Guard hi-res PDF rendering against oversized pages

[CyMule]

Adds a preflight check for hi-res PDF rendering that estimates per-page rendered pixels before rasterization and rejects pages that exceed the configured safe limit.

---

Note

Medium Risk
Changes hi-res PDF partitioning to fail early with UnprocessableEntityError when a page would render beyond a new per-page pixel cap, which may reject documents that previously processed (or OOMed) depending on DPI/page size and env configuration.

Overview
Adds a hi-res PDF render preflight guard that estimates per-page rasterized pixel counts (from cropbox/mediabox at the configured DPI) and raises UnprocessableEntityError before any rendering when a page exceeds a configurable maximum.

Introduces new env config PDF_RENDER_MAX_PIXELS_PER_PAGE (default 1_000_000_000, 0 disables), wires the check into partition_pdf_or_image for non-image hi-res PDFs, and adds unit tests covering limit exceeded/allowed, disablement, and file-cursor restoration. Also bumps version to 0.22.24 and documents the fix in the changelog.

^{Reviewed by Cursor Bugbot for commit 37f65d2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 37f65d2. Configure here.}

[cursor]

Default pixel limit exceeds PIL safety limit

Medium Severity

The default PDF_RENDER_MAX_PIXELS_PER_PAGE of 1,000,000,000 (1 billion) is twice as high as PILImage.MAX_IMAGE_PIXELS which is set to 5e8 (500 million) at module level. This means with default settings, any page rendering between 500M and 1B pixels passes the preflight check but still triggers a PIL DecompressionBombError during rasterization — after the resource-intensive poppler rendering step has already allocated memory. The preflight check's purpose is to reject oversized pages before rasterization, but the default value makes it ineffective for the exact range PIL would catch.

Additional Locations (1)

• unstructured/partition/pdf.py#L111-L112

 

^{Reviewed by Cursor Bugbot for commit 37f65d2. Configure here.}