feat(tuning): release V0.8.2 Agent 界面重构

training_server/tests/test_tuning.py

@@ -12,6 +12,7 @@ from tuning.schema import (  # noqa: E402
     RewardConfigError,
     merge_proposal,
     validate_configuration,
+    validate_constraints,
     validate_proposal,
 )
 from tuning.scoring import (  # noqa: E402
@@ -19,7 +20,7 @@ from tuning.scoring import (  # noqa: E402
     EvaluationError,
     score_evaluation,
 )
-from tuning.storage import TuningStorage  # noqa: E402
+from tuning.storage import StorageConflict, TuningStorage  # noqa: E402
 
 
 class RewardSchemaTest(unittest.TestCase):
@@ -69,6 +70,33 @@ class RewardSchemaTest(unittest.TestCase):
                 BASE_REWARD_CONFIGURATION,
             )
 
+    def test_session_constraints_reject_unknown_out_of_range_and_fixed_changes(self):
+        constraints = validate_constraints(
+            {
+                "weights.track_linear_velocity": {"kind": "fixed", "value": 1.0},
+                "params.foot_gait.period": {"kind": "range", "min": 0.5, "max": 0.7},
+            }
+        )
+        validate_proposal(
+            {"params": {"foot_gait.period": 0.65}},
+            BASE_REWARD_CONFIGURATION,
+            constraints,
+        )
+        with self.assertRaisesRegex(RewardConfigError, "已固定"):
+            validate_proposal(
+                {"weights": {"track_linear_velocity": 1.1}},
+                BASE_REWARD_CONFIGURATION,
+                constraints,
+            )
+        with self.assertRaisesRegex(RewardConfigError, "工程锁定范围"):
+            validate_proposal(
+                {"params": {"foot_gait.period": 0.75}},
+                BASE_REWARD_CONFIGURATION,
+                constraints,
+            )
+        with self.assertRaisesRegex(RewardConfigError, "未知参数约束"):
+            validate_constraints({"weights.not_allowed": {"kind": "fixed", "value": 1.0}})
+
 
 class ScoringTest(unittest.TestCase):
     baseline = {
@@ -174,8 +202,24 @@ class StorageTest(unittest.TestCase):
         self.assertEqual(sampled[0]["step"], 0)
         self.assertEqual(sampled[-1]["step"], 99)
         self.assertIn(50.0, [point["value"] for point in sampled])
+        control = self.storage.replace_constraints(
+            session["id"],
+            0,
+            {"weights.pose": {"kind": "range", "min": 0.5, "max": 1.5}},
+        )
+        self.assertEqual(control["constraintsRevision"], 1)
+        with self.assertRaises(StorageConflict):
+            self.storage.replace_constraints(session["id"], 0, {})
+        self.storage.grant_dispatch_token(session["id"])
+        with self.assertRaises(StorageConflict):
+            self.storage.grant_dispatch_token(session["id"])
+        self.assertTrue(self.storage.use_dispatch_token(session["id"]))
+        self.assertFalse(self.storage.use_dispatch_token(session["id"]))
+        incremental = self.storage.metrics(trial["id"], max_points=100, after_step=90)[0]
+        self.assertEqual(incremental["points"][0]["step"], 91)
         reopened = TuningStorage(self.storage.path)
         self.assertEqual(reopened.get_session(session["id"])["mode"], "approval")
+        self.assertEqual(reopened.get_control(session["id"])["constraintsRevision"], 1)
 
     def test_recovery_marks_inflight_records(self):
         session = self.storage.create_session(