|
-
this.workflows = new Map();
|
|
39
|
-
}
|
|
40
|
-
|
|
41
|
-
async register(workflow = {}) {
|
|
42
|
-
if (!workflow.kind || typeof workflow.kind !== 'string') throw new Error('workflow requires a kind');
|
|
43
|
-
if (workflow.kind.includes('__')) throw new Error(`workflow '${workflow.kind}' must not contain '__' (reserved as the jobId delimiter)`);
|
|
44
|
-
if (typeof workflow.dedupeKey !== 'function') throw new Error(`workflow '${workflow.kind}' requires a dedupeKey function`);
|
|
45
|
-
if (typeof workflow.prepare !== 'function') throw new Error(`workflow '${workflow.kind}' requires a prepare function`);
|
|
46
|
-
if (this.workflows.has(workflow.kind)) throw new Error(`workflow '${workflow.kind}' is already registered`);
|
|
47
|
-
if (workflow.retrySchedule !== undefined) {
|
|
48
|
-
if (!Array.isArray(workflow.retrySchedule) || !workflow.retrySchedule.length) {
|
|
49
|
-
throw new Error(`workflow '${workflow.kind}' requires a non-empty retrySchedule array`);
|
|
50
|
-
}
|
|
51
|
-
if (workflow.retrySchedule.some((minutes) => !Number.isFinite(minutes) || minutes <= 0)) {
|
|
52
|
-
throw new Error(`workflow '${workflow.kind}' requires retrySchedule entries to be positive minutes`);
|
|
53
|
-
}
|
|
54
|
-
if (workflow.retrySchedule.some((minutes) => minutes * MINUTE_MS >= RETENTION_MS)) {
|
|
55
|
-
throw new Error(`workflow '${workflow.kind}' requires retrySchedule entries under the ${RETENTION_MS / MINUTE_MS}m retention, or the record expires before the attempt fires`);
|
|
56
|
-
}
|
|
57
|
-
}
|
|
58
|
-
|
|
59
|
-
this.workflows.set(workflow.kind, workflow);
|
|
60
|
-
if (workflow.retrySchedule) {
|
|
61
|
-
await this.sweep([workflow.kind]).catch((error) => {
|
|
62
|
-
logger.error('[WorkflowRunner] Sweep on register failed', { kind: workflow.kind, error: error.message });
|
|
63
|
-
});
|
|
64
|
-
this._startSweeping();
|
|
65
|
-
return workflow;
|
|
66
|
-
}
|
|
67
|
-
|
|
68
|
-
try {
|
|
69
|
-
await this.queueAdapter.process(workflow.kind, (trigger) => workflow.prepare(trigger));
|
|
70
|
-
} catch (error) {
|
|
71
|
-
this.workflows.delete(workflow.kind);
|
|
72
|
-
throw error;
|
|
73
|
-
}
|
|
74
|
-
return workflow;
|
|
75
|
-
}
|
|
76
|
-
|
|
77
|
-
enqueue(kind, trigger, options = {}) {
|
|
78
|
-
const workflow = this.workflows.get(kind);
|
|
79
|
-
if (!workflow) throw new Error(`no workflow registered for '${kind}'`);
|
|
80
|
-
const dedupeKey = workflow.dedupeKey(trigger);
|
|
81
|
-
if (dedupeKey == null || (typeof dedupeKey === 'string' && !dedupeKey.trim())) {
|
|
82
|
-
throw new Error(`workflow '${kind}' produced an empty dedupe key`);
|
|
83
|
-
}
|
|
84
|
-
if (typeof dedupeKey !== 'string') throw new Error(`workflow '${kind}' produced a non-string dedupe key`);
|
|
85
|
-
if (!workflow.retrySchedule) {
|
|
86
|
-
return this.queueAdapter.enqueue(kind, trigger, { ...options, jobId: `${kind}__${dedupeKey}` });
|
|
87
|
-
}
|
|
88
|
-
return this._armDeferred(workflow, trigger, dedupeKey);
|
|
89
|
-
}
|
|
90
|
-
|
|
91
|
-
async sweep(only = null) {
|
|
92
|
-
const deferredKinds = [...this.workflows.values()].filter((w) => w.retrySchedule).map((w) => w.kind);
|
|
93
|
-
const kinds = only ? deferredKinds.filter((kind) => only.includes(kind)) : deferredKinds;
|
|
94
|
-
if (!kinds.length) return [];
|
|
95
|
-
|
|
96
|
-
const orphaned = await DeferredWork.find({
|
|
97
|
-
kind: { $in: kinds },
|
|
98
|
-
$or: [
|
|
99
|
-
{ status: 'pending' },
|
|
100
|
-
{ status: 'processing', claimedAt: { $lt: new Date(Date.now() - STALE_CLAIM_MS) } }
|
|
101
|
-
]
|
|
102
|
-
});
|
|
103
|
-
|
|
104
|
-
const rearmed = orphaned.map((work) => this._arm(
|
|
105
|
-
this.workflows.get(work.kind),
|
|
106
|
-
work._id,
|
|
107
|
-
this._delayUntilDue(work),
|
|
108
|
-
));
|
|
109
|
-
if (rearmed.length) logger.info('[WorkflowRunner] Swept', { kinds, rearmed: rearmed.length });
|
|
110
|
-
return rearmed;
|
|
111
|
-
}
|
|
112
|
-
|
|
113
|
-
_startSweeping() {
|
|
114
|
-
if (this.sweepTimer || !this.sweepIntervalMs) return;
|
|
115
|
-
this.sweepTimer = setInterval(() => {
|
|
116
|
-
this.sweep().catch((error) => logger.error('[WorkflowRunner] Periodic sweep failed', { error: error.message }));
|
|
117
|
-
}, this.sweepIntervalMs);
|
|
118
|
-
this.sweepTimer.unref();
|
|
119
|
-
}
|
|
120
|
-
|
|
121
|
-
stop() {
|
|
122
|
-
if (this.sweepTimer) clearInterval(this.sweepTimer);
|
|
123
|
-
this.sweepTimer = null;
|
|
124
|
-
for (const timer of this.timers.values()) clearTimeout(timer);
|
|
125
|
-
this.timers.clear();
|
|
126
|
-
}
|
|
127
|
-
|
|
128
|
-
async _armDeferred(workflow, trigger, dedupeKey) {
|
|
129
|
-
const { kind } = workflow;
|
|
130
|
-
await withIndexTimeout(ensureIndexes());
|
|
131
|
-
const work = await DeferredWork.findOne({ kind, dedupeKey })
|
|
132
|
-
|| await DeferredWork.create({ kind, dedupeKey, trigger }).catch(async (error) => {
|
|
133
|
-
if (error.code !== 11000) throw error;
|
|
134
|
-
return await DeferredWork.findOne({ kind, dedupeKey });
|
|
135
|
-
});
|
|
136
|
-
if (work.status === 'abandoned') {
|
|
137
|
-
const { modifiedCount } = await DeferredWork.updateOne(
|
|
138
|
-
{ _id: work._id, status: 'abandoned' },
|
|
139
|
-
{ $set: { status: 'pending', attempt: 0, nextAttemptAt: new Date(), lastError: null, claimedAt: null } }
|
|
140
|
-
);
|
|
141
|
-
if (!modifiedCount) return null;
|
|
142
|
-
logger.info('[WorkflowRunner] Re-arming abandoned work', { kind, dedupeKey });
|
|
143
|
-
return this._arm(workflow, work._id, 0);
|
|
144
|
-
}
|
|
145
|
-
if (work.status !== 'pending') {
|
|
146
|
-
logger.info('[WorkflowRunner] Skipping, work is in flight or complete', { kind, dedupeKey, status: work.status });
|
|
147
|
-
return null;
|
|
148
|
-
}
|
|
149
|
-
return this._arm(workflow, work._id, this._delayUntilDue(work));
|
|
150
|
-
}
|
|
151
|
-
|
|
152
|
-
_delayUntilDue(work) {
|
|
153
|
-
return Math.max(0, new Date(work.nextAttemptAt).getTime() - Date.now());
|
|
154
|
-
}
|
|
155
|
-
|
|
156
|
-
_arm(workflow, workId, delayMs) {
|
|
157
|
-
const id = String(workId);
|
|
158
|
-
clearTimeout(this.timers.get(id));
|
|
159
|
-
const timer = setTimeout(() => {
|
|
160
|
-
this.timers.delete(id);
|
|
161
|
-
this._runDeferred(workflow, id).catch((error) => {
|
|
162
|
-
logger.error('[WorkflowRunner] Deferred run failed', { kind: workflow.kind, deferredWorkId: id, error: error.message });
|
|
163
|
-
});
|
|
164
|
-
}, delayMs);
|
|
165
|
-
timer.unref();
|
|
166
|
-
this.timers.set(id, timer);
|
|
167
|
-
return id;
|
|
168
|
-
}
|
|
169
|
-
|
|
170
|
-
async _runDeferred(workflow, deferredWorkId) {
|
|
171
|
-
const claimedAt = new Date();
|
|
172
|
-
const prior = await DeferredWork.findOneAndUpdate(
|
|
173
|
-
{
|
|
174
|
-
_id: deferredWorkId,
|
|
175
|
-
$or: [
|
|
176
|
-
{ status: 'pending', nextAttemptAt: { $lte: claimedAt } },
|
|
177
|
-
{ status: 'processing', claimedAt: { $lt: new Date(claimedAt.getTime() - STALE_CLAIM_MS) } }
|
|
178
|
-
]
|
|
179
|
-
},
|
|
180
|
-
{ $set: { status: 'processing', claimedAt } },
|
|
181
|
-
{ new: false }
|
|
182
|
-
);
|
|
183
|
-
if (!prior) {
|
|
184
|
-
logger.info('[WorkflowRunner] Claimed elsewhere or not due', { kind: workflow.kind, deferredWorkId });
|
|
185
|
-
return { claimed: false };
|
|
186
|
-
}
|
|
187
|
-
|
|
188
|
-
const reclaimed = prior.status === 'processing';
|
|
189
|
-
const work = { ...prior.toObject(), attempt: prior.attempt + (reclaimed ? 1 : 0) };
|
|
190
|
-
if (reclaimed) await this._write(work, claimedAt, { attempt: work.attempt });
|
|
191
|
-
|
|
192
|
-
const checkpoints = { ...(work.checkpoints || {}) };
|
|
193
|
-
const save = async (patch) => {
|
|
194
|
-
Object.assign(checkpoints, patch);
|
|
195
|
-
const $set = {};
|
|
196
|
-
for (const [key, value] of Object.entries(patch || {})) $set[`checkpoints.${key}`] = value;
|
|
197
|
-
if (Object.keys($set).length) await this._write(work, claimedAt, $set);
|
|
198
|
-
};
|
|
199
|
-
|
|
200
|
-
try {
|
|
201
|
-
await workflow.prepare(work.trigger, checkpoints, save);
|
|
202
|
-
} catch (error) {
|
|
203
|
-
return await this._rescheduleOrAbandon(workflow, work, claimedAt, error);
|
|
204
|
-
}
|
|
205
|
-
await this._write(work, claimedAt, { status: 'done', claimedAt: null });
|
|
206
|
-
return { status: 'done' };
|
|
207
|
-
}
|
|
208
|
-
|
|
209
|
-
async _write(work, claimedAt, $set) {
|
|
210
|
-
const { modifiedCount } = await DeferredWork.updateOne({ _id: work._id, claimedAt }, { $set });
|
|
211
|
-
if (!modifiedCount) logger.warn('[WorkflowRunner] Claim lost, write discarded', { deferredWorkId: String(work._id) });
|
|
212
|
-
return modifiedCount > 0;
|
|
213
|
-
}
|
|
214
|
-
|
|
215
|
-
async _rescheduleOrAbandon(workflow, work, claimedAt, error) {
|
|
216
|
-
const delayMinutes = workflow.retrySchedule[work.attempt];
|
|
217
|
-
if (error.permanent === true || delayMinutes === undefined) {
|
|
218
|
-
await this._write(work, claimedAt, { status: 'abandoned', lastError: error.message, claimedAt: null });
|
|
219
|
-
logger.error('[WorkflowRunner] Abandoned', {
|
|
220
|
-
kind: workflow.kind,
|
|
221
|
-
dedupeKey: work.dedupeKey,
|
|
222
|
-
attempt: work.attempt,
|
|
223
|
-
permanent: error.permanent === true,
|
|
224
|
-
error: error.message
|
|
225
|
-
});
|
|
226
|
-
return { status: 'abandoned' };
|
|
227
|
-
}
|
|
228
|
-
|
|
229
|
-
const attempt = work.attempt + 1;
|
|
230
|
-
const delayMs = delayMinutes * MINUTE_MS;
|
|
231
|
-
const kept = await this._write(work, claimedAt, {
|
|
232
|
-
status: 'pending',
|
|
233
|
-
attempt,
|
|
234
|
-
nextAttemptAt: new Date(Date.now() + delayMs),
|
|
235
|
-
lastError: error.message,
|
|
236
|
-
claimedAt: null
|
|
237
|
-
});
|
|
238
|
-
if (!kept) return { claimed: false };
|
|
239
|
-
|
|
240
|
-
logger.warn('[WorkflowRunner] Rescheduled', {
|
|
241
|
-
kind: workflow.kind,
|
|
242
|
-
dedupeKey: work.dedupeKey,
|
|
243
|
-
attempt,
|
|
244
|
-
delayMinutes,
|
|
245
|
-
error: error.message
|
|
246
|
-
});
|
|
247
|
-
this._arm(workflow, work._id, delayMs);
|
|
248
|
-
return { status: 'pending', attempt };
|
|
249
|
-
}
|
|
250
|
-
}
|
|
251
|
-
|
|
252
|
-
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 };
|