Report the setup failure in the task when available (#781)

src/api-service/__app__/onefuzzlib/agent_events.py

@@ -125,24 +125,30 @@ def on_state_update(
                 node_task.save()
 
     elif state == NodeState.done:
-        # if tasks are running on the node when it reports as Done
-        # those are stopped early
-        node.mark_tasks_stopped_early()
-        node.to_reimage(done=True)
-
         # Model-validated.
         #
         # This field will be required in the future.
         # For now, it is optional for back compat.
         done_data = cast(Optional[NodeDoneEventData], state_update.data)
+        error = None
         if done_data:
-            # TODO: do something with this done data
             if done_data.error:
+                error_text = done_data.json(exclude_none=True)
+                error = Error(
+                    code=ErrorCode.TASK_FAILED,
+                    errors=[error_text],
+                )
                 logging.error(
                     "node 'done' with error: machine_id:%s, data:%s",
                     machine_id,
-                    done_data,
+                    error_text,
                 )
+
+        # if tasks are running on the node when it reports as Done
+        # those are stopped early
+        node.mark_tasks_stopped_early(error=error)
+        node.to_reimage(done=True)
+
     return None
 
 

src/api-service/__app__/onefuzzlib/workers/nodes.py

@@ -220,18 +220,19 @@ class Node(BASE_NODE, ORMMixin):
         self.stop()
         return True
 
-    def mark_tasks_stopped_early(self) -> None:
+    def mark_tasks_stopped_early(self, error: Optional[Error] = None) -> None:
         from ..tasks.main import Task
 
+        if error is None:
+            error = Error(
+                code=ErrorCode.TASK_FAILED,
+                errors=["node reimaged during task execution"],
+            )
+
         for entry in NodeTasks.get_by_machine_id(self.machine_id):
             task = Task.get_by_task_id(entry.task_id)
             if isinstance(task, Task):
-                task.mark_failed(
-                    Error(
-                        code=ErrorCode.TASK_FAILED,
-                        errors=["node reimaged during task execution"],
-                    )
-                )
+                task.mark_failed(error)
 
     def could_shrink_scaleset(self) -> bool:
         from .scalesets import ScalesetShrinkQueue