Docling: convert long PDFs in page ranges

The situation

Mail archives hold the occasional very long PDF: standards, manuals, books. Docling converts a document in one job and holds all of it in memory while it does.

What works today, and why it is not enough

A ~600-page PDF can't finish within the default 10-minute job limit and ends terminal as provider_timeout. The limit is part of the document profile, so raising it for a few files means re-indexing the whole archive. On a CPU container, the same files peak around 10 GB, enough to crash the server at a smaller memory cap.

Ask

Above a page threshold, submit the PDF as consecutive page_range jobs (Docling Serve accepts page_range) and publish the combined page units once every range succeeds. Each job stays within the time limit and bounded memory, and a failed range can retry on its own.