zart

Durable multi-step workflows for Rust that survive process restarts.

Zart persists every step result to PostgreSQL. When a worker crashes and restarts, execution resumes from the last completed step — no work is repeated, no step runs twice. Concurrency is handled automatically via skip-locked polling.

At a glance

• Durable execution — steps are checkpointed; replay is transparent
• Retries — configurable per-step retry policies with backoff
• Wait groups — fan-out/fan-in across parallel sub-tasks
• Event-driven steps — pause execution until an external event arrives
• Timeouts — per-handler and per-step deadlines
• Observability — structured tracing and optional Prometheus metrics

Quick example

use zart::prelude::*;

#[zart_durable("onboard-user")]
async fn onboard(data: UserId) -> Result<(), MyError> {
    zart::require(SendWelcomeEmail { id: data }).await?;
    zart::require(ProvisionAccount { id: data }).await?;
    Ok(())
}

Each require call is a durable step. On the first run it executes and persists the result; on replay it returns the cached value instantly.

Learn more

• Website: https://www.zart.run/
• Repository: https://github.com/paulosuzart/zart
• Crates: zart-scheduler · zart-macros · zart-api · zart-cli