fix(haystack): pass through top_k request to joiner, to get more results

Author: philwinder

api/pkg/rag/rag_haystack.go

@@ -233,6 +233,7 @@ func (h *HaystackRAG) Query(ctx context.Context, q *types.SessionRAGQuery) ([]*t
 			},
 		},
 	}
+	logger.Trace().Interface("query_request", queryReq).Msg("query request")
 
 	// Add document ID filter if there are any document IDs
 	if len(documentIDConditions) > 0 {

haystack_service/app/service.py

@@ -174,6 +174,7 @@ def _init_query_pipeline(self):
         # Save retrievers for parameter updates
         self.vector_retriever = vector_retriever
         self.bm25_retriever = bm25_retriever
+        self.document_joiner = document_joiner
 
         logger.info("Initialized query pipeline")
 
@@ -377,6 +378,10 @@ async def query(self, query_text: str, filters: Dict[str, Any] = None, top_k: in
 
         self.bm25_retriever.top_k = top_k
         self.bm25_retriever.filters = formatted_filters
+        # At this point there are 2*top_k results from the retrievers
+
+        logger.info(f"Chopping joined results in half to meet the request for top_k: {top_k}")
+        self.document_joiner.top_k = top_k
 
         # Set up the parameters for the query pipeline
         params = {