Skip to content

TypeScript SDK

The TypeScript SDK provides a full-featured client for building durable workflows with Absurd. It uses the pg (node-postgres) library for database access.

Installation

npm install absurd-sdk

Then import it:

import { Absurd } from 'absurd-sdk';

Creating a Client

import { Absurd } from 'absurd-sdk';

// From a connection string (or ABSURD_DATABASE_URL env var)
const app = new Absurd({
  db: 'postgresql://user:pass@localhost:5432/mydb',
  queueName: 'default',
});

// From an existing pg.Pool
import * as pg from 'pg';
const pool = new pg.Pool({ connectionString: '...' });
const app = new Absurd({ db: pool, queueName: 'default' });

// Minimal — uses ABSURD_DATABASE_URL and queue "default"
const app = new Absurd();

AbsurdOptions

Option Type Default Description
db pg.Pool | string ABSURD_DATABASE_URL or postgresql://localhost/absurd Database connection
queueName string "default" Default queue for operations
defaultMaxAttempts number 5 Default retry limit for spawned tasks
log Log console Logger (must have log, info, warn, error)
hooks AbsurdHooks — Lifecycle hooks for tracing / context propagation

Registering Tasks

app.registerTask({ name: 'send-email' }, async (params, ctx) => {
  const rendered = await ctx.step('render', async () => {
    return renderTemplate(params.template, params.data);
  });

  await ctx.step('send', async () => {
    return await mailer.send({ to: params.to, html: rendered });
  });
});

TaskRegistrationOptions

Option Type Default Description
name string required Task name (must match when spawning)
queue string Client queue Queue this task belongs to
defaultMaxAttempts number Client default Default max attempts
defaultCancellation CancellationPolicy — Default cancellation policy

Spawning Tasks

const { taskID, runID, attempt, created } = await app.spawn(
  'send-email',
  { to: 'user@example.com', template: 'welcome' },
  {
    maxAttempts: 10,
    retryStrategy: { kind: 'exponential', baseSeconds: 2, factor: 2, maxSeconds: 300 },
    headers: { traceId: '...' },
    idempotencyKey: 'welcome:user-42',
  }
);

SpawnOptions

Option Type Description
maxAttempts number Max retry attempts
retryStrategy RetryStrategy Backoff configuration
headers JsonObject Metadata attached to the task
queue string Target queue (must match registration if registered)
cancellation CancellationPolicy Auto-cancellation policy
idempotencyKey string Dedup key (existing task returned if key matches)

SpawnResult

Field Type Description
taskID string Unique task identifier (UUIDv7)
runID string Current run identifier
attempt number Attempt number
created boolean false if an existing task was returned (idempotency)

Task Results

app.fetchTaskResult(taskID, options?)

Returns the current task result snapshot, or null if the task does not exist.

const snapshot = await app.fetchTaskResult(taskID);

app.awaitTaskResult(taskID, options?)

Polls until the task reaches a terminal state (completed, failed, cancelled). Throws TimeoutError if options.timeout is reached.

const final = await app.awaitTaskResult(taskID, { timeout: 30 });

Task Context (TaskContext)

The context object passed to every task handler.

ctx.taskID

The unique identifier of the current task.

ctx.headers

Read-only JSON object of headers attached to the task.

ctx.step(name, fn)

Run an idempotent step. The return value is cached in Postgres — on retries the cached value is returned without calling fn again.

const result = await ctx.step('fetch-data', async () => {
  return await fetchFromAPI();
});

The return value must be JSON-serializable.

ctx.beginStep(name) + ctx.completeStep(handle, value)

Advanced form of ctx.step() when you need to split step handling into two calls (for example integrating with agent/event loops that expose separate "before" and "after" hooks).

const handle = await ctx.beginStep<MyState>('agent-turn');

if (handle.done) {
  return handle.state; // cached checkpoint value
}

const state = await runTurn();
await ctx.completeStep(handle, state);

handle.checkpointName contains the concrete checkpoint key (name, name#2, ...) after Absurd's automatic step numbering.

ctx.sleepFor(stepName, seconds)

Suspend the task for a duration.

await ctx.sleepFor('cooldown', 3600); // sleep 1 hour

ctx.sleepUntil(stepName, date)

Suspend the task until an absolute time.

await ctx.sleepUntil('deadline', new Date('2025-12-31T00:00:00Z'));

ctx.awaitEvent(eventName, options?)

Suspend until a named event is emitted. Returns the event payload.

const payload = await ctx.awaitEvent('order.shipped', {
  stepName: 'wait-for-shipment',  // optional custom checkpoint name
  timeout: 86400,                  // optional timeout in seconds
});

Throws TimeoutError if the timeout expires before the event arrives.

ctx.awaitTaskResult(taskID, options?)

Durably wait for another task's terminal result from inside a running task. The wait itself is checkpointed as a step (default step name: $awaitTaskResult:<taskID>).

options.queue must point to a different queue than the current task context queue.

const child = await app.spawn('child-task', {}, { queue: 'child-workers' });
const childResult = await ctx.awaitTaskResult(child.taskID, {
  queue: 'child-workers',
  timeout: 60,
});

ctx.heartbeat(seconds?)

Extend the current run's lease. Useful in long-running steps that don't write checkpoints frequently.

await ctx.heartbeat(300); // extend lease by 5 minutes

ctx.emitEvent(eventName, payload?)

Emit an event on the current queue. First emit per name wins.

await ctx.emitEvent('order.completed', { orderId: '42' });

Events

Emit events from outside a task handler:

await app.emitEvent('order.shipped', { trackingNumber: 'XYZ' });

// Emit to a specific queue
await app.emitEvent('order.shipped', { trackingNumber: 'XYZ' }, 'orders');

Cancellation

await app.cancelTask(taskID);
await app.cancelTask(taskID, 'other-queue');

Running tasks detect cancellation at the next step(), heartbeat(), or awaitEvent() call.

Retrying Failed Tasks

const result = await app.retryTask(taskID, {
  maxAttempts: 5,       // increase attempt limit
  spawnNewTask: false,  // retry in-place (default) or spawn a new task
});

Queue Management

await app.createQueue('emails');
await app.dropQueue('emails');
const queues = await app.listQueues(); // ['default', 'emails']

Starting a Worker

const worker = await app.startWorker({
  concurrency: 4,        // parallel tasks (default: 1)
  claimTimeout: 120,     // lease duration in seconds (default: 120)
  batchSize: 4,          // tasks to claim per poll (default: concurrency)
  pollInterval: 0.25,    // seconds between idle polls (default: 0.25)
  workerId: 'web-1',     // identifier for tracking (default: hostname:pid)
  fatalOnLeaseTimeout: true, // exit process if task exceeds 2x lease (default: true)
  onError: (err) => console.error(err),
});

// Graceful shutdown
await worker.close();

Single-Batch Processing

For cron-style or serverless workloads, process a single batch without a long-lived worker:

await app.workBatch('worker-1', 120, 10);

Binding to a Connection

Use bindToConnection() to run Absurd operations on a specific database connection (e.g., inside a transaction):

const client = await pool.connect();
try {
  const bound = app.bindToConnection(client);
  await bound.spawn('my-task', { key: 'value' });
} finally {
  client.release();
}

Hooks

Hooks let you integrate with tracing, logging, and context propagation systems.

beforeSpawn

Called before every spawn(). Modify options to inject headers:

const app = new Absurd({
  hooks: {
    beforeSpawn: (taskName, params, options) => {
      const traceId = getCurrentTraceId();
      return {
        ...options,
        headers: { ...options.headers, traceId },
      };
    },
  },
});

wrapTaskExecution

Wraps task handler execution. Must call execute():

const app = new Absurd({
  hooks: {
    wrapTaskExecution: async (ctx, execute) => {
      const traceId = ctx.headers.traceId;
      return await runWithTraceContext(traceId, execute);
    },
  },
});

Error Types

Error Description
SuspendTask Internal — task suspended (sleep/event). Never visible to user code.
CancelledTask Internal — task was cancelled. Never visible to user code.
FailedTask Internal — run already failed (e.g., claim expired).
TimeoutError Thrown by awaitEvent() / awaitTaskResult() when the timeout expires.

Closing

await app.close(); // stops worker, closes pool if owned

Retry Strategies

Kind Description
"fixed" Wait baseSeconds between each retry
"exponential" Wait baseSeconds * factor^attempt, capped at maxSeconds
"none" No automatic retries 
{ kind: 'exponential', baseSeconds: 1, factor: 2, maxSeconds: 300 }

Cancellation Policies

Field Type Description
maxDuration number Cancel task after N seconds total lifetime
maxDelay number Cancel task if no checkpoint written for N seconds