fix(router): return None instead of raising ValueError in least-busy strategy

- Replace raise ValueError with return None when no healthy deployments
  are available, matching the None-return contract used by lowest_tpm and
  lowest_latency strategies; the router's `if deployment is None` path
  then raises the proper RouterRateLimitError with cooldown context
- Add Optional[dict] return type annotation to _get_available_deployments
- Add test_should_return_none_when_no_healthy_deployments to cover the
  None-return path
- Seed random.seed(42) in probabilistic distribution tests for
  reproducible CI runs

Made-with: Cursor

litellm/router_strategy/least_busy.py

@@ -193,9 +193,11 @@ class LeastBusyLoggingHandler(CustomLogger):
         self,
         healthy_deployments: list,
         all_deployments: dict,
-    ):
+    ) -> Optional[dict]:
         """
-        Helper to get deployments using least busy strategy
+        Helper to get deployments using least busy strategy.
+        Returns None when no healthy deployments are available, consistent
+        with other routing strategies (lowest_tpm, lowest_latency).
         """
         healthy_ids = set()
         for d in healthy_deployments:
@@ -216,14 +218,14 @@ class LeastBusyLoggingHandler(CustomLogger):
                 min_deployment_ids.append(k)
 
         if not min_deployment_ids:
-            raise ValueError("No healthy deployments available")
+            return None
 
         chosen_id = random.choice(min_deployment_ids)
         for m in healthy_deployments:
             if m["model_info"]["id"] == chosen_id:
                 return m
 
-        raise ValueError(f"Chosen deployment id {chosen_id!r} not found in healthy_deployments")
+        return None
 
     def get_available_deployments(
         self,

tests/test_litellm/router_strategy/test_least_busy.py

@@ -1,4 +1,5 @@
 import os
+import random
 import sys
 from collections import Counter
 
@@ -24,6 +25,7 @@ class TestLeastBusyTieBreaking:
     """Tests that least-busy strategy distributes requests across tied deployments."""
 
     def test_should_randomly_distribute_when_all_counts_are_zero(self):
+        random.seed(42)
         cache = DualCache()
         handler = LeastBusyLoggingHandler(router_cache=cache)
         deployments = [_make_deployment(i) for i in range(3)]
@@ -44,6 +46,7 @@ class TestLeastBusyTieBreaking:
             )
 
     def test_should_randomly_distribute_when_counts_are_tied(self):
+        random.seed(42)
         cache = DualCache()
         handler = LeastBusyLoggingHandler(router_cache=cache)
         deployments = [_make_deployment(i) for i in range(2)]
@@ -65,6 +68,15 @@ class TestLeastBusyTieBreaking:
                 f"Deployment {dep_id} selected only {count}/200 times — distribution is too skewed"
             )
 
+    def test_should_return_none_when_no_healthy_deployments(self):
+        cache = DualCache()
+        handler = LeastBusyLoggingHandler(router_cache=cache)
+
+        result = handler._get_available_deployments(
+            healthy_deployments=[], all_deployments={}
+        )
+        assert result is None
+
     def test_should_pick_unique_minimum(self):
         cache = DualCache()
         handler = LeastBusyLoggingHandler(router_cache=cache)