Umbral Tasks

Some work shouldn't happen inside the request — sending the welcome email, resizing the upload, calling the flaky third-party webhook. Umbral Tasks is the Celery-equivalent: annotate an async function with #[task], enqueue it from a handler, and a separate worker drains the queue durably. The queue is a DB table, so a crash doesn't lose jobs, and retries back off exponentially instead of hammering a service that's already down.

Install

cargo add umbral-tasks

Wire it up

use umbral::prelude::*;
use umbral_tasks::TasksPlugin;

let app = App::builder()
    .database("default", pool)
    .plugin(TasksPlugin::default())
    .build()?;

Run the worker beside your web process:

cargo run -- worker        # drains the queue
cargo run -- tasks-beat    # fires periodic / cron schedules

Target: offload the welcome email

Define the job once, enqueue it from anywhere, and let the worker deliver it.

use umbral_tasks::{enqueue, EnqueueOptions};

#[umbral::task]
async fn send_welcome(payload: WelcomePayload) -> Result<(), String> {
    email_user(payload.user_id).await.map_err(|e| e.to_string())
}

// From a signup handler:
register_send_welcome();  // once at startup
enqueue("send_welcome", &WelcomePayload { user_id }, EnqueueOptions::default()).await?;

What you get

• #[task] macro turns an async fn into an enqueueable job
• DB-backed queue that survives a process crash
• Retries with exponential backoff + per-task timeouts
• Priority queues and future eta / delay scheduling
• Periodic "beat" tasks (cron or fixed interval)
• Read-only queue browser in the admin, with "Retry selected"