[PD metrics] Fix some uncompleted PD related metrics (#8627)

python/sglang/srt/managers/scheduler.py

@@ -1513,6 +1513,20 @@ class Scheduler(
             self.stats.gen_throughput = 0
             self.stats.num_queue_reqs = len(self.waiting_queue)
             self.stats.num_grammar_queue_reqs = len(self.grammar_queue)
+            if self.disaggregation_mode == DisaggregationMode.PREFILL:
+                self.stats.num_prefill_prealloc_queue_reqs = len(
+                    self.disagg_prefill_bootstrap_queue.queue
+                )
+                self.stats.num_prefill_inflight_queue_reqs = len(
+                    self.disagg_prefill_inflight_queue
+                )
+            if self.disaggregation_mode == DisaggregationMode.DECODE:
+                self.stats.num_decode_prealloc_queue_reqs = len(
+                    self.disagg_decode_prealloc_queue.queue
+                )
+                self.stats.num_decode_transfer_queue_reqs = len(
+                    self.disagg_decode_transfer_queue.queue
+                )
             self.metrics_collector.log_stats(self.stats)
         self._publish_kv_events()
 

python/sglang/srt/managers/scheduler_metrics_mixin.py

@@ -230,7 +230,7 @@ class SchedulerMetricsMixin:
             self.stats.num_grammar_queue_reqs = len(self.grammar_queue)
             self.stats.spec_accept_length = spec_accept_length
             self.stats.total_retracted_reqs = self.total_retracted_reqs
-            self.metrics_collector.log_stats(self.stats)
+            self.stats.avg_request_queue_latency = 0.0
             if self.disaggregation_mode == DisaggregationMode.DECODE:
                 self.stats.num_decode_prealloc_queue_reqs = len(
                     self.disagg_decode_prealloc_queue.queue
@@ -238,6 +238,7 @@ class SchedulerMetricsMixin:
                 self.stats.num_decode_transfer_queue_reqs = len(
                     self.disagg_decode_transfer_queue.queue
                 )
+            self.metrics_collector.log_stats(self.stats)
             self._emit_kv_metrics()
         self._publish_kv_events()
 

python/sglang/srt/metrics/collector.py

@@ -539,6 +539,7 @@ class SchedulerMetricsCollector:
             self.num_running_reqs_offline_batch, stats.num_running_reqs_offline_batch
         )
         self._log_gauge(self.cache_hit_rate, stats.cache_hit_rate)
+        self._log_gauge(self.avg_request_queue_latency, stats.avg_request_queue_latency)
 
         # Speculative decoding
         self._log_gauge(self.spec_accept_length, stats.spec_accept_length)