Design an Asynchronous Text-to-Video Generation Service

System Design: Asynchronous Text-to-Video Generation Service

Design a service that accepts text prompts from users and generates short-form videos. The generation process is computationally expensive (minutes to hours per job) and must be handled asynchronously.

Functional Requirements

1. Submit Jobs: Users submit a prompt and receive a Job ID immediately.
2. Status Tracking: Users can poll or be notified via webhook regarding the status of their generation (Pending, Processing, Completed, Failed).
3. Retrieve Media: Once complete, users can fetch the final video URL.
4. Idempotency: Users should be able to retry submission without creating duplicate jobs for the same request.

Non-Functional Requirements

1. Scalability: The system should support high bursts of user submissions.
2. Reliability: No jobs should be lost due to worker crashes or transient infrastructure failures.
3. Fairness: Ensure that heavy users cannot monopolize the limited GPU worker pool.

Task

1. Propose a high-level architecture. Which messaging patterns and data stores would you use?
2. Define the lifecycle of a job, including state transitions and how you handle "long-running" tasks that exceed typical API timeouts.
3. Explain how you would manage worker failures. What happens if a worker dies halfway through a generation process?

Include a diagram description or API definition (request/response schemas) and explain the trade-offs of your chosen queueing strategy (e.g., SQS vs. Kafka/Pub-Sub).