[metrics] Standardize metrics.Count API to prometheus counter (#14498)

python/ray/serve/backend_worker.py

@@ -190,7 +190,7 @@ class RayServeReplica:
 
         self.num_ongoing_requests = 0
 
-        self.request_counter = metrics.Count(
+        self.request_counter = metrics.Counter(
             "serve_backend_request_counter",
             description=("The number of queries that have been "
                          "processed in this replica."),
@@ -321,7 +321,7 @@ class RayServeReplica:
         try:
             result = await method_to_call(arg)
             result = await self.ensure_serializable_response(result)
-            self.request_counter.record(1)
+            self.request_counter.inc()
         except Exception as e:
             import os
             if "RAY_PDB" in os.environ:
@@ -355,7 +355,7 @@ class RayServeReplica:
                     "Please only send the same type of requests in batching "
                     "mode.")
 
-            self.request_counter.record(batch_size)
+            self.request_counter.inc(batch_size)
 
             call_method = sync_to_async(call_methods.pop())
             result_list = await call_method(args)

python/ray/serve/examples/doc/snippet_custom_metric.py

@@ -10,7 +10,7 @@ client = serve.start()
 
 class MyBackendClass:
     def __init__(self):
-        self.my_counter = metrics.Count(
+        self.my_counter = metrics.Counter(
             "my_counter",
             description=("The number of excellent requests to this backend."),
             tag_keys=("backend", ))
@@ -20,7 +20,7 @@ class MyBackendClass:
 
     def __call__(self, request):
         if "excellent" in request.query_params:
-            self.my_counter.record(1)
+            self.my_counter.inc()
 
 
 client.create_backend("my_backend", MyBackendClass)

python/ray/serve/handle.py

@@ -61,7 +61,7 @@ class RayServeHandle:
         self.handle_options = handle_options or HandleOptions()
         self.handle_tag = f"{self.endpoint_name}#{get_random_letters()}"
 
-        self.request_counter = metrics.Count(
+        self.request_counter = metrics.Counter(
             "serve_handle_request_counter",
             description=("The number of handle.remote() calls that have been "
                          "made on this handle."),
@@ -115,7 +115,7 @@ class RayServeHandle:
             ``**kwargs``: All keyword arguments will be available in
                 ``request.query_params``.
         """
-        self.request_counter.record(1)
+        self.request_counter.inc()
         return await self.router._remote(
             self.endpoint_name, self.handle_options, request_data, kwargs)
 
@@ -148,7 +148,7 @@ class RayServeSyncHandle(RayServeHandle):
             ``**kwargs``: All keyword arguments will be available in
                 ``request.args``.
         """
-        self.request_counter.record(1)
+        self.request_counter.inc()
         coro = self.router._remote(self.endpoint_name, self.handle_options,
                                    request_data, kwargs)
         future: concurrent.futures.Future = asyncio.run_coroutine_threadsafe(

python/ray/serve/http_proxy.py

@@ -106,7 +106,7 @@ class HTTPProxy:
             LongPollKey.ROUTE_TABLE: self._update_route_table,
         })
 
-        self.request_counter = metrics.Count(
+        self.request_counter = metrics.Counter(
             "serve_num_http_requests",
             description="The number of HTTP requests processed.",
             tag_keys=("route", ))
@@ -156,7 +156,7 @@ class HTTPProxy:
         assert scope["type"] == "http"
         current_path = scope["path"]
 
-        self.request_counter.record(1, tags={"route": current_path})
+        self.request_counter.inc(tags={"route": current_path})
 
         await self.router(scope, receive, send)
 

python/ray/serve/router.py

@@ -191,7 +191,7 @@ class Router:
         self._pending_endpoints: Dict[str, asyncio.Future] = dict()
 
         # -- Metrics Registration -- #
-        self.num_router_requests = metrics.Count(
+        self.num_router_requests = metrics.Counter(
             "serve_num_router_requests",
             description="The number of requests processed by the router.",
             tag_keys=("endpoint", ))
@@ -286,7 +286,7 @@ class Router:
             (await self._get_or_create_replica_set(backend)
              .assign_replica(query))
 
-        self.num_router_requests.record(1, tags={"endpoint": endpoint})
+        self.num_router_requests.inc(tags={"endpoint": endpoint})
 
         return result_ref
 

python/ray/tests/test_metrics_agent.py

@@ -30,15 +30,15 @@ def _setup_cluster_for_test(ray_start_cluster):
 
     # Generate a metric in the driver.
     counter = Count("test_driver_counter", description="desc")
-    counter.record(1)
+    counter.inc()
 
     # Generate some metrics from actor & tasks.
     @ray.remote
     def f():
         counter = Count("test_counter", description="desc")
-        counter.record(1)
+        counter.inc()
         counter = ray.get(ray.put(counter))  # Test serialization.
-        counter.record(1)
+        counter.inc()
         ray.get(worker_should_exit.wait.remote())
 
     @ray.remote
@@ -211,6 +211,11 @@ def test_basic_custom_metrics(metric_mock):
     # Make sure each of metric works as expected.
     # -- Count --
     count = Count("count", tag_keys=("a", ))
+    with pytest.raises(TypeError):
+        count.inc("hi")
+    with pytest.raises(ValueError):
+        count.inc(0)
+        count.inc(-1)
     count._metric = metric_mock
     count.record(1, {"a": "1"})
     metric_mock.record.assert_called_with(1, tags={"a": "1"})

python/ray/util/metrics.py

@@ -69,7 +69,7 @@ class Metric:
         self._default_tags = default_tags
         return self
 
-    def record(self, value: float, tags: dict = None) -> None:
+    def record(self, value: float, tags: Dict[str, str] = None) -> None:
         """Record the metric point of the metric.
 
         Tags passed in will take precedence over the metric's default tags.
@@ -125,10 +125,11 @@ class Metric:
         }
 
 
-class Count(Metric):
-    """The count of the number of metric points.
+class Counter(Metric):
+    """A cumulative metric that is monotonically increasing.
 
-    This is corresponding to Prometheus' Count metric.
+    This corresponds to Prometheus' counter metric:
+    https://prometheus.io/docs/concepts/metric_types/#counter
 
     Args:
         name(str): Name of the metric.
@@ -149,6 +150,43 @@ class Count(Metric):
         serialized_data = (self._name, self._description, self._tag_keys)
         return deserializer, serialized_data
 
+    def inc(self, value: float = 1.0, tags: Dict[str, str] = None):
+        """Increment the counter by `value` (defaults to 1).
+
+        Args:
+            value(int, float): Value to increment the counter by (default=1).
+            tags(Dict[str, str]): Tags to set or override for this counter.
+        """
+        if not isinstance(value, (int, float)):
+            raise TypeError(f"value must be int or float, got {type(value)}.")
+        if value <= 0:
+            raise ValueError(f"value must be >0, got {value}")
+
+        self.record(value, tags)
+
+
+class Count(Counter):
+    """The count of the number of metric points.
+
+    This corresponds to Prometheus' 'Count' metric.
+
+    This class is DEPRECATED, please use ray.util.metrics.Counter instead.
+
+    Args:
+        name(str): Name of the metric.
+        description(str): Description of the metric.
+        tag_keys(tuple): Tag keys of the metric.
+    """
+
+    def __init__(self,
+                 name: str,
+                 description: str = "",
+                 tag_keys: Optional[Tuple[str]] = None):
+        logger.warning(
+            "`metrics.Count` has been renamed to `metrics.Counter`. "
+            "`metrics.Count` will be removed in a future release.")
+        super().__init__(name, description, tag_keys)
+
 
 class Histogram(Metric):
     """Histogram distribution of metric points.