fix(tasks): prevent filename overwrite in extract_metadata_with_gpt and fix embed retry logic

Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
2026-02-11 08:59:27 +00:00
parent 26dc165b37
commit f7799e409d
3 changed files with 28 additions and 33 deletions
@@ -456,35 +456,18 @@ def _retry_pipeline_step(file_record: FileRecord, step_name: str, db: Session) -
        filename = os.path.basename(file_record.local_filename)
        task = extract_metadata_with_gpt.delay(filename, extracted_text, file_id)
    elif step_name == "embed_metadata_into_pdf":
-        from app.tasks.embed_metadata_into_pdf import embed_metadata_into_pdf
-
-        # Retrieve the last successful metadata extraction result from processing logs
-        last_metadata_log = (
-            db.query(ProcessingLog)
-            .filter(
-                ProcessingLog.file_id == file_id,
-                ProcessingLog.step_name == "extract_metadata_with_gpt",
-                ProcessingLog.status == "success",
-            )
-            .order_by(ProcessingLog.timestamp.desc())
-            .first()
-        )
-        if not last_metadata_log:
-            raise HTTPException(
-                status_code=400,
-                detail="No successful metadata extraction found. Retry extract_metadata_with_gpt first.",
-            )
+        from app.tasks.extract_metadata_with_gpt import extract_metadata_with_gpt as extract_metadata_task

+        # Retrying embed requires re-running metadata extraction first, because
+        # embed_metadata_into_pdf needs the actual metadata dict (not empty).
+        # Re-trigger extract_metadata_with_gpt which will chain into embed_metadata_into_pdf.
        if not file_record.local_filename or not os.path.exists(file_record.local_filename):
            raise HTTPException(
                status_code=400, detail="Local file not found on disk. Cannot retry metadata embedding."
            )
        extracted_text = _extract_text_from_pdf(file_record.local_filename)
        filename = os.path.basename(file_record.local_filename)
-        # Empty metadata dict: embed_metadata_into_pdf will re-run with the provided text.
-        # The GPT extraction step must succeed first (validated above) to ensure
-        # the pipeline can produce new metadata during the subsequent re-extraction.
-        task = embed_metadata_into_pdf.delay(filename, extracted_text, {}, file_id)
+        task = extract_metadata_task.delay(filename, extracted_text, file_id)
    else:
        raise HTTPException(status_code=400, detail=f"Unsupported pipeline step: {step_name}")

@@ -126,17 +126,18 @@ def extract_metadata_with_gpt(self, filename: str, cleaned_text: str, file_id: i
        # SECURITY: Validate filename format from GPT to prevent path traversal
        # The prompt requests filenames with only letters, numbers, periods, and underscores
        # Enforce this constraint to prevent malicious filenames
-        import re
-        filename = metadata.get("filename", "")
-        if filename:
+        suggested_filename = metadata.get("filename", "")
+        if suggested_filename:
            # Check if filename contains only safe characters AND explicitly check for ".."
            # Defense in depth: While the regex [\w\-\. ]+ already excludes / and \,
            # we explicitly reject ".." to guard against:
            # 1. Potential locale-specific \w behavior
            # 2. Files literally named ".." which are valid but problematic
            # 3. Future code changes that might relax the regex
-            if not re.match(r'^[\w\-\. ]+$', filename) or ".." in filename:
-                logger.warning(f"[{task_id}] Invalid filename format from GPT: '{filename}', using fallback")
+            if not re.match(r'^[\w\-\. ]+$', suggested_filename) or ".." in suggested_filename:
+                logger.warning(
+                    f"[{task_id}] Invalid filename format from GPT: '{suggested_filename}', using fallback"
+                )
                # Reset to empty to trigger fallback to original filename
                metadata["filename"] = ""
        
@@ -146,6 +147,7 @@ def extract_metadata_with_gpt(self, filename: str, cleaned_text: str, file_id: i
        )

        # Trigger the next step: embedding metadata into the PDF
+        # Pass the original filename (UUID-based) so embed_metadata_into_pdf can find the file on disk
        logger.info(f"[{task_id}] Queueing metadata embedding task")
        log_task_progress(
            task_id, "extract_metadata_with_gpt", "success", "Metadata extracted, queuing embed task", file_id=file_id