feat: automatically setup and inject prometheus configuration

Signed-off-by: Julien Mancuso <jmancuso@nvidia.com>

Author: julienmancuso

components/backends/vllm/deploy/disagg_planner.yaml

@@ -53,9 +53,6 @@ spec:
         mainContainer:
           image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
           workingDir: /workspace/components/planner/src/dynamo/planner
-          ports:
-            - name: metrics
-              containerPort: 9085
           command:
             - /bin/sh
             - -c
@@ -66,7 +63,6 @@ spec:
               --backend=vllm
               --adjustment-interval=60
               --profile-results-dir=/workspace/profiling_results
-              --prometheus-port=9085
     Prometheus: # NOTE: this is set on Prometheus to ensure a service is created for the Prometheus component. This is a workaround and should be managed differently.
       dynamoNamespace: vllm-disagg-planner
       componentType: frontend

deploy/cloud/helm/platform/components/operator/templates/prometheus.yaml

@@ -16,51 +16,7 @@
 apiVersion: monitoring.coreos.com/v1
 kind: PodMonitor
 metadata:
-  name: dynamo-frontend
-spec:
-  {{- if .Values.namespaceRestriction.enabled }}
-  namespaceSelector:
-    matchNames:
-    - {{ .Release.Namespace }}
-  {{- else }}
-  namespaceSelector:
-    any: true
-  {{- end }}
-  podMetricsEndpoints:
-  - interval: 30s
-    path: /metrics
-    port: http
-  selector:
-    matchLabels:
-      nvidia.com/dynamo-component-type: frontend
-      nvidia.com/metrics-enabled: "true"
---- 
-apiVersion: monitoring.coreos.com/v1
-kind: PodMonitor
-metadata:
-  name: dynamo-worker
-spec:
-  {{- if .Values.namespaceRestriction.enabled }}
-  namespaceSelector:
-    matchNames:
-    - {{ .Release.Namespace }}
-  {{- else }}
-  namespaceSelector:
-    any: true
-  {{- end }}
-  podMetricsEndpoints:
-  - interval: 30s
-    path: /metrics
-    port: system
-  selector:
-    matchLabels:
-      nvidia.com/dynamo-component-type: worker
-      nvidia.com/metrics-enabled: "true"
---- 
-apiVersion: monitoring.coreos.com/v1
-kind: PodMonitor
-metadata:
-  name: dynamo-planner
+  name: dynamo-metrics
 spec:
   {{- if .Values.namespaceRestriction.enabled }}
   namespaceSelector:
@@ -76,6 +32,5 @@ spec:
     port: metrics
   selector:
     matchLabels:
-      nvidia.com/dynamo-component-type: planner
       nvidia.com/metrics-enabled: "true"
 {{- end }}

deploy/cloud/helm/platform/components/operator/values.yaml

@@ -95,7 +95,7 @@ dynamo:
     password: ""
     existingSecretName: ''
     secure: true
-  
+
   metrics:
     prometheusEndpoint: ""
 

deploy/cloud/operator/internal/consts/consts.go

@@ -16,6 +16,10 @@ const (
 	DynamoServicePortName   = "http"
 	DynamoContainerPortName = "http"
 
+	DynamoMetricsPortName = "metrics"
+
+	DynamoPlannerMetricsPort = 9085
+
 	DynamoSystemPort     = 9090
 	DynamoSystemPortName = "system"
 

deploy/cloud/operator/internal/dynamo/component_frontend.go

@@ -30,13 +30,17 @@ func (f *FrontendDefaults) GetBaseContainer(context ComponentContext) (corev1.Co
 	container.Command = []string{"python3"}
 	container.Args = []string{"-m", "dynamo.frontend"}
 
-	// Add HTTP port
 	container.Ports = []corev1.ContainerPort{
 		{
 			Protocol:      corev1.ProtocolTCP,
 			Name:          commonconsts.DynamoContainerPortName,
 			ContainerPort: int32(commonconsts.DynamoServicePort),
 		},
+		{
+			Protocol:      corev1.ProtocolTCP,
+			Name:          commonconsts.DynamoMetricsPortName,
+			ContainerPort: int32(commonconsts.DynamoServicePort),
+		},
 	}
 
 	// Add frontend-specific defaults

deploy/cloud/operator/internal/dynamo/component_planner.go

@@ -6,6 +6,8 @@
 package dynamo
 
 import (
+	"fmt"
+
 	commonconsts "github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/consts"
 	corev1 "k8s.io/api/core/v1"
 )
@@ -21,6 +23,19 @@ func NewPlannerDefaults() *PlannerDefaults {
 
 func (p *PlannerDefaults) GetBaseContainer(context ComponentContext) (corev1.Container, error) {
 	container := p.getCommonContainer(context)
+	container.Ports = []corev1.ContainerPort{
+		{
+			Protocol:      corev1.ProtocolTCP,
+			Name:          commonconsts.DynamoMetricsPortName,
+			ContainerPort: int32(commonconsts.DynamoPlannerMetricsPort),
+		},
+	}
+	container.Env = append(container.Env, []corev1.EnvVar{
+		{
+			Name:  "PROMETHEUS_PORT",
+			Value: fmt.Sprintf("%d", commonconsts.DynamoPlannerMetricsPort),
+		},
+	}...)
 	return container, nil
 }
 

deploy/cloud/operator/internal/dynamo/component_planner_test.go

@@ -6,8 +6,10 @@
 package dynamo
 
 import (
+	"fmt"
 	"testing"
 
+	commonconsts "github.com/ai-dynamo/dynamo/deploy/cloud/operator/internal/consts"
 	"github.com/google/go-cmp/cmp"
 	corev1 "k8s.io/api/core/v1"
 )
@@ -46,10 +48,14 @@ func TestPlannerDefaults_GetBaseContainer(t *testing.T) {
 					"/bin/sh",
 					"-c",
 				},
+				Ports: []corev1.ContainerPort{
+					{Name: commonconsts.DynamoMetricsPortName, ContainerPort: commonconsts.DynamoPlannerMetricsPort, Protocol: corev1.ProtocolTCP},
+				},
 				Env: []corev1.EnvVar{
 					{Name: "DYN_NAMESPACE", Value: "dynamo-namespace"},
 					{Name: "DYN_PARENT_DGD_K8S_NAME", Value: "name"},
 					{Name: "DYN_PARENT_DGD_K8S_NAMESPACE", Value: "namespace"},
+					{Name: "PROMETHEUS_PORT", Value: fmt.Sprintf("%d", commonconsts.DynamoPlannerMetricsPort)},
 				},
 			},
 		},

deploy/cloud/operator/internal/dynamo/component_worker.go

@@ -25,13 +25,17 @@ func NewWorkerDefaults() *WorkerDefaults {
 func (w *WorkerDefaults) GetBaseContainer(context ComponentContext) (corev1.Container, error) {
 	container := w.getCommonContainer(context)
 
-	// Add system port
 	container.Ports = []corev1.ContainerPort{
 		{
 			Protocol:      corev1.ProtocolTCP,
 			Name:          commonconsts.DynamoSystemPortName,
 			ContainerPort: int32(commonconsts.DynamoSystemPort),
 		},
+		{
+			Protocol:      corev1.ProtocolTCP,
+			Name:          commonconsts.DynamoMetricsPortName,
+			ContainerPort: int32(commonconsts.DynamoSystemPort),
+		},
 	}
 
 	container.LivenessProbe = &corev1.Probe{