Move heavy work off the request path

CPU/memory-heavy or slow work (document conversion, OCR, large bulk extraction) shouldn't run inside the request that holds a user's connection. As load grows, move it to background workers fed by a queue:

• The request enqueues a job and returns quickly (with a processing status; Chapter 9).
• Workers pull jobs, do the heavy work, and update status; the client polls or subscribes for completion.
• Workers scale independently of the web tier, so a burst of uploads doesn't starve interactive chat.

As emphasized in Chapter 10, design for this from the start with status fields even if you begin synchronous; the migration to workers is then localized.