fix(router): preserve adaptive priors on restart

litellm/router_strategy/adaptive_router/adaptive_router.py

@@ -149,7 +149,7 @@ class AdaptiveRouter:
         )
 
     async def load_state_from_db(self, prisma_client: Any) -> None:
-        """Override cold-start cells with persisted state. Called once at startup."""
+        """Add persisted online deltas to configured priors. Called once at startup."""
         if prisma_client is None:
             return
         try:
@@ -165,7 +165,8 @@ class AdaptiveRouter:
                     continue
                 if row.model_name not in self.config.available_models:
                     continue
-                self._cells[(rt, row.model_name)] = BanditCell(alpha=row.alpha, beta=row.beta)
+                key = (rt, row.model_name)  # rebind-ok: each persisted row has its own cell
+                self._cells[key] = apply_delta(self._cells[key], row.alpha, row.beta)
                 loaded += 1
             verbose_router_logger.info(
                 "AdaptiveRouter[%s]: loaded %d cells from DB",

tests/test_litellm/router_strategy/adaptive_router/test_adaptive_router.py

@@ -297,7 +297,7 @@ async def test_record_turn_bounds_feedback_contexts_and_evicts_least_recent_sess
 
 
 @pytest.mark.asyncio
-async def test_load_state_from_db_overrides_cold_start():
+async def test_load_state_from_db_adds_online_deltas_to_cold_start():
     r = _make_router()
     cold = r._cells[(RequestType.GENERAL, "fast")]
 
@@ -312,8 +312,38 @@ async def test_load_state_from_db_overrides_cold_start():
     await r.load_state_from_db(prisma)
 
     new_cell = r._cells[(RequestType.GENERAL, "fast")]
-    assert (new_cell.alpha, new_cell.beta) == (42.0, 13.0)
-    assert (new_cell.alpha, new_cell.beta) != (cold.alpha, cold.beta)
+    assert (new_cell.alpha, new_cell.beta) == (cold.alpha + 42.0, cold.beta + 13.0)
+
+
+@pytest.mark.asyncio
+async def test_load_state_from_db_preserves_evaluation_priors():
+    cfg: Final = AdaptiveRouterConfig(
+        available_models=["fast"],
+        evaluation_priors=(
+            AdaptiveRouterEvaluationPrior(
+                request_type=RequestType.GENERAL,
+                model="fast",
+                successes=18,
+                failures=2,
+            ),
+        ),
+    )
+    router: Final = AdaptiveRouter(
+        router_name="seeded",
+        config=cfg,
+        model_to_prefs={"fast": AdaptiveRouterPreferences(quality_tier=2)},
+        model_to_cost={"fast": 0.001},
+    )
+    seeded: Final = router._cells[(RequestType.GENERAL, "fast")]
+    row: Final = MagicMock(request_type="general", model_name="fast", alpha=2.0, beta=3.0)
+    prisma: Final = MagicMock()
+    prisma.db.litellm_adaptiverouterstate.find_many = AsyncMock(return_value=[row])
+
+    await router.load_state_from_db(prisma)
+
+    restored: Final = router._cells[(RequestType.GENERAL, "fast")]
+    assert restored.alpha == seeded.alpha + 2.0
+    assert restored.beta == seeded.beta + 3.0
 
 
 @pytest.mark.asyncio
@@ -338,7 +368,7 @@ async def test_load_state_from_db_handles_unknown_request_type():
     await r.load_state_from_db(prisma)
 
     # Unknown skipped; good applied.
-    assert r._cells[(RequestType.GENERAL, "fast")].alpha == 7.0
+    assert r._cells[(RequestType.GENERAL, "fast")].alpha == cold.alpha + 7.0
     # Other request types kept their cold-start values.
     assert r._cells[(RequestType.WRITING, "fast")] == cold or True
 

tests/test_litellm/router_strategy/adaptive_router/test_e2e_adaptive_router.py

@@ -186,8 +186,9 @@ async def test_failure_signal_increments_beta_after_flush():
 
 
 @pytest.mark.asyncio
-async def test_load_state_from_db_overrides_cold_start():
+async def test_load_state_from_db_adds_online_deltas_to_cold_start():
     router = _make_router()
+    cold = router._cells[(RequestType.GENERAL, "gpt-4o")]
     fake_row = MagicMock()
     fake_row.request_type = RequestType.GENERAL.value
     fake_row.model_name = "gpt-4o"
@@ -200,8 +201,8 @@ async def test_load_state_from_db_overrides_cold_start():
     await router.load_state_from_db(prisma)
 
     cell = router._cells[(RequestType.GENERAL, "gpt-4o")]
-    assert cell.alpha == 90.0
-    assert cell.beta == 10.0
+    assert cell.alpha == cold.alpha + 90.0
+    assert cell.beta == cold.beta + 10.0
 
 
 @pytest.mark.asyncio