changes

src/scheduler.py

@@ -4,7 +4,7 @@ import csv
 import io
 import json
 import chardet
-from LibreQoS import ValidationFailure, refreshShapers, refreshShapersUpdateOnly
+from LibreQoS import RefreshFailure, ValidationFailure, refreshShapers, refreshShapersUpdateOnly
 import subprocess
 import sys
 import tempfile
@@ -1026,7 +1026,12 @@ def importAndShapeFullReload():
         return False
     publish_scheduler_progress(True, "initial_shaping_reload", "Refreshing shaper state", 5, SCHEDULER_STARTUP_STEP_COUNT)
     if not enable_insight_topology():
-        refreshShapers()
+        try:
+            refreshShapers()
+        except Exception as e:
+            if _defer_stale_shaping_inputs_failure(e, startup=True):
+                return False
+            raise
         shaping_runtime_hash = calculate_shaping_runtime_hash()
     else:
         shaping_runtime_hash = calculate_shaping_runtime_hash()
@@ -1072,6 +1077,8 @@ def importAndShapePartialReload():
             report_scheduler_validation_failure("Scheduled shaping refresh blocked by validation", e)
             return
         except Exception as e:
+            if _defer_stale_shaping_inputs_failure(e, startup=False):
+                return
             report_scheduler_runtime_failure("Scheduled shaping refresh failed", e)
             return
         shaping_runtime_hash = calculate_shaping_runtime_hash()
@@ -1223,6 +1230,29 @@ def _topology_runtime_not_ready_is_transient(detail: str) -> bool:
     return True
 
 
+def _is_transient_stale_shaping_inputs_failure(exc: Exception) -> bool:
+    if not isinstance(exc, RefreshFailure):
+        return False
+    detail = str(exc or "").strip()
+    return detail.startswith("Missing or stale shaping_inputs.json.")
+
+
+def _defer_stale_shaping_inputs_failure(exc: Exception, *, startup: bool) -> bool:
+    if not _is_transient_stale_shaping_inputs_failure(exc):
+        return False
+
+    ready, detail, generation = topology_runtime_readiness_detail()
+    if detail and not _topology_runtime_not_ready_is_transient(detail):
+        return False
+
+    clear_scheduler_error()
+    if startup:
+        _begin_startup_topology_runtime_wait(generation if not ready else current_topology_source_generation())
+    else:
+        _begin_partial_topology_runtime_wait(generation if not ready else current_topology_source_generation())
+    return True
+
+
 def _publish_startup_topology_runtime_wait(generation: str | None):
     global startup_topology_runtime_last_report_state
 
@@ -1349,6 +1379,8 @@ def _continue_startup_topology_runtime_wait():
             shaping_runtime_hash = 0
             return
         except Exception as e:
+            if _defer_stale_shaping_inputs_failure(e, startup=True):
+                return
             _reset_startup_topology_runtime_wait()
             report_scheduler_runtime_failure(
                 "Scheduler startup shaping refresh failed",
@@ -1453,6 +1485,8 @@ def _continue_partial_topology_runtime_wait():
                 )
                 return
             except Exception as e:
+                if _defer_stale_shaping_inputs_failure(e, startup=False):
+                    return
                 _reset_partial_topology_runtime_wait()
                 report_scheduler_runtime_failure("Scheduled shaping refresh failed", e)
                 return
@@ -1546,6 +1580,8 @@ def topology_runtime_refresh_tick():
         report_scheduler_validation_failure("Topology runtime refresh blocked by validation", e)
         return
     except Exception as e:
+        if _defer_stale_shaping_inputs_failure(e, startup=False):
+            return
         report_scheduler_runtime_failure("Topology runtime refresh failed", e)
         return
     shaping_runtime_hash = calculate_shaping_runtime_hash()

src/test_scheduler.py

@@ -22,8 +22,11 @@ scheduler = None
 
 def install_scheduler_stubs():
     libre = types.ModuleType("LibreQoS")
+    class RefreshFailure(Exception):
+        pass
     class ValidationFailure(Exception):
         pass
+    libre.RefreshFailure = RefreshFailure
     libre.ValidationFailure = ValidationFailure
     libre.refreshShapers = Mock()
     libre.refreshShapersUpdateOnly = Mock()
@@ -377,6 +380,55 @@ class TestSchedulerLogging(unittest.TestCase):
             "Topology runtime refresh failed: bad runtime"
         )
 
+    def test_topology_runtime_refresh_tick_waits_for_stale_shaping_inputs(self):
+        stale_exc = scheduler.RefreshFailure(
+            "Missing or stale shaping_inputs.json. Run topology runtime before shaping."
+        )
+
+        with patch.object(scheduler, "ensure_topology_runtime_process"):
+            with patch.object(
+                scheduler,
+                "topology_runtime_readiness_detail",
+                return_value=(True, "", "generation-1"),
+            ):
+                with patch.object(
+                    scheduler,
+                    "current_topology_source_generation",
+                    return_value="generation-1",
+                ):
+                    with patch.object(scheduler, "calculate_shaping_runtime_hash", return_value=5):
+                        with patch.object(
+                            scheduler,
+                            "refreshShapers",
+                            side_effect=stale_exc,
+                        ):
+                            with patch.object(scheduler, "clear_scheduler_error") as mock_clear_error:
+                                with patch.object(
+                                    scheduler,
+                                    "report_scheduler_runtime_failure",
+                                ) as mock_report:
+                                    with patch.object(
+                                        scheduler,
+                                        "publish_scheduler_progress",
+                                    ) as mock_progress:
+                                        with patch("builtins.print"):
+                                            scheduler.shaping_runtime_hash = 1
+                                            scheduler.topology_runtime_refresh_tick()
+
+        mock_clear_error.assert_called_once()
+        mock_report.assert_not_called()
+        self.assertTrue(scheduler.partial_topology_runtime_pending)
+        self.assertTrue(
+            any(
+                call.args[:3] == (
+                    True,
+                    "waiting_for_topology_runtime",
+                    "Waiting for topology runtime",
+                )
+                for call in mock_progress.call_args_list
+            )
+        )
+
     def test_topology_runtime_refresh_tick_reports_validation_failure(self):
         validation_exc = scheduler.ValidationFailure("Validation failed. Will now exit.")
 
@@ -689,6 +741,59 @@ class TestSchedulerLogging(unittest.TestCase):
             )
         )
 
+    def test_partial_reload_waits_for_stale_shaping_inputs(self):
+        stale_exc = scheduler.RefreshFailure(
+            "Missing or stale shaping_inputs.json. Run topology runtime before shaping."
+        )
+
+        with patch.object(scheduler, "importFromCRM"):
+            with patch.object(scheduler, "ensure_topology_runtime_process", return_value=True):
+                with patch.object(scheduler, "calculate_shaping_runtime_hash", return_value=2):
+                    with patch.object(scheduler, "shaping_runtime_hash", 1):
+                        with patch.object(
+                            scheduler,
+                            "topology_runtime_readiness_detail",
+                            return_value=(True, "", "generation-1"),
+                        ):
+                            with patch.object(
+                                scheduler,
+                                "current_topology_source_generation",
+                                return_value="generation-1",
+                            ):
+                                with patch.object(
+                                    scheduler,
+                                    "refreshShapers",
+                                    side_effect=stale_exc,
+                                ):
+                                    with patch.object(
+                                        scheduler,
+                                        "clear_scheduler_error",
+                                    ) as mock_clear_error:
+                                        with patch.object(
+                                            scheduler,
+                                            "report_scheduler_runtime_failure",
+                                        ) as mock_report:
+                                            with patch.object(
+                                                scheduler,
+                                                "publish_scheduler_progress",
+                                            ) as mock_progress:
+                                                with patch("builtins.print"):
+                                                    scheduler.importAndShapePartialReload()
+
+        mock_clear_error.assert_called_once()
+        mock_report.assert_not_called()
+        self.assertTrue(scheduler.partial_topology_runtime_pending)
+        self.assertTrue(
+            any(
+                call.args[:3] == (
+                    True,
+                    "waiting_for_topology_runtime",
+                    "Waiting for topology runtime",
+                )
+                for call in mock_progress.call_args_list
+            )
+        )
+
 
 class TestTopologyRuntimeReadiness(unittest.TestCase):
     def test_missing_status_is_not_ready(self):
@@ -898,6 +1003,54 @@ class TestTopologyRuntimeGating(unittest.TestCase):
         mock_refresh.assert_not_called()
         mock_report.assert_called_once()
 
+    def test_full_reload_waits_for_stale_shaping_inputs(self):
+        stale_exc = scheduler.RefreshFailure(
+            "Missing or stale shaping_inputs.json. Run topology runtime before shaping."
+        )
+
+        with patch.object(scheduler, "importFromCRM"):
+            with patch.object(scheduler, "ensure_topology_runtime_process", return_value=True):
+                with patch.object(scheduler, "enable_insight_topology", return_value=False):
+                    with patch.object(
+                        scheduler,
+                        "topology_runtime_readiness_detail",
+                        return_value=(True, "", "generation-1"),
+                    ):
+                        with patch.object(
+                            scheduler,
+                            "current_topology_source_generation",
+                            return_value="generation-1",
+                        ):
+                            with patch.object(
+                                scheduler,
+                                "refreshShapers",
+                                side_effect=stale_exc,
+                            ):
+                                with patch.object(
+                                    scheduler,
+                                    "clear_scheduler_error",
+                                ) as mock_clear_error:
+                                    with patch.object(
+                                        scheduler,
+                                        "publish_scheduler_progress",
+                                    ) as mock_progress:
+                                        with patch("builtins.print"):
+                                            ready = scheduler.importAndShapeFullReload()
+
+        self.assertFalse(ready)
+        mock_clear_error.assert_called_once()
+        self.assertTrue(scheduler.startup_topology_runtime_pending)
+        self.assertTrue(
+            any(
+                call.args[:3] == (
+                    True,
+                    "waiting_for_topology_runtime",
+                    "Waiting for topology runtime",
+                )
+                for call in mock_progress.call_args_list
+            )
+        )
+
 
 class TestSchedulerOverrideMerge(unittest.TestCase):
     def test_merge_rows_replaces_matching_device_id(self):