feat: DIS-678 kvbm modularity: standalone metrics endpoint (#3433)

Signed-off-by: richardhuo-nv <rihuo@nvidia.com>

deploy/metrics/grafana_dashboards/grafana-kvbm-dashboard.json

@@ -118,8 +118,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_matched_tokens{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_matched_tokens",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -227,8 +227,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_offload_requests{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_offload_requests",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -323,8 +323,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_offload_blocks_d2h{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_offload_blocks_d2h",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -336,102 +336,6 @@
       "title": "Offload Blocks - Device to Host",
       "type": "timeseries"
     },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "P1809F7CD0C75ACF3"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "color": {
-            "mode": "palette-classic"
-          },
-          "custom": {
-            "axisBorderShow": false,
-            "axisCenteredZero": false,
-            "axisColorMode": "text",
-            "axisLabel": "",
-            "axisPlacement": "auto",
-            "barAlignment": 0,
-            "barWidthFactor": 0.6,
-            "drawStyle": "line",
-            "fillOpacity": 0,
-            "gradientMode": "none",
-            "hideFrom": {
-              "legend": false,
-              "tooltip": false,
-              "viz": false
-            },
-            "insertNulls": false,
-            "lineInterpolation": "linear",
-            "lineWidth": 1,
-            "pointSize": 5,
-            "scaleDistribution": {
-              "type": "linear"
-            },
-            "showPoints": "auto",
-            "spanNulls": false,
-            "stacking": {
-              "group": "A",
-              "mode": "none"
-            },
-            "thresholdsStyle": {
-              "mode": "off"
-            }
-          },
-          "mappings": [],
-          "thresholds": {
-            "mode": "absolute",
-            "steps": [
-              {
-                "color": "green"
-              },
-              {
-                "color": "red",
-                "value": 80
-              }
-            ]
-          }
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 8,
-        "w": 12,
-        "x": 0,
-        "y": 18
-      },
-      "id": 1,
-      "options": {
-        "legend": {
-          "calcs": [],
-          "displayMode": "list",
-          "placement": "bottom",
-          "showLegend": true
-        },
-        "tooltip": {
-          "hideZeros": false,
-          "mode": "single",
-          "sort": "none"
-        }
-      },
-      "pluginVersion": "12.0.1",
-      "targets": [
-        {
-          "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_save_kv_layer_requests{dynamo_namespace=\"kvbm_connector_worker\"}",
-          "fullMetaSearch": false,
-          "includeNullMetadata": true,
-          "legendFormat": "__auto",
-          "range": true,
-          "refId": "A",
-          "useBackend": false
-        }
-      ],
-      "title": "Save KV Layer Requests",
-      "type": "timeseries"
-    },
     {
       "collapsed": false,
       "gridPos": {
@@ -528,8 +432,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_onboard_requests{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_onboard_requests",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -624,8 +528,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_onboard_blocks_h2d{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_onboard_blocks_h2d",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -720,8 +624,8 @@
       "targets": [
         {
           "disableTextWrap": false,
-          "editorMode": "builder",
-          "expr": "dynamo_component_onboard_blocks_d2d{dynamo_namespace=\"kvbm_connector_leader\"}",
+          "editorMode": "code",
+          "expr": "kvbm_onboard_blocks_d2d",
           "fullMetaSearch": false,
           "includeNullMetadata": true,
           "legendFormat": "__auto",
@@ -750,4 +654,4 @@
   "title": "KVBM Dashboard",
   "uid": "3f679257-70a5-402c-92b4-05382337b548",
   "version": 7
-}
+}
\ No newline at end of file

deploy/metrics/prometheus.yml

@@ -59,16 +59,10 @@ scrape_configs:
       - targets: ['host.docker.internal:9091']  # metrics aggregation service on host
 
   # KVBM leader related metrics
-  - job_name: 'kvbm-leader-metrics'
+  - job_name: 'kvbm-metrics'
     scrape_interval: 2s
     static_configs:
-      - targets: ['host.docker.internal:6882']
-
-  # KVBM worker related metrics
-  - job_name: 'kvbm-worker-metrics'
-    scrape_interval: 2s
-    static_configs:
-      - targets: ['host.docker.internal:6881']
+      - targets: ['host.docker.internal:6880']
 
   # Uncomment to see its own Prometheus metrics
   # - job_name: 'prometheus'

docs/guides/run_kvbm_in_trtllm.md

@@ -109,18 +109,51 @@ Follow below steps to enable metrics collection and view via Grafana dashboard:
 # Start the basic services (etcd & natsd), along with Prometheus and Grafana
 docker compose -f deploy/docker-compose.yml --profile metrics up -d
 
-# set env var DYN_SYSTEM_ENABLED to true, DYN_SYSTEM_PORT to 6880, DYN_KVBM_SLEEP to 5, when launch via dynamo
-# NOTE: Make sure port 6881 (for KVBM worker metrics) and port 6882 (for KVBM leader metrics) are available.
-# NOTE: DYN_KVBM_SLEEP is needed to avoid metrics port conflict between KVBM leader and worker
-DYN_SYSTEM_ENABLED=true DYN_SYSTEM_PORT=6880 DYN_KVBM_SLEEP=5 \
+# set env var DYN_KVBM_METRICS to true, when launch via dynamo
+# Optionally set DYN_KVBM_METRICS_PORT to choose the /metrics port (default: 6880).
+DYN_KVBM_METRICS=true \
 python3 -m dynamo.trtllm \
   --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --extra-engine-args /tmp/kvbm_llm_api_config.yaml &
 
 # optional if firewall blocks KVBM metrics ports to send prometheus metrics
-sudo ufw allow 6881/tcp
-sudo ufw allow 6882/tcp
+sudo ufw allow 6880/tcp
 ```
 
 View grafana metrics via http://localhost:3001 (default login: dynamo/dynamo) and look for KVBM Dashboard
+
+## Benchmark KVBM
+
+Once the model is loaded ready, follow below steps to use LMBenchmark to benchmark KVBM performance:
+```bash
+git clone https://github.com/LMCache/LMBenchmark.git
+
+# show case of running the synthetic multi-turn chat dataset.
+# we are passing model, endpoint, output file prefix and qps to the sh script.
+cd LMBenchmark/synthetic-multi-round-qa
+./long_input_short_output_run.sh \
+    "deepseek-ai/DeepSeek-R1-Distill-Llama-8B" \
+    "http://localhost:8000" \
+    "benchmark_kvbm" \
+    1
+
+# Average TTFT and other perf numbers would be in the output from above cmd
+```
+More details about how to use LMBenchmark could be found [here](https://github.com/LMCache/LMBenchmark).
+
+`NOTE`: if metrics are enabled as mentioned in the above section, you can observe KV offloading, and KV onboarding in the grafana dashboard.
+
+To compare, you can remove the `kv_connector_config` section from the LLM API config and run `trtllm-serve` with the updated config as the baseline.
+```bash
+cat > "/tmp/llm_api_config.yaml" <<EOF
+backend: pytorch
+cuda_graph_config: null
+kv_cache_config:
+  enable_partial_reuse: false
+  free_gpu_memory_fraction: 0.80
+EOF
+
+# run trtllm-serve for the baseline for comparison
+trtllm-serve deepseek-ai/DeepSeek-R1-Distill-Llama-8B --host localhost --port 8000 --backend pytorch --extra_llm_api_options /tmp/llm_api_config.yaml &
+```

docs/guides/run_kvbm_in_vllm.md

@@ -77,23 +77,22 @@ Follow below steps to enable metrics collection and view via Grafana dashboard:
 # Start the basic services (etcd & natsd), along with Prometheus and Grafana
 docker compose -f deploy/docker-compose.yml --profile metrics up -d
 
-# set env var DYN_SYSTEM_ENABLED to true, DYN_SYSTEM_PORT to 6880, DYN_KVBM_SLEEP to 5, when launch via dynamo
-# NOTE: Make sure port 6881 (for KVBM worker metrics) and port 6882 (for KVBM leader metrics) are available.
-DYN_SYSTEM_ENABLED=true DYN_SYSTEM_PORT=6880 \
+# set env var DYN_KVBM_METRICS to true, when launch via dynamo
+# Optionally set DYN_KVBM_METRICS_PORT to choose the /metrics port (default: 6880).
+DYN_KVBM_METRICS=true \
 python -m dynamo.vllm \
     --model deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
     --connector kvbm &
 
 # optional if firewall blocks KVBM metrics ports to send prometheus metrics
-sudo ufw allow 6881/tcp
-sudo ufw allow 6882/tcp
+sudo ufw allow 6880/tcp
 ```
 
 View grafana metrics via http://localhost:3001 (default login: dynamo/dynamo) and look for KVBM Dashboard
 
 ## Benchmark KVBM
 
-Once vllm serve is ready, follow below steps to use LMBenchmark to benchmark KVBM performance:
+Once the model is loaded ready, follow below steps to use LMBenchmark to benchmark KVBM performance:
 ```bash
 git clone https://github.com/LMCache/LMBenchmark.git
 

lib/bindings/python/rust/llm/block_manager/vllm/connector/leader.rs

@@ -5,7 +5,7 @@ pub mod recorder;
 pub mod slot;
 
 use super::*;
-use dynamo_llm::block_manager::metrics_kvbm::KvbmMetrics;
+use dynamo_llm::block_manager::metrics_kvbm::{KvbmMetrics, KvbmMetricsRegistry};
 use dynamo_runtime::DistributedRuntime;
 use slot::{ConnectorSlotManager, SlotError, SlotManager, SlotState};
 
@@ -15,7 +15,6 @@ use crate::llm::block_manager::{
     VllmBlockManager, distributed::KvbmLeader as PyKvbmLeader, vllm::KvbmRequest,
     vllm::connector::leader::slot::VllmConnectorSlot,
 };
-use dynamo_runtime::metrics::prometheus_names::kvbm_connector;
 
 use dynamo_llm::block_manager::{
     BasicMetadata, DiskStorage, ImmutableBlock, PinnedStorage,
@@ -103,11 +102,11 @@ impl KvConnectorLeader {
         let drt = drt.inner().clone();
         let handle: Handle = drt.runtime().primary();
 
-        let ns = drt
-            .namespace(kvbm_connector::KVBM_CONNECTOR_LEADER)
-            .unwrap();
-
-        let kvbm_metrics = KvbmMetrics::new(&ns);
+        let kvbm_metrics = KvbmMetrics::new(
+            &KvbmMetricsRegistry::default(),
+            kvbm_metrics_endpoint_enabled(),
+            parse_kvbm_metrics_port(),
+        );
         let kvbm_metrics_clone = kvbm_metrics.clone();
 
         let slot_manager_cell = Arc::new(OnceLock::new());
@@ -615,3 +614,30 @@ impl PyKvConnectorLeader {
             .map_err(to_pyerr)
     }
 }
+
+pub fn kvbm_metrics_endpoint_enabled() -> bool {
+    std::env::var("DYN_KVBM_METRICS")
+        .map(|v| v == "1" || v.eq_ignore_ascii_case("true"))
+        .unwrap_or(false)
+}
+
+pub fn parse_kvbm_metrics_port() -> u16 {
+    match std::env::var("DYN_KVBM_METRICS_PORT") {
+        Ok(val) => match val.trim().parse::<u16>() {
+            Ok(port) => port,
+            Err(_) => {
+                tracing::warn!(
+                    "[kvbm] Invalid DYN_KVBM_METRICS_PORT='{}', falling back to 6880",
+                    val
+                );
+                6880
+            }
+        },
+        Err(_) => {
+            tracing::warn!(
+                "DYN_KVBM_METRICS_PORT not present or couldn’t be interpreted, falling back to 6880"
+            );
+            6880
+        }
+    }
+}

lib/bindings/python/rust/llm/block_manager/vllm/connector/leader/recorder.rs

@@ -100,11 +100,11 @@ impl KvConnectorLeaderRecorder {
         let drt = drt.inner().clone();
         let handle: Handle = drt.runtime().primary();
 
-        let ns = drt
-            .namespace(kvbm_connector::KVBM_CONNECTOR_LEADER)
-            .unwrap();
-
-        let kvbm_metrics = KvbmMetrics::new(&ns);
+        let kvbm_metrics = KvbmMetrics::new(
+            &KvbmMetricsRegistry::default(),
+            kvbm_metrics_endpoint_enabled(),
+            parse_kvbm_metrics_port(),
+        );
         let kvbm_metrics_clone = kvbm_metrics.clone();
 
         let token = CancellationToken::new();

lib/bindings/python/rust/llm/block_manager/vllm/connector/trtllm_leader.rs

@@ -8,10 +8,12 @@ use crate::llm::block_manager::BlockManagerBuilder;
 use crate::llm::block_manager::vllm::connector::leader::slot::{
     ConnectorSlotManager, SlotManager, SlotState,
 };
+use crate::llm::block_manager::vllm::connector::leader::{
+    kvbm_metrics_endpoint_enabled, parse_kvbm_metrics_port,
+};
 use crate::llm::block_manager::{distributed::KvbmLeader as PyKvbmLeader, vllm::KvbmRequest};
 use anyhow;
-use dynamo_llm::block_manager::metrics_kvbm::KvbmMetrics;
-use dynamo_runtime::metrics::prometheus_names::kvbm_connector;
+use dynamo_llm::block_manager::metrics_kvbm::{KvbmMetrics, KvbmMetricsRegistry};
 use std::collections::HashSet;
 use std::sync::{Arc, OnceLock};
 use tokio::runtime::Handle;
@@ -76,11 +78,12 @@ impl KvConnectorLeader {
         let drt = drt.inner().clone();
         let handle: Handle = drt.runtime().primary();
 
-        let ns = drt
-            .namespace(kvbm_connector::KVBM_CONNECTOR_LEADER)
-            .unwrap();
+        let kvbm_metrics = KvbmMetrics::new(
+            &KvbmMetricsRegistry::default(),
+            kvbm_metrics_endpoint_enabled(),
+            parse_kvbm_metrics_port(),
+        );
 
-        let kvbm_metrics = KvbmMetrics::new(&ns);
         let kvbm_metrics_clone = kvbm_metrics.clone();
 
         let slot_manager_cell = Arc::new(OnceLock::new());

lib/bindings/python/rust/llm/block_manager/vllm/connector/trtllm_worker.rs

@@ -20,10 +20,8 @@ use crate::{
 use anyhow;
 use dynamo_llm::block_manager::distributed::{KvbmWorker, KvbmWorkerConfig};
 use dynamo_llm::block_manager::layout::LayoutType;
-use dynamo_llm::block_manager::metrics_kvbm::KvbmMetrics;
 use dynamo_llm::block_manager::storage::torch::TorchTensor;
 use dynamo_runtime::DistributedRuntime;
-use dynamo_runtime::metrics::prometheus_names::kvbm_connector;
 use dynamo_runtime::utils::task::CriticalTaskExecutionHandle;
 
 pub trait Worker: Send + Sync {
@@ -71,8 +69,6 @@ pub struct KvConnectorWorker {
 
     /// cuda events created by the python side
     layer_events: Vec<u64>,
-
-    kvbm_metrics: KvbmMetrics,
 }
 
 impl KvConnectorWorker {
@@ -98,11 +94,6 @@ impl KvConnectorWorker {
             trtllm_rank
         );
 
-        let kvbm_metrics = KvbmMetrics::new(
-            &drt.namespace(kvbm_connector::KVBM_CONNECTOR_WORKER)
-                .unwrap(),
-        );
-
         Ok(Self {
             drt,
             kvbm_worker: OnceLock::new(),
@@ -116,7 +107,6 @@ impl KvConnectorWorker {
             iteration: 0,
             layers_complete: 0,
             layer_events: Vec::new(),
-            kvbm_metrics,
         })
     }
 }
@@ -236,7 +226,6 @@ impl Worker for KvConnectorWorker {
                 self.connector.enqueue_request(operation);
             }
         }
-        self.kvbm_metrics.save_kv_layer_requests.inc();
         Ok(())
     }
 

lib/bindings/python/rust/llm/block_manager/vllm/connector/worker.rs

@@ -5,7 +5,6 @@ use dynamo_llm::block_manager::connector::protocol::TransferType;
 use dynamo_llm::block_manager::connector::scheduler::{
     Scheduler, TransferSchedulerClient, WorkerSchedulerClient,
 };
-use dynamo_llm::block_manager::metrics_kvbm::KvbmMetrics;
 
 use std::collections::HashSet;
 use std::sync::{Arc, OnceLock};
@@ -16,7 +15,6 @@ use crate::{
     DistributedRuntime as PyDistributedRuntime, llm::block_manager::distributed::VllmTensor,
     to_pyerr,
 };
-use dynamo_runtime::metrics::prometheus_names::kvbm_connector;
 
 use crate::llm::block_manager::distributed::PyLayoutType;
 use anyhow;
@@ -75,8 +73,6 @@ pub struct KvConnectorWorker {
 
     /// cuda events created by the python side
     layer_events: Vec<u64>,
-
-    kvbm_metrics: KvbmMetrics,
 }
 
 impl KvConnectorWorker {
@@ -97,11 +93,6 @@ impl KvConnectorWorker {
         )?
         .detach();
 
-        let kvbm_metrics = KvbmMetrics::new(
-            &drt.namespace(kvbm_connector::KVBM_CONNECTOR_WORKER)
-                .unwrap(),
-        );
-
         tracing::info!(
             "KvConnectorWorker initialized with worker_id: {}",
             vllm_worker_id
@@ -120,7 +111,6 @@ impl KvConnectorWorker {
             layers_complete: 0,
             kv_cache_layers: Vec::new(),
             layer_events: Vec::new(),
-            kvbm_metrics,
         })
     }
 }
@@ -324,7 +314,6 @@ impl Worker for KvConnectorWorker {
                 self.connector.enqueue_request(operation);
             }
         }
-        self.kvbm_metrics.save_kv_layer_requests.inc();
         Ok(())
     }
 

lib/bindings/python/src/dynamo/llm/trtllm_integration/connector/kvbm_connector_leader.py

@@ -17,17 +17,12 @@ from dynamo.llm.trtllm_integration.rust import (
     KvConnectorLeader as RustKvConnectorLeader,
 )
 from dynamo.llm.trtllm_integration.rust import SchedulerOutput as RustSchedulerOutput
-from dynamo.llm.utils import find_and_set_available_port_from_env, maybe_sleep
 from dynamo.runtime import DistributedRuntime
 
 
 class DynamoKVBMConnectorLeader(KvCacheConnectorScheduler):
     def __init__(self, llm_args: TorchLlmArgs):
         super().__init__(llm_args)
-        # NOTE: this is needed in TRTLLM to avoid metrics port conflict with KVBM worker,
-        # since there is no startup order in TRTLLM and race condition is possible.
-        maybe_sleep()
-        find_and_set_available_port_from_env("DYN_SYSTEM_PORT")
         self.drt = DistributedRuntime.detached()
 
         mappings = self._llm_args.parallel_config.to_mapping()

lib/bindings/python/src/dynamo/llm/trtllm_integration/connector/kvbm_connector_worker.py

@@ -9,7 +9,6 @@ from tensorrt_llm.llmapi.llm_args import TorchLlmArgs
 from dynamo.llm.trtllm_integration.rust import (
     KvConnectorWorker as RustKvConnectorWorker,
 )
-from dynamo.llm.utils import find_and_set_available_port_from_env
 from dynamo.runtime import DistributedRuntime
 
 
@@ -17,7 +16,6 @@ class DynamoKVBMConnectorWorker(KvCacheConnectorWorker):
     def __init__(self, llm_args: TorchLlmArgs):
         super().__init__(llm_args)
 
-        find_and_set_available_port_from_env("DYN_SYSTEM_PORT")
         self.drt = DistributedRuntime.detached()
 
         mappings = self._llm_args.parallel_config.to_mapping()

lib/bindings/python/src/dynamo/llm/utils.py

-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-
-import os
-import socket
-import time
-
-
-def maybe_sleep():
-    """
-    Maybe sleep for the duration specified in the environment variable if it is set.
-    """
-    sleep_duration = int(os.environ.get("DYN_KVBM_SLEEP", "0"))
-    if sleep_duration > 0:
-        print(f"Sleeping {sleep_duration} seconds to avoid metrics port conflict")
-        time.sleep(sleep_duration)
-
-
-# TODO(keiven|ziqi): Auto port selection to be done in Rust
-def find_and_set_available_port_from_env(env_var="DYN_SYSTEM_PORT"):
-    """
-    Find an available port from the environment variable.
-    """
-    port = int(os.environ.get(env_var, "0"))
-    if port == 0:
-        # No port specified, let system pick
-        pass
-    while True:
-        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-        try:
-            # Port is available
-            s.bind(("127.0.0.1", port))
-            s.close()
-            os.environ[env_var] = str(port)
-            print(f"Port {port} is available, setting env var {env_var} to {port}")
-            break
-        except OSError:
-            # Port is in use, try next
-            port += 1
-            s.close()
-        except Exception as e:
-            raise RuntimeError(f"Error finding available port: {e}")

lib/bindings/python/src/dynamo/llm/vllm_integration/connector_leader.py

@@ -29,7 +29,6 @@ if TYPE_CHECKING:
 # from dynamo.llm.vllm_integration.rust import SchedulerOutput as RustSchedulerOutput
 
 from dynamo.llm import KvbmLeader
-from dynamo.llm.utils import find_and_set_available_port_from_env
 from dynamo.llm.vllm_integration.rust import KvbmRequest
 from dynamo.llm.vllm_integration.rust import KvConnectorLeader as RustKvConnectorLeader
 from dynamo.llm.vllm_integration.rust import SchedulerOutput as RustSchedulerOutput
@@ -54,7 +53,6 @@ class KvConnectorLeader:
     def __init__(self, vllm_config: "VllmConfig", engine_id: str, **kwargs):
         drt = kwargs.get("drt", None)
         if drt is None:
-            find_and_set_available_port_from_env("DYN_SYSTEM_PORT")
             self.drt = DistributedRuntime.detached()
         else:
             self.drt = drt

lib/bindings/python/src/dynamo/llm/vllm_integration/connector_worker.py

@@ -28,7 +28,6 @@ if TYPE_CHECKING:
 #     KvConnectorWorker as RustKvConnectorWorker,
 # )
 
-from dynamo.llm.utils import find_and_set_available_port_from_env
 from dynamo.llm.vllm_integration.rust import KvConnectorWorker as RustKvConnectorWorker
 from dynamo.runtime import DistributedRuntime
 
@@ -43,8 +42,6 @@ class KvConnectorWorker:
     def __init__(self, vllm_config: "VllmConfig", engine_id: str, **kwargs):
         drt = kwargs.get("drt", None)
         if drt is None:
-            # this is needed to avoid metrics port conflict with KVBM leader side DRT if metrics is enabled
-            find_and_set_available_port_from_env("DYN_SYSTEM_PORT")
             self.drt = DistributedRuntime.detached()
         else:
             self.drt = drt

lib/llm/src/block_manager/metrics_kvbm.rs

 // SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 // SPDX-License-Identifier: Apache-2.0
 
-use dynamo_runtime::metrics::MetricsRegistry;
-use prometheus::IntCounter;
+use axum::Router;
+use dynamo_runtime::metrics::prometheus_names::{
+    kvbm::{
+        MATCHED_TOKENS, OFFLOAD_BLOCKS_D2H, OFFLOAD_REQUESTS, ONBOARD_BLOCKS_D2D,
+        ONBOARD_BLOCKS_H2D, ONBOARD_REQUESTS,
+    },
+    sanitize_prometheus_name,
+};
+use prometheus::{IntCounter, Opts, Registry};
+use std::{collections::HashMap, net::SocketAddr, sync::Arc, thread};
+use tokio::{net::TcpListener, sync::Notify};
+
+use crate::http::service::{RouteDoc, metrics::router};
 
 #[derive(Clone, Debug)]
 pub struct KvbmMetrics {
@@ -21,60 +32,167 @@ pub struct KvbmMetrics {
     // number of blocks onboarded from disk to device
     pub onboard_blocks_d2d: IntCounter,
 
-    // number of save kv layer requests
-    pub save_kv_layer_requests: IntCounter,
-
     // number of matched tokens from KVBM
     pub matched_tokens: IntCounter,
+
+    shutdown_notify: Option<Arc<Notify>>,
 }
 
 impl KvbmMetrics {
-    pub fn new(mr: &dyn MetricsRegistry) -> Self {
+    /// Create raw metrics and (once per process) spawn an axum server exposing `/metrics` at metrics_port.
+    /// Non-blocking: the HTTP server runs on a background task.
+    pub fn new(mr: &KvbmMetricsRegistry, create_endpoint: bool, metrics_port: u16) -> Self {
+        // 1) register kvbm metrics
         let offload_requests = mr
-            .create_intcounter("offload_requests", "The number of offload requests", &[])
+            .create_intcounter(OFFLOAD_REQUESTS, "The number of offload requests", &[])
             .unwrap();
         let offload_blocks_d2h = mr
             .create_intcounter(
-                "offload_blocks_d2h",
+                OFFLOAD_BLOCKS_D2H,
                 "The number of offload blocks from device to host",
                 &[],
             )
             .unwrap();
         let onboard_requests = mr
-            .create_intcounter("onboard_requests", "The number of onboard requests", &[])
+            .create_intcounter(ONBOARD_REQUESTS, "The number of onboard requests", &[])
             .unwrap();
         let onboard_blocks_h2d = mr
             .create_intcounter(
-                "onboard_blocks_h2d",
+                ONBOARD_BLOCKS_H2D,
                 "The number of onboard blocks from host to device",
                 &[],
             )
             .unwrap();
         let onboard_blocks_d2d = mr
             .create_intcounter(
-                "onboard_blocks_d2d",
+                ONBOARD_BLOCKS_D2D,
                 "The number of onboard blocks from disk to device",
                 &[],
             )
             .unwrap();
-        let save_kv_layer_requests = mr
-            .create_intcounter(
-                "save_kv_layer_requests",
-                "The number of save kv layer requests",
-                &[],
-            )
-            .unwrap();
         let matched_tokens = mr
-            .create_intcounter("matched_tokens", "The number of matched tokens", &[])
+            .create_intcounter(MATCHED_TOKENS, "The number of matched tokens", &[])
             .unwrap();
+
+        // early return if no endpoint is needed
+        if !create_endpoint {
+            return Self {
+                offload_requests,
+                offload_blocks_d2h,
+                onboard_requests,
+                onboard_blocks_h2d,
+                onboard_blocks_d2d,
+                matched_tokens,
+                shutdown_notify: None,
+            };
+        }
+
+        // 2) start HTTP server in background with graceful shutdown via Notify
+        let registry = mr.inner(); // Arc<Registry>
+        let notify = Arc::new(Notify::new());
+        let notify_for_task = notify.clone();
+
+        let addr = SocketAddr::from(([0, 0, 0, 0], metrics_port));
+        let (_route_docs, app): (Vec<RouteDoc>, Router) = router(
+            (*registry).clone(), // take owned Registry (Clone) for router to wrap in Arc
+            None,                // or Some("/metrics".to_string()) to override the path
+        );
+
+        let run_server = async move {
+            let listener = match TcpListener::bind(addr).await {
+                Ok(listener) => listener,
+                Err(err) => {
+                    panic!("failed to bind metrics server to {addr}: {err}");
+                }
+            };
+
+            if let Err(err) = axum::serve(listener, app)
+                .with_graceful_shutdown(async move {
+                    // wait for shutdown signal
+                    notify_for_task.notified().await;
+                })
+                .await
+            {
+                tracing::error!("[kvbm] metrics server error: {err}");
+            }
+        };
+
+        // Spawn on existing runtime if present, otherwise start our own.
+        if tokio::runtime::Handle::try_current().is_ok() {
+            tokio::spawn(run_server);
+        } else {
+            thread::spawn(move || {
+                let rt = tokio::runtime::Builder::new_multi_thread()
+                    .enable_all()
+                    .build()
+                    .expect("build tokio runtime");
+                rt.block_on(run_server);
+            });
+        }
+
         Self {
             offload_requests,
             offload_blocks_d2h,
             onboard_requests,
             onboard_blocks_h2d,
             onboard_blocks_d2d,
-            save_kv_layer_requests,
             matched_tokens,
+            shutdown_notify: Some(notify),
+        }
+    }
+}
+
+impl Drop for KvbmMetrics {
+    fn drop(&mut self) {
+        if let Some(n) = &self.shutdown_notify {
+            // (all KvbmMetrics clones) + 1 (held by server task)
+            // strong_count == 2 means this is the last metrics instance
+            if Arc::strong_count(n) == 2 {
+                n.notify_waiters();
+            }
         }
     }
 }
+
+/// A raw, standalone Prometheus metrics registry implementation using the fixed prefix: `kvbm_`
+#[derive(Debug, Clone)]
+pub struct KvbmMetricsRegistry {
+    registry: Arc<Registry>,
+    prefix: String,
+}
+
+impl KvbmMetricsRegistry {
+    pub fn new() -> Self {
+        Self {
+            registry: Arc::new(Registry::new()),
+            prefix: "kvbm".to_string(),
+        }
+    }
+
+    pub fn create_intcounter(
+        &self,
+        name: &str,
+        description: &str,
+        labels: &[(&str, &str)],
+    ) -> anyhow::Result<IntCounter> {
+        let metrics_name = sanitize_prometheus_name(&format!("{}_{}", self.prefix, name))?;
+        let const_labels: HashMap<String, String> = labels
+            .iter()
+            .map(|(k, v)| (k.to_string(), v.to_string()))
+            .collect();
+        let opts = Opts::new(metrics_name, description).const_labels(const_labels);
+        let c = IntCounter::with_opts(opts)?;
+        self.registry.register(Box::new(c.clone()))?;
+        Ok(c)
+    }
+
+    pub fn inner(&self) -> Arc<Registry> {
+        Arc::clone(&self.registry)
+    }
+}
+
+impl Default for KvbmMetricsRegistry {
+    fn default() -> Self {
+        Self::new()
+    }
+}

lib/runtime/src/metrics/prometheus_names.rs

@@ -318,13 +318,25 @@ pub mod distributed_runtime {
     pub const UPTIME_SECONDS: &str = "uptime_seconds";
 }
 
-/// KVBM connector
-pub mod kvbm_connector {
-    /// KVBM connector leader
-    pub const KVBM_CONNECTOR_LEADER: &str = "kvbm_connector_leader";
+/// KVBM
+pub mod kvbm {
+    /// The number of offload requests
+    pub const OFFLOAD_REQUESTS: &str = "offload_requests";
 
-    /// KVBM connector worker
-    pub const KVBM_CONNECTOR_WORKER: &str = "kvbm_connector_worker";
+    /// The number of offload blocks from device to host
+    pub const OFFLOAD_BLOCKS_D2H: &str = "offload_blocks_d2h";
+
+    /// The number of onboard requests
+    pub const ONBOARD_REQUESTS: &str = "onboard_requests";
+
+    /// The number of onboard blocks from host to device
+    pub const ONBOARD_BLOCKS_H2D: &str = "onboard_blocks_h2d";
+
+    /// The number of onboard blocks from disk to device
+    pub const ONBOARD_BLOCKS_D2D: &str = "onboard_blocks_d2d";
+
+    /// The number of matched tokens
+    pub const MATCHED_TOKENS: &str = "matched_tokens";
 }
 
 /// KvStats metrics from LLM workers