feat: add graph workflow fallback and retry metadata

backend/tests/domains/test_workflow_graph_runtime.py

@@ -16,8 +16,9 @@ from app.domains.orders.models import Order, OrderLine, OrderStatus
 from app.domains.products.models import CadFile, Product
 from app.domains.rendering.models import OutputType, RenderTemplate, WorkflowRun
 from app.domains.rendering.workflow_executor import prepare_workflow_context
-from app.domains.rendering.workflow_graph_runtime import execute_graph_workflow
+from app.domains.rendering.workflow_graph_runtime import WorkflowGraphRuntimeError, execute_graph_workflow
 from app.domains.rendering.workflow_run_service import create_workflow_run
+from app.domains.rendering.workflow_runtime_services import OrderLineRenderSetupResult
 
 import app.models  # noqa: F401
 
@@ -239,3 +240,116 @@ def test_execute_graph_workflow_persists_bridge_outputs_and_queues_render_task(
             {"InnerRing": "Steel", "OuterRing": "Rubber"},
         )
     ]
+
+
+def test_execute_graph_workflow_retries_bridge_node_and_persists_attempt_metadata(
+    sync_session,
+    monkeypatch,
+):
+    attempts = {"count": 0}
+
+    def _flaky_prepare(_session, _context_id):
+        attempts["count"] += 1
+        if attempts["count"] == 1:
+            raise RuntimeError("temporary setup failure")
+        return OrderLineRenderSetupResult(status="skip", reason="line_cancelled")
+
+    monkeypatch.setattr(
+        "app.domains.rendering.workflow_graph_runtime.prepare_order_line_render_context",
+        _flaky_prepare,
+    )
+
+    workflow_context = prepare_workflow_context(
+        {
+            "version": 1,
+            "nodes": [
+                {
+                    "id": "setup",
+                    "step": "order_line_setup",
+                    "params": {"retry_policy": {"max_attempts": 2}},
+                },
+            ],
+            "edges": [],
+        },
+        context_id=str(uuid.uuid4()),
+        execution_mode="graph",
+    )
+    run = create_workflow_run(
+        sync_session,
+        workflow_def_id=None,
+        order_line_id=None,
+        workflow_context=workflow_context,
+    )
+
+    dispatch_result = execute_graph_workflow(sync_session, workflow_context)
+    sync_session.commit()
+
+    refreshed_run = sync_session.execute(
+        select(WorkflowRun)
+        .where(WorkflowRun.id == run.id)
+        .options(selectinload(WorkflowRun.node_results))
+    ).scalar_one()
+    setup_result = next(node for node in refreshed_run.node_results if node.node_name == "setup")
+
+    assert dispatch_result.task_ids == []
+    assert refreshed_run.status == "completed"
+    assert setup_result.status == "skipped"
+    assert setup_result.output["attempt_count"] == 2
+    assert setup_result.output["max_attempts"] == 2
+    assert setup_result.output["retry_state"] == "recovered"
+    assert setup_result.output["last_error"] == "temporary setup failure"
+    assert setup_result.output["retry_policy"]["max_attempts"] == 2
+
+
+def test_execute_graph_workflow_marks_failed_node_with_retry_exhausted_metadata(
+    sync_session,
+    monkeypatch,
+):
+    monkeypatch.setattr(
+        "app.domains.rendering.workflow_graph_runtime.prepare_order_line_render_context",
+        lambda _session, _context_id: (_ for _ in ()).throw(RuntimeError("permanent setup failure")),
+    )
+
+    workflow_context = prepare_workflow_context(
+        {
+            "version": 1,
+            "nodes": [
+                {
+                    "id": "setup",
+                    "step": "order_line_setup",
+                    "params": {
+                        "retry_policy": {"max_attempts": 2},
+                        "failure_policy": {"fallback_to_legacy": True},
+                    },
+                },
+            ],
+            "edges": [],
+        },
+        context_id=str(uuid.uuid4()),
+        execution_mode="graph",
+    )
+    run = create_workflow_run(
+        sync_session,
+        workflow_def_id=None,
+        order_line_id=None,
+        workflow_context=workflow_context,
+    )
+
+    with pytest.raises(WorkflowGraphRuntimeError, match="permanent setup failure"):
+        execute_graph_workflow(sync_session, workflow_context)
+
+    sync_session.commit()
+
+    refreshed_run = sync_session.execute(
+        select(WorkflowRun)
+        .where(WorkflowRun.id == run.id)
+        .options(selectinload(WorkflowRun.node_results))
+    ).scalar_one()
+    setup_result = next(node for node in refreshed_run.node_results if node.node_name == "setup")
+
+    assert setup_result.status == "failed"
+    assert setup_result.output["attempt_count"] == 2
+    assert setup_result.output["max_attempts"] == 2
+    assert setup_result.output["retry_exhausted"] is True
+    assert setup_result.output["last_error"] == "permanent setup failure"
+    assert setup_result.output["failure_policy"]["fallback_to_legacy"] is True