Fix buckets for time based historgram metrics

sgl-project · Nov 8, 2024 · 588f2c3 · 588f2c3
1 parent 3fa9361
commit 588f2c3
Showing 1 changed file with 51 additions and 3 deletions.
diff --git a/python/sglang/srt/metrics/metrics_collector.py b/python/sglang/srt/metrics/metrics_collector.py
@@ -213,19 +213,67 @@ def __init__(self, labelnames: List[str], max_model_len):
             name="sglang:e2e_request_latency_seconds",
             documentation="Histogram of End-to-end request latency in seconds",
             labelnames=labelnames,
-            buckets=build_1_2_5_buckets(max_model_len),
+            buckets=[
+                0.3,
+                0.5,
+                0.8,
+                1.0,
+                1.5,
+                2.0,
+                2.5,
+                5.0,
+                10.0,
+                15.0,
+                20.0,
+                30.0,
+                40.0,
+                50.0,
+                60.0,
+            ],
         )
         self.histogram_time_waiting_requests = Histogram(
             name="sglang:waiting_request_latency_seconds",
             documentation="Histogram of request waiting time in seconds",
             labelnames=labelnames,
-            buckets=build_1_2_5_buckets(max_model_len),
+            buckets=[
+                0.3,
+                0.5,
+                0.8,
+                1.0,
+                1.5,
+                2.0,
+                2.5,
+                5.0,
+                10.0,
+                15.0,
+                20.0,
+                30.0,
+                40.0,
+                50.0,
+                60.0,
+            ],
         )
         self.histogram_time_decode_requests = Histogram(
             name="sglang:decode_request_latency_seconds",
             documentation="Histogram of request decoding time in seconds",
             labelnames=labelnames,
-            buckets=build_1_2_5_buckets(max_model_len),
+            buckets=[
+                0.3,
+                0.5,
+                0.8,
+                1.0,
+                1.5,
+                2.0,
+                2.5,
+                5.0,
+                10.0,
+                15.0,
+                20.0,
+                30.0,
+                40.0,
+                50.0,
+                60.0,
+            ],
         )