[Feat] Production Stack Router: Add OpenTelemetry tracing support with W3C context propagation

helm/README.md

@@ -201,6 +201,14 @@ This table documents all available configuration values for the Production Stack
 | `routerSpec.readinessProbe.failureThreshold` | integer |`3`| Failure threshold for router's readiness probe |
 | `routerSpec.readinessProbe.httpGet.path` | string |`"/health"`| Endpoint that the router's readiness probe will be testing |
 
+#### Router OpenTelemetry Configuration
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `routerSpec.otel.endpoint` | string | `""` | OTLP endpoint for tracing (e.g., "otel-collector:4317"). Tracing is enabled when this is set. |
+| `routerSpec.otel.serviceName` | string | `"vllm-router"` | Service name for OpenTelemetry traces |
+| `routerSpec.otel.secure` | boolean | `false` | Use secure (TLS) connection for OTLP exporter |
+
 #### Router Ingress Configuration
 
 | Field | Type | Default | Description |

helm/templates/deployment-router.yaml

@@ -136,6 +136,15 @@ spec:
           - "--lmcache-controller-port"
           - "{{ .Values.routerSpec.lmcacheControllerPort }}"
           {{- end }}
+          {{- if .Values.routerSpec.otel.endpoint }}
+          - "--otel-endpoint"
+          - "{{ .Values.routerSpec.otel.endpoint }}"
+          - "--otel-service-name"
+          - "{{ .Values.routerSpec.otel.serviceName | default "vllm-router" }}"
+          {{- if .Values.routerSpec.otel.secure }}
+          - "--otel-secure"
+          {{- end }}
+          {{- end }}
         {{- if .Values.routerSpec.resources }}
         resources:
           {{- if .Values.routerSpec.resources.requests }}

helm/tests/routerOtel_test.yaml

@@ -0,0 +1,77 @@
+suite: test router OpenTelemetry configuration
+templates:
+  - deployment-router.yaml
+tests:
+  - it: should not include otel args when endpoint is not set
+    set:
+      routerSpec:
+        enableRouter: true
+        otel:
+          endpoint: ""
+    asserts:
+      - template: deployment-router.yaml
+        notContains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-endpoint"
+
+  - it: should include otel args when endpoint is set
+    set:
+      routerSpec:
+        enableRouter: true
+        otel:
+          endpoint: "otel-collector:4317"
+          serviceName: "vllm-router"
+          secure: false
+    asserts:
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-endpoint"
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "otel-collector:4317"
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-service-name"
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "vllm-router"
+      - template: deployment-router.yaml
+        notContains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-secure"
+
+  - it: should use custom service name when specified
+    set:
+      routerSpec:
+        enableRouter: true
+        otel:
+          endpoint: "jaeger:4317"
+          serviceName: "my-custom-router"
+          secure: false
+    asserts:
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "my-custom-router"
+
+  - it: should include otel-secure flag when secure is true
+    set:
+      routerSpec:
+        enableRouter: true
+        otel:
+          endpoint: "otel-collector:4317"
+          serviceName: "vllm-router"
+          secure: true
+    asserts:
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-endpoint"
+      - template: deployment-router.yaml
+        contains:
+          path: spec.template.spec.containers[0].args
+          content: "--otel-secure"

helm/values.schema.json

@@ -580,6 +580,26 @@
           "additionalProperties": {
             "type": "string"
           }
+        },
+        "otel": {
+          "type": "object",
+          "description": "OpenTelemetry tracing configuration for the router",
+          "properties": {
+            "endpoint": {
+              "type": "string",
+              "description": "OTLP endpoint for tracing (e.g., 'otel-collector:4317'). Tracing is enabled when this is set."
+            },
+            "serviceName": {
+              "type": "string",
+              "description": "Service name for OpenTelemetry traces",
+              "default": "vllm-router"
+            },
+            "secure": {
+              "type": "boolean",
+              "description": "Use secure (TLS) connection for OTLP exporter",
+              "default": false
+            }
+          }
         }
       }
     }

helm/values.yaml

@@ -377,6 +377,16 @@ routerSpec:
   # -- Window size in seconds to calculate the request statistics
   requestStatsWindow: 60
 
+  # -- OpenTelemetry tracing configuration
+  # When otelEndpoint is set, tracing is automatically enabled
+  otel:
+    # -- OTLP endpoint for tracing (e.g., "localhost:4317" or "otel-collector:4317")
+    endpoint: ""
+    # -- Service name for traces (default: "vllm-router")
+    serviceName: "vllm-router"
+    # -- Use secure (TLS) connection for OTLP exporter (default: false, i.e., insecure)
+    secure: false
+
   # -- deployment strategy
   strategy: {}
 

pyproject.toml

@@ -26,6 +26,9 @@ dependencies = [
     "xxhash==3.5.0",
     "psutil==7.0.0",
     "pyyaml>=6.0.2",
+    "opentelemetry-api>=1.28.0",
+    "opentelemetry-sdk>=1.28.0",
+    "opentelemetry-exporter-otlp>=1.28.0",
 ]
 
 [project.scripts]

src/tests/test_otel_tracing.py

@@ -0,0 +1,81 @@
+import pytest
+from opentelemetry.trace import SpanKind
+
+import vllm_router.experimental.otel.tracing as tracing_module
+from vllm_router.experimental.otel.tracing import (
+    end_span,
+    extract_context,
+    initialize_tracing,
+    inject_context,
+    is_tracing_enabled,
+    shutdown_tracing,
+    start_span,
+)
+
+
+@pytest.fixture(autouse=True)
+def reset_tracing_state():
+    """Reset global tracing state before each test."""
+    tracing_module._tracer = None
+    tracing_module._provider = None
+    tracing_module._tracing_enabled = False
+    yield
+    # Cleanup after test
+    if tracing_module._tracing_enabled:
+        shutdown_tracing()
+
+
+class TestTracingIntegration:
+    def test_full_request_flow(self):
+        """Test a complete request tracing flow."""
+        initialize_tracing(service_name="vllm-router", otlp_endpoint="localhost:4317")
+
+        # Simulate incoming request with trace context
+        incoming_headers = {
+            "traceparent": "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01",
+        }
+        incoming_context = extract_context(incoming_headers)
+
+        # Create parent span (router)
+        parent_span, parent_context = start_span(
+            name="router /v1/chat/completions",
+            parent_context=incoming_context,
+            kind=SpanKind.SERVER,
+            attributes={
+                "http.method": "POST",
+                "vllm.model": "Qwen/Qwen2.5-7B-Instruct",
+            },
+        )
+
+        # Create child span (backend request)
+        child_span, child_context = start_span(
+            name="backend_request",
+            parent_context=parent_context,
+            kind=SpanKind.CLIENT,
+            attributes={
+                "http.url": "http://backend:8000/v1/chat/completions",
+            },
+        )
+
+        # Inject context into outgoing headers
+        outgoing_headers = {}
+        inject_context(outgoing_headers, child_context)
+
+        assert "traceparent" in outgoing_headers
+
+        # End spans in reverse order
+        end_span(child_span, status_code=200)
+        end_span(parent_span, status_code=200)
+
+    def test_tracing_disabled_flow(self):
+        """Test that operations handle disabled tracing gracefully."""
+        assert is_tracing_enabled() is False
+
+        # These should not raise even when tracing is disabled
+        headers = {}
+        inject_context(headers)
+        end_span(None)
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

src/vllm_router/app.py

@@ -79,6 +79,18 @@
 except ImportError:
     semantic_cache_available = False
 
+try:
+    # OpenTelemetry tracing integration
+    from vllm_router.experimental.otel import (
+        initialize_tracing,
+        is_tracing_enabled,
+        shutdown_tracing,
+    )
+
+    otel_available = True
+except ImportError:
+    otel_available = False
+
 logger = logging.getLogger("uvicorn")
 
 
@@ -121,6 +133,11 @@ async def lifespan(app: FastAPI):
     logger.info("Closing routing logic instances")
     cleanup_routing_logic()
 
+    # Shutdown OpenTelemetry tracing if enabled
+    if otel_available and app.state.otel_enabled:
+        logger.info("Shutting down OpenTelemetry tracing")
+        shutdown_tracing()
+
 
 def initialize_all(app: FastAPI, args):
     """
@@ -142,6 +159,23 @@ def initialize_all(app: FastAPI, args):
             profile_session_sample_rate=args.sentry_profile_session_sample_rate,
         )
 
+    if otel_available and args.otel_endpoint:
+        initialize_tracing(
+            service_name=args.otel_service_name,
+            otlp_endpoint=args.otel_endpoint,
+            insecure=not args.otel_secure,
+        )
+        app.state.otel_enabled = is_tracing_enabled()
+        if app.state.otel_enabled:
+            logger.info(
+                f"OpenTelemetry tracing enabled, exporting to {args.otel_endpoint}"
+            )
+    elif args.otel_endpoint and not otel_available:
+        logger.warning(
+            "OpenTelemetry endpoint specified but OpenTelemetry packages not installed. "
+            "Install with: pip install opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp"
+        )
+
     if args.service_discovery == "static":
         initialize_service_discovery(
             ServiceDiscoveryType.STATIC,
@@ -292,6 +326,7 @@ def initialize_all(app: FastAPI, args):
 app.include_router(metrics_router)
 app.state.aiohttp_client_wrapper = AiohttpClientWrapper()
 app.state.semantic_cache_available = semantic_cache_available
+app.state.otel_enabled = False
 
 
 def main():

src/vllm_router/experimental/otel/__init__.py

@@ -0,0 +1,37 @@
+# Copyright 2024-2025 The vLLM Production Stack Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""OpenTelemetry tracing module for vLLM Router."""
+
+from vllm_router.experimental.otel.tracing import (
+    end_span,
+    extract_context,
+    get_tracer,
+    initialize_tracing,
+    inject_context,
+    is_tracing_enabled,
+    shutdown_tracing,
+    start_span,
+)
+
+__all__ = [
+    "initialize_tracing",
+    "shutdown_tracing",
+    "get_tracer",
+    "is_tracing_enabled",
+    "extract_context",
+    "inject_context",
+    "start_span",
+    "end_span",
+]