GPU Priority Queue Puts Search Embeddings Ahead of Crawler Batch Jobs

AI embeddings now run through a GPU priority queue. A user-facing search request always jumps ahead of background crawler batch jobs, so a heavy indexing run no longer slows down live search.

01 Two kinds of work, one GPU

Search embeddings and crawler batch embeddings compete for the same hardware, but they are not equally urgent. A crawler batch can wait a second; a person watching a search box cannot. The queue makes that difference explicit rather than leaving it to whichever job arrived first.

02 A timeout rather than a wait

The queue includes a configurable timeout. If the GPU is busy, search falls back to lexical results instantly rather than holding the request open. A slightly different ranking is a better outcome than a slow page.