[Cache] Forward fast prefix matching capability (#37844)

python/sglang/srt/mem_cache/unified_cache/unified_tree_core_interface.py

@@ -369,6 +369,10 @@ class UnifiedTreeCoreInterface(ABC):
         """Match a key against the tree; returns device indices + boundary NodeIds."""
         ...
 
+    def supports_fast_match_prefix(self) -> bool:
+        """Whether matching every waiting request is cheap enough for scheduling."""
+        return False
+
     @property
     @abstractmethod
     def empty_match_result(self) -> MatchResult:

python/sglang/srt/mem_cache/unified_radix_cache.py

@@ -538,6 +538,9 @@ class UnifiedRadixCache(BasePrefixCache):
             result = self.linker.match(params.key, params.req, result)
         return result
 
+    def supports_fast_match_prefix(self) -> bool:
+        return self.tree_core.supports_fast_match_prefix()
+
     def is_chunk_cache(self) -> bool:
         return self.disable
 

test/registered/unit/managers/test_load_inquirer.py

@@ -31,11 +31,14 @@ class TestSchedulePolicyWaitingQueueMatching(unittest.TestCase):
         policy.tree_cache = SimpleNamespace(supports_fast_match_prefix=lambda: True)
         self.assertTrue(policy.waiting_queue_prefix_matched([]))
 
-    def test_lpm_queue_limit_can_disable_matching(self):
+    def test_lpm_queue_limit_respects_fast_matching_capability(self):
         policy = self.make_policy(CacheAwarePolicy.LPM, False)
         self.assertTrue(policy.waiting_queue_prefix_matched([None] * 128))
         self.assertFalse(policy.waiting_queue_prefix_matched([None] * 129))
 
+        policy.tree_cache = SimpleNamespace(supports_fast_match_prefix=lambda: True)
+        self.assertTrue(policy.waiting_queue_prefix_matched([None] * 129))
+
 
 class TestSchedulerLoadInquirer(unittest.TestCase):
     def make_inquirer(self, waiting_queue_prefix_matched):