Add per-request retraction count (#11177)

python/sglang/srt/metrics/collector.py

@@ -811,6 +811,34 @@ class TokenizerMetricsCollector:
             buckets=bucket_e2e_request_latency,
         )
 
+        # Retraction count histogram
+        self.num_retractions = Histogram(
+            name="sglang:num_retractions",
+            documentation="Histogram of retraction counts per request.",
+            labelnames=labels.keys(),
+            buckets=[
+                0,
+                1,
+                2,
+                3,
+                4,
+                5,
+                6,
+                7,
+                8,
+                9,
+                10,
+                15,
+                20,
+                25,
+                30,
+                40,
+                50,
+                75,
+                100,
+            ],
+        )
+
     def observe_one_finished_request(
         self,
         labels: Dict[str, str],
@@ -819,6 +847,7 @@ class TokenizerMetricsCollector:
         cached_tokens: int,
         e2e_latency: float,
         has_grammar: bool,
+        retraction_count: int,
     ):
         self.prompt_tokens_total.labels(**labels).inc(prompt_tokens)
         self.generation_tokens_total.labels(**labels).inc(generation_tokens)
@@ -833,6 +862,7 @@ class TokenizerMetricsCollector:
             self.generation_tokens_histogram.labels(**labels).observe(
                 float(generation_tokens)
             )
+        self.num_retractions.labels(**labels).observe(retraction_count)
 
     def observe_time_to_first_token(self, labels: Dict[str, str], value: float):
         self.histogram_time_to_first_token.labels(**labels).observe(value)