Skip to content


Repository files navigation



A light-weight job scheduling library for Node.js

This was originally a fork of agenda.js, it differs from the original version in following points:

  • Complete rewrite in Typescript (fully typed!)
  • mongodb4 driver (supports mongodb 5.x)
  • Supports mongoDB sharding by name
  • touch() can have an optional progress parameter (0-100)
  • Bugfixes and improvements for locking & job processing (concurrency, lockLimit,..)
  • Breaking change: define() config paramter moved from 2nd position to 3rd
  • getRunningStats()
  • automatically waits for agenda to be connected before calling any database operations
  • uses a database abstraction layer behind the scene
  • does not create a database index by default, you can set ensureIndex: true when initializing Agenda or run manually:
    "name" : 1,
    "nextRunAt" : 1,
    "priority" : -1,
    "lockedAt" : 1,
    "disabled" : 1
}, "findAndLockNextJobIndex")

Agenda offers

  • Minimal overhead. Agenda aims to keep its code base small.
  • Mongo backed persistence layer.
  • Promises based API.
  • Scheduling with configurable priority, concurrency, repeating and persistence of job results.
  • Scheduling via cron or human readable syntax.
  • Event backed job queue that you can hook into.
  • Agenda-rest: optional standalone REST API.
  • Inversify-agenda - Some utilities for the development of agenda workers with Inversify.
  • Agendash: optional standalone web-interface.

Feature Comparison

Since there are a few job queue solutions, here a table comparing them to help you use the one that better suits your needs.

Feature Bull Bee Agenda
Backend redis redis mongo
Delayed jobs
Global events
Rate Limiter
Sandboxed worker
Repeatable jobs
Atomic ops ~
Central (Scalable) Queue
Supports long running jobs
Optimized for Jobs / Messages Messages Jobs

Kudos for making the comparison chart goes to Bull maintainers.


Install via NPM

npm install @hokify/agenda

You will also need a working Mongo database (v4+) to point it to.

Example Usage

const mongoConnectionString = 'mongodb://';

const agenda = new Agenda({ db: { address: mongoConnectionString } });

// Or override the default collection name:
// const agenda = new Agenda({db: {address: mongoConnectionString, collection: 'jobCollectionName'}});

// or pass additional connection options:
// const agenda = new Agenda({db: {address: mongoConnectionString, collection: 'jobCollectionName', options: {ssl: true}}});

// or pass in an existing mongodb-native MongoClient instance
// const agenda = new Agenda({mongo: myMongoClient});

agenda.define('delete old users', async job => {
	await User.remove({ lastLogIn: { $lt: twoDaysAgo } });

(async function () {
	// IIFE to give access to async/await
	await agenda.start();

	await agenda.every('3 minutes', 'delete old users');

	// Alternatively, you could also do:
	await agenda.every('*/3 * * * *', 'delete old users');
	'send email report',
	async job => {
		const { to } =;
		await emailClient.send({
			from: '',
			subject: 'Email Report',
			body: '...'
	{ priority: 'high', concurrency: 10 }

(async function () {
	await agenda.start();
	await agenda.schedule('in 20 minutes', 'send email report', { to: '' });
(async function () {
	const weeklyReport = agenda.create('send email report', { to: '' });
	await agenda.start();
	await weeklyReport.repeatEvery('1 week').save();

Full documentation

See also

Agenda's basic control structure is an instance of an agenda. Agenda's are mapped to a database collection and load the jobs from within.

Table of Contents

Configuring an agenda

All configuration methods are chainable, meaning you can do something like:

const agenda = new Agenda();
  .processEvery('3 minutes')

Possible agenda config options:

	name: string;
	defaultConcurrency: number;
	processEvery: number;
	maxConcurrency: number;
	defaultLockLimit: number;
	lockLimit: number;
	defaultLockLifetime: number;
	ensureIndex: boolean;
	sort: SortOptionObject<IJobParameters>;
	db: {
		collection: string;
		address: string;
		options: MongoClientOptions;
	mongo: Db;

Agenda uses Human Interval for specifying the intervals. It supports the following units:

seconds, minutes, hours, days,weeks, months -- assumes 30 days, years -- assumes 365 days

More sophisticated examples

agenda.processEvery('one minute');
agenda.processEvery('1.5 minutes');
agenda.processEvery('3 days and 4 hours');
agenda.processEvery('3 days, 4 hours and 36 seconds');

database(url, [collectionName], [MongoClientOptions])

Specifies the database at the url specified. If no collection name is given, agendaJobs is used.

By default useNewUrlParser and useUnifiedTopology is set to true,

agenda.database('localhost:27017/agenda-test', 'agendaJobs');

You can also specify it during instantiation.

const agenda = new Agenda({
	db: { address: 'localhost:27017/agenda-test', collection: 'agendaJobs' }

Agenda will emit a ready event (see Agenda Events) when properly connected to the database. It is safe to call agenda.start() without waiting for this event, as this is handled internally. If you're using the db options, or call database, then you may still need to listen for ready before saving jobs.

mongo(dbInstance, [collectionName])

Use an existing mongodb-native MongoClient/Db instance. This can help consolidate connections to a database. You can instead use .database to have agenda handle connecting for you.

You can also specify it during instantiation:

const agenda = new Agenda({ mongo: mongoClientInstance.db('agenda-test') });

Note that MongoClient.connect() returns a mongoClientInstance since node-mongodb-native 3.0.0, while it used to return a dbInstance that could then be directly passed to agenda.


Sets the lastModifiedBy field to name in the jobs collection. Useful if you have multiple job processors (agendas) and want to see which job queue last ran the job. + '-' +;

You can also specify it during instantiation

const agenda = new Agenda({ name: 'test queue' });


Takes a string interval which can be either a traditional javascript number, or a string such as 3 minutes

Specifies the frequency at which agenda will query the database looking for jobs that need to be processed. Agenda internally uses setTimeout to guarantee that jobs run at (close to ~3ms) the right time.

Decreasing the frequency will result in fewer database queries, but more jobs being stored in memory.

Also worth noting is that if the job queue is shutdown, any jobs stored in memory that haven't run will still be locked, meaning that you may have to wait for the lock to expire. By default it is '5 seconds'.

agenda.processEvery('1 minute');

You can also specify it during instantiation

const agenda = new Agenda({ processEvery: '30 seconds' });


Takes a number which specifies the max number of jobs that can be running at any given moment. By default it is 20.


You can also specify it during instantiation

const agenda = new Agenda({ maxConcurrency: 20 });


Takes a number which specifies the default number of a specific job that can be running at any given moment. By default it is 5.


You can also specify it during instantiation

const agenda = new Agenda({ defaultConcurrency: 5 });


Takes a number which specifies the max number jobs that can be locked at any given moment. By default it is 0 for no max.


You can also specify it during instantiation

const agenda = new Agenda({ lockLimit: 0 });


Takes a number which specifies the default number of a specific job that can be locked at any given moment. By default it is 0 for no max.


You can also specify it during instantiation

const agenda = new Agenda({ defaultLockLimit: 0 });


Takes a number which specifies the default lock lifetime in milliseconds. By default it is 10 minutes. This can be overridden by specifying the lockLifetime option to a defined job.

A job will unlock if it is finished (ie. the returned Promise resolves/rejects or done is specified in the params and done() is called) before the lockLifetime. The lock is useful if the job crashes or times out.


You can also specify it during instantiation

const agenda = new Agenda({ defaultLockLifetime: 10000 });


Takes a query which specifies the sort query to be used for finding and locking the next job.

By default it is { nextRunAt: 1, priority: -1 }, which obeys a first in first out approach, with respect to priority.


Optional. Disables the automatic creation of the default index on the jobs table. By default, Agenda creates an index to optimize its queries against Mongo while processing jobs.

This is useful if you want to use your own index in specific use-cases.

Agenda Events

An instance of an agenda will emit the following events:

  • ready - called when Agenda mongo connection is successfully opened and indices created. If you're passing agenda an existing connection, you shouldn't need to listen for this, as agenda.start() will not resolve until indices have been created. If you're using the db options, or call database, then you may still need to listen for the ready event before saving jobs. agenda.start() will still wait for the connection to be opened.
  • error - called when Agenda mongo connection process has thrown an error
await agenda.start();

Defining Job Processors

Before you can use a job, you must define its processing behavior.

define(jobName, fn, [options])

Defines a job with the name of jobName. When a job of jobName gets run, it will be passed to fn(job, done). To maintain asynchronous behavior, you may either provide a Promise-returning function in fn or provide done as a second parameter to fn. If done is specified in the function signature, you must call done() when you are processing the job. If your function is synchronous or returns a Promise, you may omit done from the signature.

options is an optional argument which can overwrite the defaults. It can take the following:

  • concurrency: number maximum number of that job that can be running at once (per instance of agenda)
  • lockLimit: number maximum number of that job that can be locked at once (per instance of agenda)
  • lockLifetime: number interval in ms of how long the job stays locked for (see multiple job processors for more info). A job will automatically unlock once a returned promise resolves/rejects (or if done is specified in the signature and done() is called).
  • priority: (lowest|low|normal|high|highest|number) specifies the priority of the job. Higher priority jobs will run first. See the priority mapping below
  • shouldSaveResult: boolean flag that specifies whether the result of the job should also be stored in the database. Defaults to false

Priority mapping:

  highest: 20,
  high: 10,
  normal: 0,
  low: -10,
  lowest: -20

Async Job:

agenda.define('some long running job', async job => {
	const data = await doSomelengthyTask();
	await formatThatData(data);
	await sendThatData(data);

Async Job (using done):

agenda.define('some long running job', (job, done) => {
	doSomelengthyTask(data => {

Sync Job:

agenda.define('say hello', job => {

define() acts like an assignment: if define(jobName, ...) is called multiple times (e.g. every time your script starts), the definition in the last call will overwrite the previous one. Thus, if you define the jobName only once in your code, it's safe for that call to execute multiple times.

Creating Jobs

every(interval, name, [data], [options])

Runs job name at the given interval. Optionally, data and options can be passed in. Every creates a job of type single, which means that it will only create one job in the database, even if that line is run multiple times. This lets you put it in a file that may get run multiple times, such as webserver.js which may reboot from time to time.

interval can be a human-readable format String, a cron format String, or a Number.

data is an optional argument that will be passed to the processing function under

options is an optional argument that will be passed to job.repeatEvery. In order to use this argument, data must also be specified.

Returns the job.

agenda.define('printAnalyticsReport', async job => {
	const users = await User.doSomethingReallyIntensive();
	console.log('I print a report!');

agenda.every('15 minutes', 'printAnalyticsReport');

Optionally, name could be array of job names, which is convenient for scheduling different jobs for same interval.

agenda.every('15 minutes', ['printAnalyticsReport', 'sendNotifications', 'updateUserRecords']);

In this case, every returns array of jobs.

schedule(when, name, [data])

Schedules a job to run name once at a given time. when can be a Date or a String such as tomorrow at 5pm.

data is an optional argument that will be passed to the processing function under

Returns the job.

agenda.schedule('tomorrow at noon', 'printAnalyticsReport', { userCount: 100 });

Optionally, name could be array of job names, similar to the every method.

agenda.schedule('tomorrow at noon', [

In this case, schedule returns array of jobs.

now(name, [data])

Schedules a job to run name once immediately.

data is an optional argument that will be passed to the processing function under

Returns the job.'do the hokey pokey');

create(jobName, data)

Returns an instance of a jobName with data. This does NOT save the job in the database. See below to learn how to manually work with jobs.

const job = agenda.create('printAnalyticsReport', { userCount: 100 });
console.log('Job successfully saved');

Managing Jobs

jobs(mongodb-native query, mongodb-native sort, mongodb-native limit, mongodb-native skip)

Lets you query (then sort, limit and skip the result) all of the jobs in the agenda job's database. These are full mongodb-native find, sort, limit and skip commands. See mongodb-native's documentation for details.

const jobs = await{ name: 'printAnalyticsReport' }, { data: -1 }, 3, 1);
// Work with jobs (see below)

cancel(mongodb-native query)

Cancels any jobs matching the passed mongodb-native query, and removes them from the database. Returns a Promise resolving to the number of cancelled jobs, or rejecting on error.

const numRemoved = await agenda.cancel({ name: 'printAnalyticsReport' });

This functionality can also be achieved by first retrieving all the jobs from the database using, looping through the resulting array and calling job.remove() on each. It is however preferable to use agenda.cancel() for this use case, as this ensures the operation is atomic.

disable(mongodb-native query)

Disables any jobs matching the passed mongodb-native query, preventing any matching jobs from being run by the Job Processor.

const numDisabled = await agenda.disable({ name: 'pollExternalService' });

Similar to agenda.cancel(), this functionality can be acheived with a combination of and job.disable()

enable(mongodb-native query)

Enables any jobs matching the passed mongodb-native query, allowing any matching jobs to be run by the Job Processor.

const numEnabled = await agenda.enable({ name: 'pollExternalService' });

Similar to agenda.cancel(), this functionality can be acheived with a combination of and job.enable()


Removes all jobs in the database without defined behaviors. Useful if you change a definition name and want to remove old jobs. Returns a Promise resolving to the number of removed jobs, or rejecting on error.

IMPORTANT: Do not run this before you finish defining all of your jobs. If you do, you will nuke your database of jobs.

const numRemoved = await agenda.purge();

Starting the job processor

To get agenda to start processing jobs from the database you must start it. This will schedule an interval (based on processEvery) to check for new jobs and run them. You can also stop the queue.


Starts the job queue processing, checking processEvery time to see if there are new jobs. Must be called after processEvery, and before any job scheduling (e.g. every).


Stops the job queue processing. Unlocks currently running jobs.

This can be very useful for graceful shutdowns so that currently running/grabbed jobs are abandoned so that other job queues can grab them / they are unlocked should the job queue start again. Here is an example of how to do a graceful shutdown.

async function graceful() {
	await agenda.stop();

process.on('SIGTERM', graceful);
process.on('SIGINT', graceful);

Multiple job processors

Sometimes you may want to have multiple node instances / machines process from the same queue. Agenda supports a locking mechanism to ensure that multiple queues don't process the same job.

You can configure the locking mechanism by specifying lockLifetime as an interval when defining the job.

	(job, cb) => {
		// Do something in 10 seconds or less...
	{ lockLifetime: 10000 }

This will ensure that no other job processor (this one included) attempts to run the job again for the next 10 seconds. If you have a particularly long running job, you will want to specify a longer lockLifetime.

By default it is 10 minutes. Typically you shouldn't have a job that runs for 10 minutes, so this is really insurance should the job queue crash before the job is unlocked.

When a job is finished (i.e. the returned promise resolves/rejects or done is specified in the signature and done() is called), it will automatically unlock.

Manually working with a job

A job instance has many instance methods. All mutating methods must be followed with a call to await in order to persist the changes to the database.

repeatEvery(interval, [options])

Specifies an interval on which the job should repeat. The job runs at the time of defining as well in configured intervals, that is "run now and in intervals".

interval can be a human-readable format String, a cron format String, or a Number.

options is an optional argument containing:

options.timezone: should be a string as accepted by moment-timezone and is considered when using an interval in the cron string format.

options.skipImmediate: true | false (default) Setting this true will skip the immediate run. The first run will occur only in configured interval.

options.startDate: Date the first time the job runs, should be equal or after the start date.

options.endDate: Date the job should not repeat after the endDate. The job can run on the end-date itself, but not after that.

options.skipDays: human readable string ('2 days'). After each run, it will skip the duration of 'skipDays'

job.repeatEvery('10 minutes');
job.repeatEvery('3 minutes', {
	skipImmediate: true
job.repeatEvery('0 6 * * *', {
	timezone: 'America/New_York'


Specifies a time when the job should repeat. Possible values



Specifies the next time at which the job should run.

job.schedule('tomorrow at 6pm');


Specifies the priority weighting of the job. Can be a number or a string from the above priority table.



Specifies whether the result of the job should also be stored in the database. Defaults to false.


The data returned by the job will be available on the result attribute after it succeeded and got retrieved again from the database, e.g. via or through the success job event).

unique(properties, [options])

Ensure that only one instance of this job exists with the specified properties

options is an optional argument which can overwrite the defaults. It can take the following:

  • insertOnly: boolean will prevent any properties from persisting if the job already exists. Defaults to false.
job.unique({ 'data.type': 'active', 'data.userId': '123', nextRunAt: date });

IMPORTANT: To avoid high CPU usage by MongoDB, make sure to create an index on the used fields, like data.type and data.userId for the example above.


Sets job.attrs.failedAt to now, and sets job.attrs.failReason to reason.

Optionally, reason can be an error, in which case job.attrs.failReason will be set to error.message'insufficient disk space');
// or Error('insufficient disk space'));


Runs the given job and calls callback(err, job) upon completion. Normally you never need to call this manually., job) => {
	console.log("I don't know why you would need to do this...");


Saves the job.attrs into the database. Returns a Promise resolving to a Job instance, or rejecting on error.

try {
	cosole.log('Successfully saved job to collection');
} catch (e) {
	console.error('Error saving job to collection');


Removes the job from the database. Returns a Promise resolving to the number of jobs removed, or rejecting on error.

try {
	await job.remove();
	console.log('Successfully removed job from collection');
} catch (e) {
	console.error('Error removing job from collection');


Disables the job. Upcoming runs won't execute.


Enables the job if it got disabled before. Upcoming runs will execute.


Resets the lock on the job. Useful to indicate that the job hasn't timed out when you have very long running jobs. The call returns a promise that resolves when the job's lock has been renewed.

agenda.define('super long job', async job => {
	await doSomeLongTask();
	await job.touch();
	await doAnotherLongTask();
	await job.touch();
	await finishOurLongTasks();

Job Queue Events

An instance of an agenda will emit the following events:

  • start - called just before a job starts
  • start:job name - called just before the specified job starts
agenda.on('start', job => {
	console.log('Job %s starting',;
  • complete - called when a job finishes, regardless of if it succeeds or fails
  • complete:job name - called when a job finishes, regardless of if it succeeds or fails
agenda.on('complete', job => {
	console.log(`Job ${} finished`);
  • success - called when a job finishes successfully
  • success:job name - called when a job finishes successfully
agenda.on('success:send email', job => {
	console.log(`Sent Email Successfully to ${}`);
  • fail - called when a job throws an error
  • fail:job name - called when a job throws an error
agenda.on('fail:send email', (err, job) => {
	console.log(`Job failed with error: ${err.message}`);

Frequently Asked Questions

What is the order in which jobs run?

Jobs are run with priority in a first in first out order (so they will be run in the order they were scheduled AND with respect to highest priority).

For example, if we have two jobs named "send-email" queued (both with the same priority), and the first job is queued at 3:00 PM and second job is queued at 3:05 PM with the same priority value, then the first job will run first if we start to send "send-email" jobs at 3:10 PM. However if the first job has a priority of 5 and the second job has a priority of 10, then the second will run first (priority takes precedence) at 3:10 PM.

The default MongoDB sort object is { nextRunAt: 1, priority: -1 } and can be changed through the option sort when configuring Agenda.

What is the difference between lockLimit and maxConcurrency?

Agenda will lock jobs 1 by one, setting the lockedAt property in mongoDB, and creating an instance of the Job class which it caches into the _lockedJobs array. This defaults to having no limit, but can be managed using lockLimit. If all jobs will need to be run before agenda's next interval (set via agenda.processEvery), then agenda will attempt to lock all jobs.

Agenda will also pull jobs from _lockedJobs and into _runningJobs. These jobs are actively being worked on by user code, and this is limited by maxConcurrency (defaults to 20).

If you have multiple instances of agenda processing the same job definition with a fast repeat time you may find they get unevenly loaded. This is because they will compete to lock as many jobs as possible, even if they don't have enough concurrency to process them. This can be resolved by tweaking the maxConcurrency and lockLimit properties.

Sample Project Structure?

Agenda doesn't have a preferred project structure and leaves it to the user to choose how they would like to use it. That being said, you can check out the example project structure below.

Can I Donate?

Thanks! I'm flattered, but it's really not necessary. If you really want to, you can find my gittip here.

Web Interface?

Agenda itself does not have a web interface built in but we do offer stand-alone web interface Agendash:

Agendash interface

Mongo vs Redis

The decision to use Mongo instead of Redis is intentional. Redis is often used for non-essential data (such as sessions) and without configuration doesn't guarantee the same level of persistence as Mongo (should the server need to be restarted/crash).

Agenda decides to focus on persistence without requiring special configuration of Redis (thereby degrading the performance of the Redis server on non-critical data, such as sessions).

Ultimately if enough people want a Redis driver instead of Mongo, I will write one. (Please open an issue requesting it). For now, Agenda decided to focus on guaranteed persistence.

Spawning / forking processes

Ultimately Agenda can work from a single job queue across multiple machines, node processes, or forks. If you are interested in having more than one worker, Bars3s has written up a fantastic example of how one might do it:

const cluster = require('cluster');
const os = require('os');

const httpServer = require('./app/http-server');
const jobWorker = require('./app/job-worker');

const jobWorkers = [];
const webWorkers = [];

if (cluster.isMaster) {
	const cpuCount = os.cpus().length;
	// Create a worker for each CPU
	for (let i = 0; i < cpuCount; i += 1) {

	cluster.on('exit', (worker, code, signal) => {
		if (jobWorkers.indexOf( !== -1) {
				`job worker ${} exited (signal: ${signal}). Trying to respawn...`

		if (webWorkers.indexOf( !== -1) {
				`http worker ${} exited (signal: ${signal}). Trying to respawn...`
} else {
	if (process.env.web) {
		console.log(`start http server: ${}`);
		// Initialize the http server here

	if (process.env.job) {
		console.log(`start job server: ${}`);
		// Initialize the Agenda here

function addWebWorker() {
	webWorkers.push(cluster.fork({ web: 1 }).id);

function addJobWorker() {
	jobWorkers.push(cluster.fork({ job: 1 }).id);

function removeWebWorker(id) {
	webWorkers.splice(webWorkers.indexOf(id), 1);

function removeJobWorker(id) {
	jobWorkers.splice(jobWorkers.indexOf(id), 1);

Recovering lost Mongo connections ("auto_reconnect")

Agenda is configured by default to automatically reconnect indefinitely, emitting an error event when no connection is available on each process tick, allowing you to restore the Mongo instance without having to restart the application.

However, if you are using an existing Mongo client you'll need to configure the reconnectTries and reconnectInterval connection settings manually, otherwise you'll find that Agenda will throw an error with the message "MongoDB connection is not recoverable, application restart required" if the connection cannot be recovered within 30 seconds.

Example Project Structure

Agenda will only process jobs that it has definitions for. This allows you to selectively choose which jobs a given agenda will process.

Consider the following project structure, which allows us to share models with the rest of our code base, and specify which jobs a worker processes, if any at all.

- server.js
- worker.js
  - agenda.js
    - user-controller.js
    - email.js
    - video-processing.js
    - image-processing.js
     - user-model.js
     - blog-post.model.js

Sample job processor (eg. jobs/email.js)

let email = require('some-email-lib'),
	User = require('../models/user-model.js');

module.exports = function (agenda) {
	agenda.define('registration email', async job => {
		const user = await User.get(;
		await email(, 'Thanks for registering', 'Thanks for registering ' +;

	agenda.define('reset password', async job => {
		// Etc

	// More email related jobs


const Agenda = require('agenda');

const connectionOpts = { db: { address: 'localhost:27017/agenda-test', collection: 'agendaJobs' } };

const agenda = new Agenda(connectionOpts);

const jobTypes = process.env.JOB_TYPES ? process.env.JOB_TYPES.split(',') : [];

jobTypes.forEach(type => {
	require('./jobs/' + type)(agenda);

if (jobTypes.length) {
	agenda.start(); // Returns a promise, which should be handled appropriately

module.exports = agenda;


let app = express(),
	User = require('../models/user-model'),
	agenda = require('../worker.js');'/users', (req, res, next) => {
	const user = new User(req.body); => {
		if (err) {
			return next(err);
		}'registration email', { userId: user.primary() });
		res.send(201, user.toJson());



Now you can do the following in your project:

node server.js

Fire up an instance with no JOB_TYPES, giving you the ability to process jobs, but not wasting resources processing jobs.

JOB_TYPES=email node server.js

Allow your http server to process email jobs.

JOB_TYPES=email node worker.js

Fire up an instance that processes email jobs.

JOB_TYPES=video-processing,image-processing node worker.js

Fire up an instance that processes video-processing/image-processing jobs. Good for a heavy hitting server.

Debugging Issues

If you think you have encountered a bug, please feel free to report it here:

Submit Issue

Please provide us with as much details as possible such as:

  • Agenda version
  • Environment (OSX, Linux, Windows, etc)
  • Small description of what happened
  • Any relevant stack track
  • Agenda logs (see below)

To turn on logging, please set your DEBUG env variable like so:

  • OSX: DEBUG="agenda:*" ts-node src/index.ts
  • Linux: DEBUG="agenda:*" ts-node src/index.ts
  • Windows CMD: set DEBUG=agenda:*
  • Windows PowerShell: $env:DEBUG = "agenda:*"

While not necessary, attaching a text file with this debug information would be extremely useful in debugging certain issues and is encouraged.

Known Issues

"Multiple order-by items are not supported. Please specify a single order-by item."

When running Agenda on Azure cosmosDB, you might run into this issue caused by Agenda's sort query used for finding and locking the next job. To fix this, you can pass custom sort option: sort: { nextRunAt: 1 }


It is recommended to set this index if you use agendash:

    "nextRunAt" : -1,
    "lastRunAt" : -1,
    "lastFinishedAt" : -1
}, "agendash2")

If you have one job definition with thousand of instances, you can add this index to improve internal sorting query for faster sortings

    "name" : 1,
    "disabled" : 1,
    "lockedAt" : 1
}, "findAndLockDeadJobs")

Sandboxed Worker - use child processes

It's possible to start jobs in a child process, this helps for example for long running processes to seperate them from the main thread. For example if one process consumes too much memory and gets killed, it will not affect any others. To use this feature, several steps are required. 1.) create a childWorker helper. The subrocess has a complete seperate context, so there are no database connections or anything else that can be shared. Therefore you have to ensure that all required connections and initializations are done here too. Furthermore you also have to load the correct job definition so that agenda nows what code it must execute. Therefore 3 parameters are passed to the childWorker: name, jobId and path to the job definition.

Example file can look like this:


import 'reflect-metadata';

process.on('message', message => {
  if (message === 'cancel') {
  } else {
    console.log('got message', message);

(async () => {
	const mongooseConnection = /** connect to database */

  /** do other required initializations */

  // get process arguments (name, jobId and path to agenda definition file)
	const [, , name, jobId, agendaDefinition] = process.argv;

  // set fancy process title
	process.title = `${process.title} (sub worker: ${name}/${jobId})`;

  // initialize Agenda in "forkedWorker" mode
	const agenda = new Agenda({ name: `subworker-${name}`, forkedWorker: true });
	// connect agenda (but do not start it)
	await agenda.mongo(mongooseConnection.db as any);

	if (!name || !jobId) {
		throw new Error(`invalid parameters: ${JSON.stringify(process.argv)}`);

  // load job definition
  /** in this case the file is for example ../some/path/definitions.js
  with a content like:
  export default (agenda: Agenda, definitionOnly = false) => {
      'some job',
      async (notification: {
        attrs: { data: { dealId: string; orderId: TypeObjectId<IOrder> } };
      }) => {
        // do something

    if (!definitionOnly) {
        // here you can create scheduled jobs or other things
	if (agendaDefinition) {
		const loadDefinition = await import(agendaDefinition);
		(loadDefinition.default || loadDefinition)(agenda, true);

  // run this job now
	await agenda.runForkedJob(jobId);

  // disconnect database and exit
})().catch(err => {
	console.error('err', err);
	if (process.send) {

Ensure to only define job definitions during this step, otherwise you create some overhead (e.g. if you create new jobs inside the defintion files). That's why I call the defintion file with agenda and a second paramter that is set to true. If this parameter is true, I do not initialize any jobs (create jobs etc..)

2.) to use this, you have to enable it on a job. Set forkMode to true:

const job = agenda.create('some job', { meep: 1 });



The MIT License