Rename _alive_streaming_session_count; use _is_streaming helper (#22755)

python/sglang/srt/managers/scheduler_runtime_checker_mixin.py

@@ -134,7 +134,7 @@ class PoolStats:
 
 
 class SchedulerRuntimeCheckerMixin:
-    def _alive_streaming_session_count(self: Scheduler) -> int:
+    def _streaming_session_count(self: Scheduler) -> int:
         return sum(
             1
             for session in self.session_controller.sessions.values()
@@ -506,7 +506,7 @@ class SchedulerRuntimeCheckerMixin:
             return
 
         self.get_pool_stats().update_scheduler_stats(self.stats)
-        self.stats.num_streaming_sessions = self._alive_streaming_session_count()
+        self.stats.num_streaming_sessions = self._streaming_session_count()
         self.stats.streaming_session_held_tokens = self._session_held_tokens()
 
         priority_enabled = self.enable_priority_scheduling

python/sglang/srt/managers/session_controller.py

@@ -312,7 +312,7 @@ class SessionController:
             )
             return
 
-        # No active request -- safe to release immediately.
+        # No owning request -- safe to release immediately.
         if session.streaming and session.req_nodes:
             req = next(iter(session.req_nodes.values())).req
             req.session = None

python/sglang/srt/mem_cache/common.py

@@ -9,7 +9,7 @@ import triton.language as tl
 
 from sglang.srt.mem_cache.base_prefix_cache import BasePrefixCache, EvictParams
 from sglang.srt.mem_cache.memory_pool import HybridReqToTokenPool, ReqToTokenPool
-from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache
+from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache, _is_streaming
 from sglang.srt.mem_cache.swa_memory_pool import SWATokenToKVPoolAllocator
 from sglang.srt.server_args import get_global_server_args
 from sglang.srt.utils import support_triton
@@ -487,10 +487,8 @@ def release_kv_cache(req: Req, tree_cache: BasePrefixCache, is_insert: bool = Tr
     # cache_finished_req below (which also sets req_pool_idx = None).
     from sglang.srt.managers.schedule_batch import FINISH_ABORT
 
-    is_streaming_session = (
+    is_streaming_session = isinstance(tree_cache, SessionAwareCache) and _is_streaming(
-        isinstance(tree_cache, SessionAwareCache)
+        req
-        and getattr(req, "session", None) is not None
-        and req.session.streaming
     )
     is_aborted_streaming = is_streaming_session and isinstance(
         getattr(req, "finished_reason", None), FINISH_ABORT

python/sglang/srt/observability/metrics_collector.py

@@ -686,7 +686,7 @@ class SchedulerMetricsCollector:
         if self.enable_streaming_session:
             self.num_streaming_sessions = Gauge(
                 name="sglang:num_streaming_sessions",
-                documentation="The number of active streaming sessions.",
+                documentation="The number of streaming sessions.",
                 labelnames=labels.keys(),
                 multiprocess_mode="mostrecent",
             )

python/sglang/srt/observability/scheduler_metrics_mixin.py

@@ -604,7 +604,7 @@ class SchedulerMetricsMixin:
             self.stats.cache_hit_rate = cache_hit_rate
 
             self.stats.max_total_num_tokens = self.max_total_num_tokens
-            self.stats.num_streaming_sessions = self._alive_streaming_session_count()
+            self.stats.num_streaming_sessions = self._streaming_session_count()
             self.stats.streaming_session_held_tokens = self._session_held_tokens()
 
             # Speculative decoding