fix: Cannot define custom duration histogram buckets via `@bentoml.se…

…rvice(metrics=...)` (#4895)

Fixes #4859

Signed-off-by: Frost Ming <me@frostming.com>

src/_bentoml_sdk/method.py

@@ -52,7 +52,7 @@ class APIMethod(t.Generic[P, R]):
     batch_dim: tuple[int, int] = attrs.field(
         default=(0, 0), converter=lambda x: (x, x) if not isinstance(x, tuple) else x
     )
-    max_batch_size: int = 100
+    max_batch_size: int = attrs.field(default=100, validator=attrs.validators.gt(1))
     max_latency_ms: int = 60000
     is_stream: bool = attrs.field(init=False)
     doc: str | None = attrs.field(init=False)

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

@@ -150,6 +150,9 @@ def __init__(
     def _setup(
         self,
         metrics_client: "PrometheusClient" = Provide[BentoMLContainer.metrics_client],
+        duration_buckets: tuple[float, ...] = Provide[
+            BentoMLContainer.duration_buckets
+        ],
     ):
         self.metrics_client = metrics_client
 
@@ -164,6 +167,7 @@ def _setup(
                 "http_response_code",
                 "runner_name",
             ],
+            buckets=duration_buckets,
         )
         self.metrics_request_total = metrics_client.Counter(
             namespace=self.namespace,