fix: preserve timing metrics in parallel iteration (#33216)

api/core/app/apps/common/workflow_response_converter.py

@@ -517,7 +517,7 @@ class WorkflowResponseConverter:
         snapshot = self._pop_snapshot(event.node_execution_id)
 
         start_at = snapshot.start_at if snapshot else event.start_at
-        finished_at = naive_utc_now()
+        finished_at = event.finished_at or naive_utc_now()
         elapsed_time = (finished_at - start_at).total_seconds()
 
         inputs, inputs_truncated = self._truncate_mapping(event.inputs)

api/core/app/apps/workflow_app_runner.py

@@ -456,6 +456,7 @@ class WorkflowBasedAppRunner:
                     node_id=event.node_id,
                     node_type=event.node_type,
                     start_at=event.start_at,
+                    finished_at=event.finished_at,
                     inputs=inputs,
                     process_data=process_data,
                     outputs=outputs,
@@ -471,6 +472,7 @@ class WorkflowBasedAppRunner:
                     node_id=event.node_id,
                     node_type=event.node_type,
                     start_at=event.start_at,
+                    finished_at=event.finished_at,
                     inputs=event.node_run_result.inputs,
                     process_data=event.node_run_result.process_data,
                     outputs=event.node_run_result.outputs,
@@ -487,6 +489,7 @@ class WorkflowBasedAppRunner:
                     node_id=event.node_id,
                     node_type=event.node_type,
                     start_at=event.start_at,
+                    finished_at=event.finished_at,
                     inputs=event.node_run_result.inputs,
                     process_data=event.node_run_result.process_data,
                     outputs=event.node_run_result.outputs,

api/core/app/entities/queue_entities.py

@@ -335,6 +335,7 @@ class QueueNodeSucceededEvent(AppQueueEvent):
     in_loop_id: str | None = None
     """loop id if node is in loop"""
     start_at: datetime
+    finished_at: datetime | None = None
 
     inputs: Mapping[str, object] = Field(default_factory=dict)
     process_data: Mapping[str, object] = Field(default_factory=dict)
@@ -390,6 +391,7 @@ class QueueNodeExceptionEvent(AppQueueEvent):
     in_loop_id: str | None = None
     """loop id if node is in loop"""
     start_at: datetime
+    finished_at: datetime | None = None
 
     inputs: Mapping[str, object] = Field(default_factory=dict)
     process_data: Mapping[str, object] = Field(default_factory=dict)
@@ -414,6 +416,7 @@ class QueueNodeFailedEvent(AppQueueEvent):
     in_loop_id: str | None = None
     """loop id if node is in loop"""
     start_at: datetime
+    finished_at: datetime | None = None
 
     inputs: Mapping[str, object] = Field(default_factory=dict)
     process_data: Mapping[str, object] = Field(default_factory=dict)

api/core/app/workflow/layers/persistence.py

@@ -268,7 +268,12 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
 
     def _handle_node_succeeded(self, event: NodeRunSucceededEvent) -> None:
         domain_execution = self._get_node_execution(event.id)
-        self._update_node_execution(domain_execution, event.node_run_result, WorkflowNodeExecutionStatus.SUCCEEDED)
+        self._update_node_execution(
+            domain_execution,
+            event.node_run_result,
+            WorkflowNodeExecutionStatus.SUCCEEDED,
+            finished_at=event.finished_at,
+        )
 
     def _handle_node_failed(self, event: NodeRunFailedEvent) -> None:
         domain_execution = self._get_node_execution(event.id)
@@ -277,6 +282,7 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
             event.node_run_result,
             WorkflowNodeExecutionStatus.FAILED,
             error=event.error,
+            finished_at=event.finished_at,
         )
 
     def _handle_node_exception(self, event: NodeRunExceptionEvent) -> None:
@@ -286,6 +292,7 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
             event.node_run_result,
             WorkflowNodeExecutionStatus.EXCEPTION,
             error=event.error,
+            finished_at=event.finished_at,
         )
 
     def _handle_node_pause_requested(self, event: NodeRunPauseRequestedEvent) -> None:
@@ -352,13 +359,14 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
         *,
         error: str | None = None,
         update_outputs: bool = True,
+        finished_at: datetime | None = None,
     ) -> None:
-        finished_at = naive_utc_now()
+        actual_finished_at = finished_at or naive_utc_now()
         snapshot = self._node_snapshots.get(domain_execution.id)
         start_at = snapshot.created_at if snapshot else domain_execution.created_at
         domain_execution.status = status
-        domain_execution.finished_at = finished_at
-        domain_execution.elapsed_time = max((finished_at - start_at).total_seconds(), 0.0)
+        domain_execution.finished_at = actual_finished_at
+        domain_execution.elapsed_time = max((actual_finished_at - start_at).total_seconds(), 0.0)
 
         if error:
             domain_execution.error = error