diff --git a/pyproject.toml b/pyproject.toml
index 919c1778987..62433a3f552 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -49,7 +49,8 @@ dependencies = [
   "pathspec",
   "pip-tools>=6.6.2",
   "pip-requirements-parser>=31.2.0",
-  "prometheus-client>=0.10.0",
+  # Lowest version of prometheus_client that supports 3.10
+  "prometheus-client>=0.12.0",
   "psutil",
   "pynvml<12",
   "python-dateutil",
diff --git a/src/bentoml/_internal/server/metrics/prometheus.py b/src/bentoml/_internal/server/metrics/prometheus.py
index 9583e2d87fa..b9295b0a441 100644
--- a/src/bentoml/_internal/server/metrics/prometheus.py
+++ b/src/bentoml/_internal/server/metrics/prometheus.py
@@ -96,9 +96,25 @@ def start_http_server(self, port: int, addr: str = "") -> None:
             registry=self.registry,
         )
 
+    start_wsgi_server = start_http_server
+
+    def write_to_textfile(self, path: str) -> None:
+        """
+        Write metrics to given path. This is intended to be used with
+        the Node expoerter textfile collector.
+
+        Args:
+            path: path to write the metrics to. This file must end
+                  with '.prom' for the textfile collector to process it.
+        """
+        self.prometheus_client.write_to_textfile(path, registry=self.registry)
+
     def make_wsgi_app(self) -> ext.WSGIApp:
         return self.prometheus_client.make_wsgi_app(registry=self.registry)  # type: ignore (unfinished prometheus types)
 
+    def make_asgi_app(self) -> ext.ASGIApp:
+        return self.prometheus_client.make_asgi_app(registry=self.registry)  # type: ignore (unfinished prometheus types)
+
     def generate_latest(self):
         if self.multiproc:
             registry = self.prometheus_client.CollectorRegistry()
diff --git a/src/bentoml/metrics.py b/src/bentoml/metrics.py
index 2c5401222bf..cbed5b723d9 100644
--- a/src/bentoml/metrics.py
+++ b/src/bentoml/metrics.py
@@ -20,23 +20,26 @@
 # This sets of functions are implemented in the PrometheusClient class
 _INTERNAL_IMPL = [
     "start_http_server",
+    "start_wsgi_server",
     "make_wsgi_app",
+    "make_asgi_app",
     "generate_latest",
     "text_string_to_metric_families",
+    "write_to_textfile",
 ]
 _NOT_IMPLEMENTED = [
     "delete_from_gateway",
     "instance_ip_grouping_key",
-    "make_asgi_app",
     "push_to_gateway",
     "pushadd_to_gateway",
-    "start_wsgi_server",
-    "write_to_textfile",
 ]
 _NOT_SUPPORTED = [
     "GC_COLLECTOR",
+    "GCCollector",
     "PLATFORM_COLLECTOR",
+    "PlatformCollector",
     "PROCESS_COLLECTOR",
+    "ProcessCollector",
     "REGISTRY",
 ] + _NOT_IMPLEMENTED
 _PROPERTY = ["CONTENT_TYPE_LATEST"]
@@ -45,8 +48,18 @@
 def __dir__() -> list[str]:
     # This is for IPython and IDE autocompletion.
     metrics_client = BentoMLContainer.metrics_client.get()
-    _dir: set[str] = set(dir(metrics_client.prometheus_client)) - set(_NOT_SUPPORTED)
-    return list(_dir)
+    return list(set(dir(metrics_client.prometheus_client)) - set(_NOT_SUPPORTED))
+
+
+def __getattr__(item: t.Any):
+    # This is the entrypoint for all bentoml.metrics.*
+    if item in _PROPERTY:
+        logger.warning(
+            "'%s' is a '@property', which means there is no lazy loading. See https://docs.bentoml.org/en/latest/guides/metrics.html.",
+            item,
+        )
+        return getattr(_LazyMetric(item), item)
+    return _LazyMetric(item)
 
 
 class _LazyMetric:
@@ -111,13 +124,3 @@ def _load_proxy(
             proxy = proxy(*self._args, **self._kwargs)
         self._initialized = True
         return proxy
-
-
-def __getattr__(item: t.Any):
-    if item in _PROPERTY:
-        logger.warning(
-            "'%s' is a '@property', which means there is no lazy loading. See https://docs.bentoml.org/en/latest/guides/metrics.html.",
-            item,
-        )
-        return getattr(_LazyMetric(item), item)
-    return _LazyMetric(item)
diff --git a/src/bentoml/start.py b/src/bentoml/start.py
index 2f3c1005017..63f72c855e2 100644
--- a/src/bentoml/start.py
+++ b/src/bentoml/start.py
@@ -42,7 +42,7 @@ def start_runner_server(
     from ._internal.utils.circus import create_standalone_arbiter
     from ._internal.utils.analytics import track_serve
 
-    ensure_prometheus_dir()
+    prometheus_dir = ensure_prometheus_dir()
     working_dir = os.path.realpath(os.path.expanduser(working_dir))
     svc = load(bento_identifier, working_dir=working_dir, standalone_load=True)
 
@@ -83,6 +83,8 @@ def start_runner_server(
                             "--no-access-log",
                             "--worker-id",
                             "$(circus.wid)",
+                            "--prometheus-dir",
+                            prometheus_dir,
                         ],
                         copy_env=True,
                         stop_children=True,
diff --git a/src/bentoml_cli/worker/grpc_prometheus_server.py b/src/bentoml_cli/worker/grpc_prometheus_server.py
index 47a1d8a4e62..ab0286272e2 100644
--- a/src/bentoml_cli/worker/grpc_prometheus_server.py
+++ b/src/bentoml_cli/worker/grpc_prometheus_server.py
@@ -68,10 +68,11 @@ def main(fd: int, backlog: int, prometheus_dir: str | None):
     configure_server_logging()
 
     BentoMLContainer.development_mode.set(False)
-    metrics_client = BentoMLContainer.metrics_client.get()
     if prometheus_dir is not None:
         BentoMLContainer.prometheus_multiproc_dir.set(prometheus_dir)
 
+    metrics_client = BentoMLContainer.metrics_client.get()
+
     # create a ASGI app that wraps around the default HTTP prometheus server.
     prom_app = Starlette(
         debug=get_debug_mode(), middleware=[Middleware(GenerateLatestMiddleware)]