admin/mcp-nextcloud
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: lower DOCUMENT_PDF_GRAPHICS_LIMIT default 5000 -> 1000

Browse Source 
The OOM hotfix (#852) set graphics_limit=5000, which caught the 955k-drawing
bomb but let a second pathology through: form/table PDFs (e.g. student records)
have ~1.5k grid-line vector drawings per page -- under 5000, so uncapped. With
those pages uncapped, pymupdf4llm's O(n^2) find_tables grinds ~17s/page, so a
7-page form hits the 120s timeout. All 6 current backfill parse failures in
tenant-blackbox-demo are this exact timeout (zero OOM, zero error).

Measured on a 7-page sample: graphics_limit=2000 -> 119s (timeout), 1000 -> 2.9s
-- with identical extracted text and ZERO recovered tables either way (the
expensive analysis produces nothing useful on these dense forms). Lowering the
default to 1000 makes them index in ~3s; the bomb file (955k >> 1000) stays
capped, and pages with genuine simple tables (<1000 drawings) still get table
detection.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
2026-06-05 00:06:40 +02:00
co-authored by Claude Opus 4.8
parent eb5f59f085
commit 77a36fa821
2 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nextcloud_mcp_server/config.py
+9 -4
View File
@@ -133,7 +133,7 @@ _DEFAULTS: dict[str, Any] = {
"document_chunk_size": 2048,
"document_chunk_overlap": 200,
# PDF parse isolation (OOM guard)
"document_pdf_graphics_limit": 5000,
"document_pdf_graphics_limit": 1000,
"document_parse_timeout_seconds": 120.0,
"document_parse_mem_limit_mb": 1536,
# Observability
@@ -712,9 +712,14 @@ class Settings:
# PDF parse isolation (OOM guard). The parse runs in a subprocess so one
# pathological file fails that doc, not the pod.
# to_markdown graphics cap; pages above it skip graphics analysis. Must be
# >=1 -- pymupdf4llm treats 0 as "no cap", which re-exposes the OOM.
document_pdf_graphics_limit: int = 5000
# to_markdown graphics cap: pages with more vector drawings than this skip
# the O(n^2) find_tables analysis. Must be >=1 -- pymupdf4llm treats 0 as
# "no cap", which re-exposes the OOM. Default 1000: form/table PDFs have
# ~1.5k grid-line drawings per page, which at the old 5000 cap slipped
# through uncapped and timed out after ~17s/page (for zero recovered
# tables); at 1000 they parse in ~3s with identical text. Pages with genuine
# simple tables (<1000 drawings) still get table detection.
document_pdf_graphics_limit: int = 1000
# wall-clock cap per parse; the worker subprocess is killed on timeout.
# float so a fractional DOCUMENT_PARSE_TIMEOUT_SECONDS is honoured, matching
# anyio.move_on_after's float seconds.