feat(metrics): init zero for request count of apis (#5186)

* feat(metrics): init zero for request count of apis

* fix: use route instead of api name

src/_bentoml_impl/worker/service.py

@@ -186,6 +186,7 @@ def main(
     BentoMLContainer.development_mode.set(development_mode)
 
     server_context.service_name = service.name
+    server_context.service_routes = [api.route for api in service.apis.values()]
     if service.bento is None:
         server_context.bento_name = service.name
         server_context.bento_version = "not available"

src/bentoml/_internal/context.py

@@ -225,6 +225,7 @@ class _ComponentContext:
     service_type: str | None = None
     service_name: str | None = None
     worker_index: int | None = None
+    service_routes: list[str] = attr.field(factory=list)
 
     @property
     def component_type(self) -> str | None:

src/bentoml/_internal/server/http/instruments.py

@@ -207,6 +207,23 @@ def _setup(
             documentation="Total number of bytes sent to websocket",
             labelnames=["endpoint", "service_name", "service_version", "runner_name"],
         )
+        # place holder metrics to initialize endpoints with 0 to facilitate rate calculation
+        # otherwise, rate() can be 0 if the endpoint is hit for the first time
+        for endpoint in server_context.service_routes:
+            self.metrics_request_total.labels(
+                endpoint=endpoint,
+                service_name=server_context.bento_name,
+                service_version=server_context.bento_version,
+                http_response_code=200,
+                runner_name=server_context.service_name,
+            )
+            self.metrics_request_total.labels(
+                endpoint=endpoint,
+                service_name=server_context.bento_name,
+                service_version=server_context.bento_version,
+                http_response_code=500,
+                runner_name=server_context.service_name,
+            )
         self._is_setup = True
 
     async def __call__(