feat(pdf): #74 au complet — fix auth 500 stream + endpoint pdf/info + HTTP Range 206 + indexation incrementale PDF + config taille/timeout

backend/pdf_reader.py

@@ -2,10 +2,19 @@
 from __future__ import annotations
 
 import logging
+import os
+from concurrent.futures import ThreadPoolExecutor
+from concurrent.futures import TimeoutError as FuturesTimeout
 from pathlib import Path
 
 logger = logging.getLogger(__name__)
 
+# Configurable limits (see ROADMAP #74 G3):
+#   OBSIGATE_PDF_MAX_SIZE_MB   — PDFs larger than this are not text-extracted (default 50)
+#   OBSIGATE_PDF_EXTRACT_TIMEOUT — seconds before extraction is abandoned (default 30)
+PDF_MAX_SIZE_MB: int = int(os.environ.get("OBSIGATE_PDF_MAX_SIZE_MB", "50"))
+PDF_EXTRACT_TIMEOUT: float = float(os.environ.get("OBSIGATE_PDF_EXTRACT_TIMEOUT", "30"))
+
 PDF_READER: str = "pypdf"
 PdfReader = None  # type: ignore
 try:
@@ -20,15 +29,51 @@ except ImportError:
         logger.warning("No PDF reader available — install pypdf or pymupdf")
 
 
+def pdf_exceeds_size_limit(file_path: Path) -> bool:
+    """True if the PDF is larger than OBSIGATE_PDF_MAX_SIZE_MB (skip text extraction)."""
+    try:
+        return file_path.stat().st_size > PDF_MAX_SIZE_MB * 1024 * 1024
+    except OSError:
+        return False
+
+
+def _run_with_timeout(fn, *args, timeout: float):
+    """Run a sync function in a worker thread with a hard timeout.
+
+    A timed-out extraction leaves its worker thread running (daemon-style pool
+    is abandoned), but the request itself is freed — acceptable trade-off for
+    pathological PDFs on a self-hosted single-user server.
+    """
+    executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="pdf-extract")
+    try:
+        future = executor.submit(fn, *args)
+        return future.result(timeout=timeout)
+    finally:
+        executor.shutdown(wait=False)
+
+
 def extract_pdf_text(file_path: Path, max_chars: int = 100000) -> str:
-    """Extract text from a PDF file. Returns empty string on failure."""
+    """Extract text from a PDF file. Returns empty string on failure.
+
+    Oversized PDFs (> OBSIGATE_PDF_MAX_SIZE_MB) and extractions exceeding
+    OBSIGATE_PDF_EXTRACT_TIMEOUT seconds return "" instead of blocking.
+    """
     if PdfReader is None and PDF_READER == "pypdf":
         return ""
+    if pdf_exceeds_size_limit(file_path):
+        logger.info("PDF too large to index (> %d MB), skipping text extraction: %s",
+                    PDF_MAX_SIZE_MB, file_path)
+        return ""
     try:
         if PDF_READER == "pymupdf":
-            return _extract_pymupdf(file_path, max_chars)
+            return _run_with_timeout(_extract_pymupdf, file_path, max_chars,
+                                     timeout=PDF_EXTRACT_TIMEOUT)
         else:
-            return _extract_pypdf(file_path, max_chars)
+            return _run_with_timeout(_extract_pypdf, file_path, max_chars,
+                                     timeout=PDF_EXTRACT_TIMEOUT)
+    except FuturesTimeout:
+        logger.warning("PDF text extraction timed out (%ss): %s", PDF_EXTRACT_TIMEOUT, file_path)
+        return ""
     except Exception as e:
         logger.warning("Failed to extract PDF text from %s: %s", file_path, e)
         return ""