[Metrics] Propagate idle gaps across all scheduler loops (#40004)

Co-authored-by: metamergebot <324680979+metamergebot@users.noreply.github.com>
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>

python/sglang/srt/disaggregation/decode.py

@@ -2656,6 +2656,7 @@ class SchedulerDisaggregationDecodeMixin:
                 self.process_batch_result(batch, result)
             else:
                 # When the server is idle, do self-check and re-init some states
+                self._sched_idled = True
                 self.on_idle()
 
             # Update last_batch
@@ -2707,6 +2708,7 @@ class SchedulerDisaggregationDecodeMixin:
                 self.result_queue.append((batch.copy(), batch_result))
             else:
                 batch_result = None
+                self._sched_idled = True
 
             # Process the last batch
             if self.last_batch:

python/sglang/srt/disaggregation/prefill.py

@@ -659,6 +659,7 @@ class SchedulerDisaggregationPrefillMixin:
                 result = self.run_batch(batch)
                 self.process_batch_result(batch, result)
             else:
+                self._sched_idled = True
                 self.on_idle()
 
             self.process_disagg_prefill_inflight_queue()
@@ -700,6 +701,7 @@ class SchedulerDisaggregationPrefillMixin:
                 self.result_queue.append((batch.copy(), batch_result))
             else:
                 batch_result = None
+                self._sched_idled = True
 
             # Process the last batch
             if self.last_batch:

python/sglang/srt/hardware_backend/mlx/scheduler_mixin.py

@@ -98,6 +98,8 @@ class SchedulerMlxOverlapMixin:
         self.forward_ct += 1
         batch.forward_iter = self.forward_ct
         batch.launch_ts = time.monotonic()
+        batch.after_idle_gap = self._sched_idled
+        self._sched_idled = False
         self.profiler_manager._profile_batch_predicate(batch)
 
     def _finalize_mlx_pending_job(self: Scheduler, pending: MlxPendingJob):
@@ -273,6 +275,7 @@ class SchedulerMlxOverlapMixin:
                 pending_curr = _launch_fresh(next_batch)
                 self.result_queue.append(pending_curr)
             else:
+                self._sched_idled = True
                 self.on_idle()
 
             self.last_batch = next_batch

python/sglang/srt/managers/scheduler_pp_mixin.py

@@ -184,6 +184,7 @@ class SchedulerPPMixin:
 
             # When the server is idle, self-check and re-init some states
             if server_is_idle:
+                self._sched_idled = True
                 self.on_idle()
 
     @DynamicGradMode()
@@ -362,8 +363,10 @@ class SchedulerPPMixin:
                 self.running_batch.batch_is_full = False
 
             # When the server is idle, self-check and re-init some states
-            if server_is_idle and len(self.disagg_prefill_inflight_queue) == 0:
-                self.on_idle()
+            if server_is_idle:
+                self._sched_idled = True
+                if len(self.disagg_prefill_inflight_queue) == 0:
+                    self.on_idle()
 
     @DynamicGradMode()
     def event_loop_pp_disagg_decode(self: Scheduler):
@@ -552,8 +555,10 @@ class SchedulerPPMixin:
             if get_disagg().disaggregation_decode_enable_offload_kvcache:
                 queue_size += len(self.decode_offload_manager.ongoing_offload)
 
-            if server_is_idle and queue_size == 0:
-                self.on_idle()
+            if server_is_idle:
+                self._sched_idled = True
+                if queue_size == 0:
+                    self.on_idle()
 
     def init_pp_loop_state(self: Scheduler):
         self.pp_loop_size: int = self.ps.pp_size + get_parallel().pp_async_batch_depth

python/sglang/srt/multiplex/multiplexing_mixin.py

@@ -135,6 +135,7 @@ class SchedulerMultiplexMixin:
                     stream_idx > 0 and running_batch.is_empty()
                 )
                 if running_batch.is_empty() and self.split_prefill_batch is None:
+                    self._sched_idled = True
                     self.on_idle()
 
             if adjust_stream_group: