What is Hatchet?
Hatchet is a modern distributed task queue system engineered to enhance web application resilience. It operates as a low-latency queue with an average start time of 25ms, combining real-time interaction capabilities with enterprise-grade reliability. The platform supports multiple queuing strategies including FIFO, LIFO, Round Robin, and Priority Queues.
Built with a focus on scalability and reliability, Hatchet offers comprehensive features such as customizable retry policies, built-in error handling, and full observability of task runs. The system supports multiple programming languages through declarative SDKs, including Python, Typescript, and Go, making it versatile for various development environments.
Features
- Low Latency Scheduling: 25ms average start time for real-time task execution
- Concurrency Management: Support for FIFO, LIFO, Round Robin, and Priority Queues
- Workflow Orchestration: Full DAG-style execution capabilities
- Error Handling: Customizable retry policies and built-in recovery mechanisms
- Observability: Searchable runs with log streaming and metrics tracking
- Multi-language Support: SDKs for Python, Typescript, and Go
- Cron Scheduling: Support for recurring task execution
- Spike Protection: Traffic smoothing and system load management
Use Cases
- Generative AI request distribution
- Large-scale document batch processing
- Multi-modal system workflow orchestration
- Event-based architecture management
- Background task processing
- Scheduled task execution
FAQs
-
What is Hatchet?
Hatchet is a managed low-latency queue for web apps to solve scaling issues like concurrency, fairness, and rate limiting. -
Does Hatchet Cloud run my compute?
Not yet, but soon. Currently Hatchet Cloud provides a managed control plane for workers. Users can run workers on any cloud provider or on-premises and connect them to Hatchet Cloud. -
Is there a self-hosted version of Hatchet?
Yes, Hatchet is an MIT-licensed Open Source project and instructions for self-hosting docker containers can be found in the documentation.
