👷 Worker Types

Choosing the right worker type is critical for maximizing Asteri's performance for your specific application workload.

🏗️ Worker Models Overview

1. Synchronous Workers (sync)

The default worker type. It handles one request at a time per worker process. - Best for: CPU-bound tasks, simple applications, or legacy code that isn't thread-safe. - Mechanism: Standard process-based concurrency. If a request is blocked (e.g., waiting for a database), the entire worker is blocked.

2. Threaded Workers (gthread)

A multi-threaded worker model where each worker process can handle multiple requests using a thread pool. - Best for: I/O-bound tasks in applications that are thread-safe (e.g., standard Django/Flask apps). - Configuration: Use --threads to set the number of threads per worker. - Mechanism: Uses a thread pool to handle concurrent requests within a single process, reducing memory overhead compared to spawning many processes.

3. Gevent Workers (gevent)

An asynchronous worker based on Greenlets (pseudo-threads) and the Gevent library. - Best for: Extremely high concurrency, long-lived connections, and heavy I/O-bound workloads. - Requirement: Must have gevent installed (pip install gevent). - Mechanism: Cooperative multitasking. It patches standard library functions to be non-blocking, allowing thousands of concurrent connections in a single process.

4. ASGI Workers (asgi)

Native support for modern asynchronous applications and protocols (HTTP/2, WebSockets). - Best for: FastAPI, Starlette, and other async frameworks. - Mechanism: Implements the ASGI 3.0 specification. It leverages Python's asyncio to handle concurrent requests efficiently.

📊 Comparison Table

💡 Recommendation

• For Flask/Django: Start with gthread (2-4 workers, 2-4 threads).
• For FastAPI: Always use asgi.
• For High Concurrency: Experiment with gevent if your application is compatible.