Background Tasks
The AI Dashboard application executes long-running data generation pipelines using asynchronous thread operations rather than offloading tasks to external queuing clusters (such as Celery or RQ).
Architectural Context: Embedded Generator Loops
The pipeline generation process involves network calls to LLM APIs, dynamic schema synthesis, and database query executions. Combined, these stages can take 3 to 10 seconds to complete.
Rather than managing background worker nodes and polling endpoints, the application handles concurrency using LangGraph generator streams integrated directly into ASGI async handlers.
Execution Mechanisms
1. Server-Sent Events Async Streaming
When a client hits POST /dashboard/generate/stream, the router spawns a background generator processing the input prompt asynchronously.
As each agent stage finishes, the generator yields real-time progress events directly to the open network socket. This avoids blocking concurrent API requests while keeping the connection active.
2. Async Graph Orchestration
Within the generator thread, workflow stages execute using async invocations (graph.ainvoke / graph.astream). This allows the server runtime to switch execution contexts during database reads or API requests, maximizing CPU utilization.
Error Catching & Recovery
Because operations run within active async scopes, runtime errors are caught and handled directly:
- Graceful Termination: Unhandled exceptions trigger the generation of explicit type: error SSE payloads before safely closing the connection socket.
- Session Cleanup: If a connection drops unexpectedly during processing, incomplete state snapshots are safely handled by the database session manager.