@nicnocquee/dataqueue 1.33.0 → 1.35.0-beta.20260224075710

package/ai/skills/dataqueue-advanced/SKILL.md

@@ -0,0 +1,370 @@
+---
+name: dataqueue-advanced
+description: Advanced DataQueue patterns — step memoization, waits, tokens, cron, timeouts, tags, idempotency.
+---
+
+# DataQueue Advanced Patterns
+
+## Step Memoization with ctx.run()
+
+Wrap side-effectful work in `ctx.run(stepName, fn)`. Results are cached in the database — when the handler re-runs after a wait, completed steps replay from cache without re-executing.
+
+```typescript
+const handler = async (payload, signal, ctx) => {
+  const data = await ctx.run('fetch-data', async () => {
+    return await fetchFromAPI(payload.url);
+  });
+
+  await ctx.run('send-notification', async () => {
+    await notify(data.userId, data.message);
+  });
+};
+```
+
+**Rules:**
+
+- Step names must be unique within a handler.
+- Step names must be stable across deployments while jobs are waiting.
+- Step order must not change conditionally between re-invocations.
+
+## Time-Based Waits
+
+### waitFor (duration)
+
+```typescript
+const handler = async (payload, signal, ctx) => {
+  await ctx.run('step-1', async () => {
+    /* ... */
+  });
+  await ctx.waitFor({ hours: 24 });
+  await ctx.run('step-2', async () => {
+    /* ... */
+  });
+};
+```
+
+Duration fields: `seconds`, `minutes`, `hours`, `days`, `weeks`, `months`, `years` (additive).
+
+### waitUntil (date)
+
+```typescript
+await ctx.waitUntil(new Date('2025-03-01T09:00:00Z'));
+```
+
+### How waits work internally
+
+1. Handler throws a `WaitSignal` internally.
+2. Job moves to `'waiting'` status — worker lock is released.
+3. After the wait expires, job becomes `'pending'` again.
+4. Handler re-runs from top; `ctx.run()` replays cached steps.
+
+Waiting jobs are idle — they hold no lock, no concurrency slot, no resources.
+
+## Token-Based Waits (Human-in-the-Loop)
+
+Create a token, send it to an external actor, and wait for them to complete it.
+
+```typescript
+const handler = async (payload, signal, ctx) => {
+  const token = await ctx.run('create-token', async () => {
+    return await ctx.createToken({ timeout: '48h', tags: ['approval'] });
+  });
+
+  await ctx.run('notify', async () => {
+    await sendSlack(`Approve: ${token.id}`);
+  });
+
+  const result = await ctx.waitForToken<{ approved: boolean }>(token.id);
+  if (result.ok) {
+    await ctx.run('process', async () => {
+      if (result.output.approved) await approve(payload.id);
+    });
+  }
+};
+```
+
+Complete tokens externally:
+
+```typescript
+await queue.completeToken(tokenId, { approved: true });
+```
+
+Expire timed-out tokens periodically:
+
+```typescript
+await queue.expireTimedOutTokens();
+```
+
+## Cron Scheduling
+
+```typescript
+const cronId = await queue.addCronJob({
+  scheduleName: 'daily-report',
+  cronExpression: '0 9 * * *',
+  jobType: 'generate_report',
+  payload: { reportId: 'daily', userId: 'system' },
+  timezone: 'America/New_York',
+  allowOverlap: false,
+});
+```
+
+The processor automatically enqueues due cron jobs before each batch — no manual triggering needed.
+
+Manage schedules:
+
+```typescript
+await queue.pauseCronJob(cronId);
+await queue.resumeCronJob(cronId);
+await queue.editCronJob(cronId, { cronExpression: '0 */2 * * *' });
+await queue.removeCronJob(cronId);
+const schedules = await queue.listCronJobs('active');
+```
+
+## Timeout Management
+
+### Proactive — ctx.prolong()
+
+```typescript
+const handler = async (payload, signal, ctx) => {
+  ctx.prolong(60_000); // set deadline to 60s from now
+  await doHeavyWork();
+  ctx.prolong(); // reset to original timeoutMs
+};
+```
+
+### Reactive — ctx.onTimeout()
+
+```typescript
+const handler = async (payload, signal, ctx) => {
+  let step = 0;
+  ctx.onTimeout(() => {
+    if (step < 3) return 30_000; // extend 30s
+  });
+  step = 1;
+  await doStep1();
+  step = 2;
+  await doStep2();
+  step = 3;
+  await doStep3();
+};
+```
+
+Both update `locked_at` in the DB, preventing premature reclamation.
+
+### Force Kill on Timeout
+
+```typescript
+await queue.addJob({
+  jobType: 'task',
+  payload: {
+    /* ... */
+  },
+  timeoutMs: 5000,
+  forceKillOnTimeout: true,
+});
+```
+
+**Limitations of forceKillOnTimeout:**
+
+- Requires Node.js (not Bun).
+- Handler must be serializable (no closures over external variables).
+- `prolong`, `onTimeout`, `ctx.run`, waits are NOT available.
+
+## Event Hooks
+
+Subscribe to real-time job lifecycle events. Works identically with PostgreSQL and Redis.
+
+```typescript
+const queue = initJobQueue<MyPayloadMap>(config);
+
+queue.on('job:completed', ({ jobId, jobType }) => {
+  console.log(`Job ${jobId} (${jobType}) completed`);
+});
+
+queue.on('job:failed', ({ jobId, jobType, error, willRetry }) => {
+  console.error(`Job ${jobId} failed: ${error.message}`);
+  if (!willRetry) {
+    alertOps(`Permanent failure for job ${jobId}`);
+  }
+});
+
+queue.on('error', (error) => {
+  Sentry.captureException(error);
+});
+```
+
+### Available events
+
+| Event            | Payload                                |
+| ---------------- | -------------------------------------- |
+| `job:added`      | `{ jobId, jobType }`                   |
+| `job:processing` | `{ jobId, jobType }`                   |
+| `job:completed`  | `{ jobId, jobType }`                   |
+| `job:failed`     | `{ jobId, jobType, error, willRetry }` |
+| `job:cancelled`  | `{ jobId }`                            |
+| `job:retried`    | `{ jobId }`                            |
+| `job:waiting`    | `{ jobId, jobType }`                   |
+| `job:progress`   | `{ jobId, progress }`                  |
+| `error`          | `Error`                                |
+
+### Listener management
+
+```typescript
+const listener = ({ jobId }) => console.log(jobId);
+queue.on('job:completed', listener);
+queue.off('job:completed', listener);
+queue.once('job:added', ({ jobId }) => console.log('First job:', jobId));
+queue.removeAllListeners('job:completed');
+queue.removeAllListeners(); // all events
+```
+
+The `error` event fires alongside `onError` callbacks in `ProcessorOptions` and `SupervisorOptions` -- both mechanisms work independently.
+
+## Tags
+
+```typescript
+await queue.addJob({
+  jobType: 'email',
+  payload: {
+    /* ... */
+  },
+  tags: ['welcome', 'onboarding'],
+});
+
+const jobs = await queue.getJobsByTags(['welcome'], 'any');
+await queue.cancelAllUpcomingJobs({
+  tags: { values: ['onboarding'], mode: 'all' },
+});
+```
+
+Tag query modes: `'exact'`, `'all'`, `'any'`, `'none'`.
+
+## Idempotency
+
+```typescript
+const jobId = await queue.addJob({
+  jobType: 'email',
+  payload: { to: 'user@example.com', subject: 'Welcome', body: '...' },
+  idempotencyKey: `welcome-${userId}`,
+});
+```
+
+If a job with the same key exists, returns the existing job ID. Key is unique across all statuses until `cleanupOldJobs` removes it.
+
+## Transactional Job Creation (PostgreSQL Only)
+
+Insert a job within an existing database transaction so the job is enqueued **atomically** with other writes:
+
+```typescript
+import { Pool } from 'pg';
+
+const pool = new Pool({ connectionString: process.env.DATABASE_URL });
+
+async function registerUser(email: string, name: string) {
+  const client = await pool.connect();
+  try {
+    await client.query('BEGIN');
+
+    await client.query('INSERT INTO users (email, name) VALUES ($1, $2)', [
+      email,
+      name,
+    ]);
+
+    const queue = getJobQueue();
+    await queue.addJob(
+      {
+        jobType: 'send_email',
+        payload: { to: email, subject: 'Welcome!', body: `Hi ${name}!` },
+      },
+      { db: client },
+    );
+
+    await client.query('COMMIT');
+  } catch (error) {
+    await client.query('ROLLBACK');
+    throw error;
+  } finally {
+    client.release();
+  }
+}
+```
+
+The `db` option accepts any object matching `DatabaseClient { query(text, values): Promise<{ rows, rowCount }> }` — works with `pg.PoolClient`, `pg.Client`, or compatible ORM query runners.
+
+The job event (`'added'`) is also inserted within the same transaction.
+
+## Retry Strategy
+
+Configure how failed jobs are retried with `retryDelay`, `retryBackoff`, and `retryDelayMax`.
+
+### Fixed delay
+
+```typescript
+await queue.addJob({
+  jobType: 'email',
+  payload: {
+    /* ... */
+  },
+  maxAttempts: 5,
+  retryDelay: 30, // 30 seconds between each retry
+  retryBackoff: false,
+});
+```
+
+### Exponential backoff with cap
+
+```typescript
+await queue.addJob({
+  jobType: 'email',
+  payload: {
+    /* ... */
+  },
+  maxAttempts: 10,
+  retryDelay: 5, // base: 5 seconds
+  retryBackoff: true, // default — delay doubles each attempt with jitter
+  retryDelayMax: 300, // never wait more than 5 minutes
+});
+```
+
+### Cron schedules with retry config
+
+```typescript
+await queue.addCronJob({
+  scheduleName: 'daily-sync',
+  cronExpression: '0 * * * *',
+  jobType: 'sync',
+  payload: { source: 'api' },
+  retryDelay: 60,
+  retryBackoff: true,
+  retryDelayMax: 600,
+});
+```
+
+Every job enqueued by the schedule inherits the retry settings.
+
+### Default behavior
+
+When no retry options are set, the legacy formula `2^attempts * 60 seconds` is used. This is fully backward compatible.
+
+## Maintenance
+
+Use `createSupervisor()` to automate all maintenance tasks in a long-running server:
+
+```typescript
+const supervisor = queue.createSupervisor({
+  intervalMs: 60_000,
+  stuckJobsTimeoutMinutes: 10,
+  cleanupJobsDaysToKeep: 30,
+  cleanupEventsDaysToKeep: 30,
+});
+supervisor.startInBackground();
+```
+
+For serverless or one-off scripts, call `supervisor.start()` (runs once) or use the manual methods:
+
+```typescript
+await queue.reclaimStuckJobs(10); // reclaim jobs stuck > 10 min
+await queue.cleanupOldJobs(30); // delete completed jobs > 30 days
+await queue.cleanupOldJobEvents(30); // delete old events > 30 days
+await queue.expireTimedOutTokens(); // expire overdue tokens
+```

package/ai/skills/dataqueue-core/SKILL.md

@@ -0,0 +1,235 @@
+---
+name: dataqueue-core
+description: Core patterns for using @nicnocquee/dataqueue — typed PayloadMap, init, handlers, adding and processing jobs.
+---
+
+# DataQueue Core Patterns
+
+## Imports
+
+Always import from `@nicnocquee/dataqueue`. There is no v2/v3 subpath.
+
+```typescript
+import { initJobQueue, JobHandlers } from '@nicnocquee/dataqueue';
+```
+
+## Step 1: Define a PayloadMap
+
+Define an object type mapping job type strings to their payload shapes. This is the foundation of type safety — every API method is generic over this map.
+
+```typescript
+export type JobPayloadMap = {
+  send_email: { to: string; subject: string; body: string };
+  generate_report: { reportId: string; userId: string };
+};
+```
+
+## Step 2: Define Handlers
+
+Create a `JobHandlers<PayloadMap>` object. TypeScript enforces that every key in the PayloadMap has a handler. Each handler receives `(payload, signal, ctx)`.
+
+```typescript
+import { JobHandlers } from '@nicnocquee/dataqueue';
+import type { JobPayloadMap } from './types';
+
+export const jobHandlers: JobHandlers<JobPayloadMap> = {
+  send_email: async (payload) => {
+    await sendEmail(payload.to, payload.subject, payload.body);
+  },
+  generate_report: async (payload, signal) => {
+    if (signal.aborted) return;
+    const url = await generateReport(payload.reportId, payload.userId);
+    return { url }; // stored as job output, readable via getJob()
+  },
+};
+```
+
+## Step 3: Initialize the Queue (Singleton)
+
+Use a module-level singleton. Each `initJobQueue` call creates a new database pool — never call it per-request.
+
+### PostgreSQL
+
+```typescript
+import { initJobQueue } from '@nicnocquee/dataqueue';
+import type { JobPayloadMap } from './types';
+
+let jobQueue: ReturnType<typeof initJobQueue<JobPayloadMap>> | null = null;
+
+export const getJobQueue = () => {
+  if (!jobQueue) {
+    jobQueue = initJobQueue<JobPayloadMap>({
+      databaseConfig: {
+        connectionString: process.env.PG_DATAQUEUE_DATABASE,
+      },
+    });
+  }
+  return jobQueue;
+};
+```
+
+### Redis
+
+```typescript
+jobQueue = initJobQueue<JobPayloadMap>({
+  backend: 'redis',
+  redisConfig: {
+    url: process.env.REDIS_URL,
+    keyPrefix: 'myapp:',
+  },
+});
+```
+
+### Bring Your Own Pool / Client
+
+You can pass an existing `pg.Pool` or `ioredis` client instead of connection config:
+
+```typescript
+import { Pool } from 'pg';
+const pool = new Pool({ connectionString: process.env.DATABASE_URL });
+
+jobQueue = initJobQueue<JobPayloadMap>({ pool });
+```
+
+```typescript
+import IORedis from 'ioredis';
+const redis = new IORedis(process.env.REDIS_URL);
+
+jobQueue = initJobQueue<JobPayloadMap>({
+  backend: 'redis',
+  client: redis,
+  keyPrefix: 'myapp:',
+});
+```
+
+When you provide your own pool/client, the library will **not** close it on shutdown — you manage its lifecycle.
+
+## Step 4: Add Jobs
+
+```typescript
+const jobId = await queue.addJob({
+  jobType: 'send_email',
+  payload: { to: 'user@example.com', subject: 'Hi', body: 'Hello' },
+  priority: 10,
+  runAt: new Date(Date.now() + 5000),
+  tags: ['welcome'],
+  idempotencyKey: 'welcome-user-123',
+});
+```
+
+### Batch Insert
+
+Use `addJobs` to insert many jobs in a single database round-trip. Returns IDs in the same order as the input array.
+
+```typescript
+const jobIds = await queue.addJobs([
+  {
+    jobType: 'send_email',
+    payload: { to: 'a@example.com', subject: 'Hi', body: '...' },
+  },
+  {
+    jobType: 'send_email',
+    payload: { to: 'b@example.com', subject: 'Hi', body: '...' },
+    priority: 10,
+  },
+  {
+    jobType: 'generate_report',
+    payload: { reportId: '1', userId: '2' },
+    tags: ['monthly'],
+  },
+]);
+```
+
+Each job can independently have its own `idempotencyKey`, `priority`, `runAt`, `tags`, etc. The `{ db }` transactional option is also supported (PostgreSQL only).
+
+### Transactional Job Creation (PostgreSQL only)
+
+Pass an external `pg.PoolClient` inside a transaction via `{ db: client }`:
+
+```typescript
+const client = await pool.connect();
+await client.query('BEGIN');
+await client.query('INSERT INTO users (email) VALUES ($1)', [email]);
+await queue.addJob(
+  {
+    jobType: 'send_email',
+    payload: { to: email, subject: 'Welcome!', body: '...' },
+  },
+  { db: client },
+);
+await client.query('COMMIT');
+client.release();
+```
+
+If the transaction rolls back, the job is never enqueued.
+
+### Retry configuration
+
+Control retry behavior per-job with `retryDelay`, `retryBackoff`, and `retryDelayMax`:
+
+```typescript
+await queue.addJob({
+  jobType: 'send_email',
+  payload: { to: 'user@example.com', subject: 'Hi', body: 'Hello' },
+  maxAttempts: 5,
+  retryDelay: 10, // base delay: 10 seconds
+  retryBackoff: true, // exponential backoff (default)
+  retryDelayMax: 300, // cap at 5 minutes
+});
+```
+
+- **Fixed delay**: set `retryBackoff: false` for constant delay between retries.
+- **Exponential backoff** (default): delay doubles each attempt with jitter.
+- **Default**: when no retry options are set, legacy `2^attempts * 60s` is used.
+
+## Step 5: Process Jobs
+
+### Serverless (one-shot)
+
+```typescript
+const processor = queue.createProcessor(handlers, {
+  batchSize: 10,
+  concurrency: 3,
+});
+const processed = await processor.start();
+```
+
+### Long-running server
+
+```typescript
+const processor = queue.createProcessor(handlers, {
+  batchSize: 10,
+  concurrency: 3,
+  pollInterval: 5000,
+});
+processor.startInBackground();
+
+// Automate maintenance (reclaim stuck jobs, cleanup old data, expire tokens)
+const supervisor = queue.createSupervisor({
+  intervalMs: 60_000,
+  stuckJobsTimeoutMinutes: 10,
+  cleanupJobsDaysToKeep: 30,
+  cleanupEventsDaysToKeep: 30,
+});
+supervisor.startInBackground();
+
+process.on('SIGTERM', async () => {
+  await Promise.all([
+    processor.stopAndDrain(30000),
+    supervisor.stopAndDrain(30000),
+  ]);
+  queue.getPool().end();
+  process.exit(0);
+});
+```
+
+## Common Mistakes
+
+1. **Creating a new queue per request** — always use a singleton. Each `initJobQueue` creates a DB pool.
+2. **Missing handler for a job type** — the job fails with `FailureReason.NoHandler`. Let TypeScript enforce completeness by typing handlers as `JobHandlers<PayloadMap>`.
+3. **Not checking `signal.aborted`** — timed-out jobs keep running in the background. Always check the signal in long-running handlers.
+4. **Skipping maintenance** — use `createSupervisor()` to automate reclaiming stuck jobs, cleaning up old data, and expiring tokens. Without it, crashed workers leave jobs stuck in `processing` and tables grow unbounded.
+5. **Forgetting to run migrations** — PostgreSQL requires `dataqueue-cli migrate` before use. Redis needs no migrations.
+6. **Not calling `stopAndDrain` on shutdown** — use `stopAndDrain()` (not `stop()`) for graceful shutdown to avoid stuck jobs.
+7. **Forgetting to commit/rollback when using `db` option** — the `addJob` INSERT sits in an open transaction. If you never `COMMIT` or `ROLLBACK`, the connection leaks and the job is invisible to other sessions.
+8. **Using `db` option with Redis** — transactional job creation is PostgreSQL only. The Redis backend throws if `db` is provided.