[Serve] Add route tags with custom metrics

python/ray/serve/metrics.py

@@ -1,10 +1,12 @@
 from ray.util import metrics
-from typing import Tuple, Optional, Dict, List
+from typing import Tuple, Optional, Dict, List, Union
 from ray.serve import context
+import ray
 
 DEPLOYMENT_TAG = "deployment"
 REPLICA_TAG = "replica"
 APPLICATION_TAG = "application"
+ROUTE_TAG = "route"
 
 
 def _add_serve_metric_tags(tag_keys: Optional[Tuple[str]] = None) -> Tuple[str]:
@@ -52,6 +54,14 @@ def _add_serve_metric_default_tags(default_tags: Dict[str, str]):
     return default_tags
 
 
+def _add_serve_context_tag_values(tag_keys: Tuple, tags: Dict[str, str]):
+    """Add serve context tag values to the metric tags"""
+
+    _request_context = ray.serve.context._serve_request_context.get()
+    if ROUTE_TAG in tag_keys and ROUTE_TAG not in tags:
+        tags[ROUTE_TAG] = _request_context.route
+
+
 class Counter(metrics.Counter):
     def __init__(
         self, name: str, description: str = "", tag_keys: Optional[Tuple[str]] = None
@@ -67,6 +77,13 @@ def __init__(
     def set_default_tags(self, default_tags: Dict[str, str]):
         super().set_default_tags(_add_serve_metric_default_tags(default_tags))
 
+    def inc(self, value: Union[int, float] = 1.0, tags: Dict[str, str] = None):
+        """Increment the counter by the given value, add serve context
+        tag values to the tags
+        """
+        _add_serve_context_tag_values(self._tag_keys, tags)
+        super().inc(value, tags)
+
 
 class Gauge(metrics.Gauge):
     def __init__(
@@ -83,6 +100,13 @@ def __init__(
     def set_default_tags(self, default_tags: Dict[str, str]):
         super().set_default_tags(_add_serve_metric_default_tags(default_tags))
 
+    def set(self, value: Union[int, float], tags: Dict[str, str] = None):
+        """Set the gauge to the given value, add serve context
+        tag values to the tags
+        """
+        _add_serve_context_tag_values(self._tag_keys, tags)
+        super().set(value, tags)
+
 
 class Histogram(metrics.Histogram):
     def __init__(
@@ -102,3 +126,10 @@ def __init__(
 
     def set_default_tags(self, default_tags: Dict[str, str]):
         super().set_default_tags(_add_serve_metric_default_tags(default_tags))
+
+    def observe(self, value: Union[int, float], tags: Dict[str, str] = None):
+        """Observe the given value, add serve context
+        tag values to the tags
+        """
+        _add_serve_context_tag_values(self._tag_keys, tags)
+        super().observe(value, tags)

python/ray/serve/tests/test_metrics.py

@@ -342,6 +342,10 @@ def _generate_metrics_summary(self, metrics):
             ]
         return metrics_summary_route, metrics_summary_app
 
+    def verify_metrics(self, metric, expected_output):
+        for key in expected_output:
+            assert metric[key] == expected_output[key]
+
     def test_request_context_pass_for_http_proxy(self, serve_start_shutdown):
         """Test HTTP proxy passing request context"""
 
@@ -411,7 +415,7 @@ def h():
             "serve_deployment_processing_latency_ms_sum",
         ]:
             metrics_route, metrics_app_name = self._generate_metrics_summary(
-                get_metric_dictionaries("serve_handle_request_counter")
+                get_metric_dictionaries(metric_name)
             )
             assert metrics_route["app1_f"] == {"/app1"}
             assert metrics_route["app2_g"] == {"/app2"}
@@ -488,6 +492,7 @@ def __init__(self):
                     tag_keys=(
                         "my_static_tag",
                         "my_runtime_tag",
+                        "route",
                     ),
                 )
                 self.counter.set_default_tags({"my_static_tag": "static_value"})
@@ -498,6 +503,7 @@ def __init__(self):
                     tag_keys=(
                         "my_static_tag",
                         "my_runtime_tag",
+                        "route",
                     ),
                 )
                 self.histogram.set_default_tags({"my_static_tag": "static_value"})
@@ -507,6 +513,7 @@ def __init__(self):
                     tag_keys=(
                         "my_static_tag",
                         "my_runtime_tag",
+                        "route",
                     ),
                 )
                 self.gauge.set_default_tags({"my_static_tag": "static_value"})
@@ -527,29 +534,42 @@ def __call__(self):
             lambda: len(get_metric_dictionaries("my_gauge")) == 1,
             timeout=20,
         )
+
         counter_metrics = get_metric_dictionaries("my_counter")
         assert len(counter_metrics) == 1
-        counter_metrics[0]["my_static_tag"] == "static_value"
-        counter_metrics[0]["my_runtime_tag"] == "100"
-        counter_metrics[0]["replica"] == replica_tag
-        counter_metrics[0]["deployment"] == deployment_name
-        counter_metrics[0]["application"] == "app"
-
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "100",
+            "replica": replica_tag,
+            "deployment": deployment_name,
+            "application": "app",
+            "route": "/app",
+        }
+        self.verify_metrics(counter_metrics[0], expected_metrics)
+
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "300",
+            "replica": replica_tag,
+            "deployment": deployment_name,
+            "application": "app",
+            "route": "/app",
+        }
         gauge_metrics = get_metric_dictionaries("my_gauge")
         assert len(counter_metrics) == 1
-        gauge_metrics[0]["my_static_tag"] == "static_value"
-        gauge_metrics[0]["my_runtime_tag"] == "300"
-        gauge_metrics[0]["replica"] == replica_tag
-        gauge_metrics[0]["deployment"] == deployment_name
-        gauge_metrics[0]["application"] == "app"
-
+        self.verify_metrics(gauge_metrics[0], expected_metrics)
+
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "200",
+            "replica": replica_tag,
+            "deployment": deployment_name,
+            "application": "app",
+            "route": "/app",
+        }
         histogram_metrics = get_metric_dictionaries("my_histogram_sum")
         assert len(histogram_metrics) == 1
-        histogram_metrics[0]["my_static_tag"] == "static_value"
-        histogram_metrics[0]["my_runtime_tag"] == "200"
-        histogram_metrics[0]["replica"] == replica_tag
-        histogram_metrics[0]["deployment"] == deployment_name
-        gauge_metrics[0]["application"] == "app"
+        self.verify_metrics(histogram_metrics[0], expected_metrics)
 
     @pytest.mark.parametrize("use_actor", [False, True])
     def test_serve_metrics_outside_serve(self, use_actor, serve_start_shutdown):
@@ -650,20 +670,30 @@ async def __call__(self):
             lambda: len(get_metric_dictionaries("my_gauge")) == 1,
             timeout=20,
         )
+
         counter_metrics = get_metric_dictionaries("my_counter")
         assert len(counter_metrics) == 1
-        counter_metrics[0]["my_static_tag"] == "static_value"
-        counter_metrics[0]["my_runtime_tag"] == "100"
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "100",
+        }
+        self.verify_metrics(counter_metrics[0], expected_metrics)
 
         gauge_metrics = get_metric_dictionaries("my_gauge")
         assert len(counter_metrics) == 1
-        gauge_metrics[0]["my_static_tag"] == "static_value"
-        gauge_metrics[0]["my_runtime_tag"] == "300"
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "300",
+        }
+        self.verify_metrics(gauge_metrics[0], expected_metrics)
 
         histogram_metrics = get_metric_dictionaries("my_histogram_sum")
         assert len(histogram_metrics) == 1
-        histogram_metrics[0]["my_static_tag"] == "static_value"
-        histogram_metrics[0]["my_runtime_tag"] == "200"
+        expected_metrics = {
+            "my_static_tag": "static_value",
+            "my_runtime_tag": "200",
+        }
+        self.verify_metrics(histogram_metrics[0], expected_metrics)
 
 
 def test_actor_summary(serve_instance):