Fixes Internal Server Error returned on simultaneous SignalWithStart requests for the same WF ID

service/history/handler.go

@@ -982,12 +982,38 @@ func (h *Handler) SignalWithStartWorkflowExecution(ctx context.Context, request
 		return nil, h.error(err1, scope, namespaceID, workflowID)
 	}
 
-	resp, err2 := engine.SignalWithStartWorkflowExecution(ctx, request)
-	if err2 != nil {
-		return nil, h.error(err2, scope, namespaceID, workflowID)
-	}
+	for {
+		resp, err2 := engine.SignalWithStartWorkflowExecution(ctx, request)
+		if err2 == nil {
+			return resp, nil
+		}
 
-	return resp, nil
+		// Two simultaneous SignalWithStart requests might try to start a workflow at the same time.
+		// This can result in one of the requests failing with one of two possible errors:
+		//    1) If it is a brand new WF ID, one of the requests can fail with WorkflowExecutionAlreadyStartedError
+		//       (createMode is persistence.CreateWorkflowModeBrandNew)
+		//    2) If it an already existing WF ID, one of the requests can fail with a CurrentWorkflowConditionFailedError
+		//       (createMode is persisetence.CreateWorkflowModeWorkflowIDReuse)
+		// If either error occurs, just go ahead and retry. It should succeed on the subsequent attempt.
+		// For simplicity, we keep trying unless the context finishes or we get an error that is not one of the
+		// two mentioned above.
+		_, isExecutionAlreadyStartedErr := err2.(*persistence.WorkflowExecutionAlreadyStartedError)
+		_, isWorkflowConditionFailedErr := err2.(*persistence.CurrentWorkflowConditionFailedError)
+
+		isContextDone := false
+		select {
+		case <-ctx.Done():
+			isContextDone = true
+			if ctxErr := ctx.Err(); ctxErr != nil {
+				err2 = ctxErr
+			}
+		default:
+		}
+
+		if (!isExecutionAlreadyStartedErr && !isWorkflowConditionFailedErr) || isContextDone {
+			return nil, h.error(err2, scope, namespaceID, workflowID)
+		}
+	}
 }
 
 // RemoveSignalMutableState is used to remove a signal request ID that was previously recorded.  This is currently