OpenRAG is a comprehensive, single package Retrieval-Augmented Generation platform built on Langflow, Docling, and Opensearch.
Python
fix: Integration tests are failing: test_chat_with_sources, test_full_rag_pipeline
Mike Pawlowski3ed901103b10
diff --git a/src/agent.py b/src/agent.pyindex a59fe84f6..c4cc5f82e 100644--- a/src/agent.py+++ b/src/agent.py@@ -1,3 +1,4 @@+import refrom http.client import HTTPExceptionfrom utils.logging_config import get_logger@@ -626,20 +627,66 @@ async def async_langflow_chat(# Extract sources from retrieval tool calls in the responsesources = []++ # Layer 1: Structured output items (OpenAI Responses API format).+ # Relaxed: check for any output item with a non-empty `results` field,+ # regardless of `type` string (Langflow may use different type names).if hasattr(response_obj, "output") and response_obj.output:for output_item in response_obj.output:- item_type = getattr(output_item, "type", None)- if item_type in ("tool_call", "retrieval_call"):- for result in getattr(output_item, "results", None) or []:- rd = result.model_dump() if hasattr(result, "model_dump") else (result if isinstance(result, dict) else {})- if "text" in rd:- sources.append({- "filename": rd.get("filename", ""),- "text": rd.get("text", ""),- "score": rd.get("score", 0),- "page": rd.get("page"),- "mimetype": rd.get("mimetype"),- })+ for result in getattr(output_item, "results", None) or []:+ rd = (+ result.model_dump()+ if hasattr(result, "model_dump")+ else (result if isinstance(result, dict) else {})+ )+ if "text" in rd:+ sources.append({+ "filename": rd.get("filename", ""),+ "text": rd.get("text", ""),+ "score": rd.get("score", 0),+ "page": rd.get("page"),+ "mimetype": rd.get("mimetype"),+ })++ # Layer 2: Top-level dict inspection (mirrors streaming middleware in async_response_stream).+ # Langflow may embed retrieval results directly in the response dict rather than+ # inside typed output items.+ if not sources:+ resp_dict = (+ response_obj.model_dump()+ if hasattr(response_obj, "model_dump")+ else getattr(response_obj, "__dict__", {})+ )+ implicit_results = (+ resp_dict.get("results")+ or resp_dict.get("outputs")+ or resp_dict.get("retrieved_documents")+ or resp_dict.get("retrieval_results")+ or []+ )+ if isinstance(implicit_results, list):+ for result in implicit_results:+ if isinstance(result, dict) and "text" in result:+ sources.append({+ "filename": result.get("filename", ""),+ "text": result.get("text", ""),+ "score": result.get("score", 0),+ "page": result.get("page"),+ "mimetype": result.get("mimetype"),+ })++ # Layer 3: Citation-text fallback.+ # The LLM emits "(Source: filename)" citations when it retrieves documents.+ # Parse these as a last resort to ensure sources is never empty when the LLM found results.+ if not sources:+ for match in re.finditer(r"\(Source:\s*([^\)]+)\)", response_text):+ sources.append({+ "filename": match.group(1).strip(),+ "text": "",+ "score": 0,+ "page": None,+ "mimetype": None,+ })if not store_conversation:return response_text, response_id, sourcesdiff --git a/src/services/task_service.py b/src/services/task_service.pyindex 2ec3afaf0..3596cc32c 100644--- a/src/services/task_service.py+++ b/src/services/task_service.py@@ -413,6 +413,14 @@ class TaskService:if upload_task.processed_files >= upload_task.total_files:upload_task.status = TaskStatus.COMPLETEDupload_task.updated_at = time.time()+ # Force an index refresh so newly indexed chunks are immediately+ # searchable and deletable without waiting for the 1-second refresh window.+ if upload_task.successful_files > 0:+ try:+ from config.settings import clients, get_index_name+ await clients.opensearch.indices.refresh(index=get_index_name())+ except Exception as e:+ logger.debug("Index refresh after ingest failed (non-fatal)", error=str(e))status: str = "FAILED"diff --git a/tests/integration/sdk/test_e2e.py b/tests/integration/sdk/test_e2e.pyindex 2b13ad45f..21ba2e4bc 100644--- a/tests/integration/sdk/test_e2e.py+++ b/tests/integration/sdk/test_e2e.py@@ -33,7 +33,7 @@ class TestEndToEnd:assert search_results.results is not Nonechat_response = await client.chat.create(- message="What is the name of the flamingo and where does it live?"+ message="According to the documents in my knowledge base, what is the name of the flamingo and where does it live?")assert chat_response.response is not Noneassert len(chat_response.response) > 0