josk

分类: 数据库与数据存储 | 上传者: veliovgroup | 下载: 0 | 版本: v1.0（最新）

用于水平扩展的 Node.js 和 Bun 应用程序的 JoSk 任务/CRON 调度程序 — 同步计划的工作，以便每个周期在集群中运行一次。当用户在多实例/集群/Kubernetes/PM2/Meteor.js 拓扑、`josk` NPM 包或 Meteor `ostrio:cron-jobs` Atmosphere 包中编写或查看重复作业、cron 样式任务、`setInterval`/`setTimeout` 工作或定期后台作业（电子邮件队列、短信发送、同步、轮询、清理）时触发。当用户提及 JoSk 的名称、询问“RedisAdapter”/“MongoAdapter”/“PostgresAdapter”、谈论一次/最多一次执行、僵尸任务恢复、调度程序租用、“zombieTime”/“execute”/“concurrency”，或者将 JoSk 与 Agenda、Bree、node-cron、Bull 或 BullMQ 进行比较时也会触发。涵盖完整的公共 API（“new JoSk”、“setInterval”、“setTimeout”、“setImmediate”、“clearInterval”、“clearTimeout”、“destroy”、“ping”、三个内置适配器和自定义适配器合约）。

更新日志: Source: GitHub https://github.com/veliovgroup/josk

目录结构

当前层级: tree/master/skills/josk/

📁 references/
- 📄 adapters.md 10.6 KB
- 📄 api.md 11.8 KB
- 📄 meteor.md 4.0 KB
- 📄 patterns.md 9.7 KB
- 📄 troubleshooting.md 9.4 KB
📄 SKILL.md 4.3 KB

SKILL.md

---
name: josk
description: "JoSk task/CRON scheduler for horizontally scaled Node.js and Bun apps — synchronizes scheduled work so each tick runs once across the cluster. Trigger when the user is writing or reviewing recurring jobs, cron-style tasks, `setInterval`/`setTimeout` work, or periodic background jobs (email queues, SMS sends, sync, polling, cleanup) in a multi-instance / cluster / Kubernetes / PM2 / Meteor.js topology, the `josk` NPM package, or the Meteor `ostrio:cron-jobs` Atmosphere package. Also trigger when the user mentions JoSk by name, asks about `RedisAdapter` / `MongoAdapter` / `PostgresAdapter`, talks about exactly-once / at-most-once execution, zombie-task recovery, scheduler leases, `zombieTime` / `execute` / `concurrency`, or compares JoSk to Agenda, Bree, node-cron, Bull, or BullMQ. Covers the full public API (`new JoSk`, `setInterval`, `setTimeout`, `setImmediate`, `clearInterval`, `clearTimeout`, `destroy`, `ping`, the three built-in adapters, and the custom-adapter contract)."
---

# JoSk — distributed task scheduler for Node.js & Bun

JoSk runs one copy of each scheduled task across a cluster. Schedule lives in Redis, MongoDB, or PostgreSQL; lease + atomic claim prevents two instances running the same tick.

Use when: multi-process/pod deployment, scaled Meteor/K8s/PM2, user mentions JoSk/adapters/zombie leases, or compares to agenda/bree/node-cron/bull.

## Source of truth

Read `references/` lazily — do not guess v4/v5/v6 semantics from memory:

- `references/api.md` — constructor options, methods, hooks, types
- `references/adapters.md` — adapter setup, cluster rules, custom-adapter contract
- `references/patterns.md` — handler styles, CRON, concurrency, shutdown
- `references/meteor.md` — `ostrio:cron-jobs` / Meteor imports
- `references/troubleshooting.md` — zombies, jitter, migrations, KeyDB/replica caveats

## Mental model

```
new JoSk({ adapter: new <Redis|Mongo|Postgres>Adapter({ ... }), onError, ... })
  .setInterval | setTimeout | setImmediate (handler, [delay], uid)
```

- **`adapter`** required. **`uid`** app-wide unique string per task.
- **Handler:** async/Promise (preferred), zero-arg sync, or `(ready) =>` callback — see `references/patterns.md`.
- **Timer id:** `set*` returns `Promise<string>`; pass string or that Promise to `clear*`.

## New integration checklist

1. Storage + connected client (JoSk does not manage connections)
2. `onError` hook (strongly recommended)
3. `jobs.destroy()` on shutdown — see `references/patterns.md`
4. Pick adapter — see below + `references/adapters.md`

## Pick the adapter

- **PostgreSQL** — multi-DC, clock skew; server-time locks, `SKIP LOCKED`. Default when strict single-claim matters.
- **Redis/KeyDB/Valkey** — single-region, high frequency. Reject active-active / multi-master.
- **MongoDB** — app already on Mongo (Meteor: `MongoInternals…mongo.db`). Official `mongodb` driver only.

## Scheduling method

| Method | Guarantee | Use when |
|---|---|---|
| `setInterval(fn, delay, uid)` | At-least-once per tick | Idempotent recurring work; handler safe if run twice |
| `setTimeout(fn, delay, uid)` | At-most-once | One-shot deferred; duplicate worse than miss; removed before handler |
| `setImmediate(fn, uid)` | At-most-once | One-shot fire-now; same guarantee as `setTimeout` with delay 0 |

Duplicate vs miss worse? Map to table. `zombieTime` (default 15 min): max handler runtime before re-claim on intervals; keep ≥ slowest handler + margin, not below 60s.

## Throughput tuning

- `execute: 'batch'` (default) — all due tasks per lease; `one` — one task per lease
- `concurrency: Infinity` (default) — parallel handlers; set integer to cap pool/API/CPU use

## Red flags

- No `onError` hook
- Reused `uid` across different tasks
- Default `zombieTime` with handlers >15 min
- `resetOnInit: true` in production cluster
- Replica reads / multi-writer Redis
- Intervals <~2s (storage + jitter overlap)
- MongoAdapter on CosmosDB/DocumentDB/Mongoose without warning
- KeyDB active-replication / multi-master

Full API, adapter ctors, handler recipes: `references/api.md`, `references/adapters.md`, `references/patterns.md`.

## Install

```sh
npm install josk redis   # or mongodb / pg
bun add josk
meteor add ostrio:cron-jobs   # Meteor; see references/meteor.md
```

Node ≥ 20.9.0 or Bun ≥ 1.1.0. Server-only — never client/browser.

登录后下载/点赞/收藏 ❤ 38 | ★ 0

josk

目录结构

SKILL.md

举报内容

提示