feature: Support ServiceMonitors for External Prometheus stack

Describe the feature

Enhancement: Support external ServiceMonitors

Add native integration with an already-running Prometheus stack (e.g. the EKS-Blueprints “kube-prometheus-stack” add-on).

Why do you need this feature?

In many production K8s environments a platform team already operates a central Prometheus stack.
But the current vllm-stack chart always creates its own ServiceMonitor only when the bundled observability stack deployment observability/kube-prom-stack.yaml is installed.

Users with existing Prometheus setups (e.g., EKS Blueprint addons) deploy vLLM without that sub-chart, which means:

• The vLLM router / engine metrics are never scraped.
• Users must craft and apply a separate ServiceMonitor manifest by hand.

---

Why it matters

• Operational simplicity – Teams adopting vLLM shouldn’t have to maintain out-of-tree YAML just to get metrics.
• Consistency – Seamless integration with existing monitoring infrastructure (e.g. kube-prom-stack, Loki, Tempo)
  • No need for manual ServiceMonitor creation
• Backward compatibility - with built-in observability stack

---

Possible enhancement (values & templates)

Add configuration options in values.yaml to conditionally create ServiceMonitors for external Prometheus instances:

externalPrometheus:
  enabled: false          # default – preserves current behaviour
  namespace: "monitoring" # where the Prometheus Operator lives
  labels:                 # label selector for Prometheus (<namespace>/serviceMonitorSelector)
    release: "kube-prometheus-stack"
  serviceSelector:        # optional extra selector for the vLLM Service
    matchExpressions:
      -  key: app.kubernetes.io/managed-by
         operator: In
         values: [ Helm]
      - key: "release"
        operator: In
        values: [test, router] 
     - key: "environment"  
        operator: In  
        values: [test, router]  # Include both test and router environments  

Possible Implementation

• Add new template helm/templates/external-servicemonitor.yaml
• Make it conditional based on externalPrometheus.enabled flag
• Support configurable namespace, labels, and selectors for different Prometheus setups

{{- if .Values.externalPrometheus.enabled }}
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: {{ include "vllm.fullname" . }}
  namespace: {{ .Values.externalPrometheus.namespace }}
  labels: {{- toYaml .Values.externalPrometheus.labels | nindent 4 }}
spec:
  selector: {{- toYaml .Values.externalPrometheus.serviceSelector | nindent 6 }}
  endpoints:
    - port: http-metrics
      interval: 30s
{{- end }}

---

Current Workaround
Users must manually create and apply ServiceMonitor resources targeting their external Prometheus operators.

Why do you need this feature?

No response

Additional context

No response

Start with values.yaml and the existing observability/kube-prom-stack.yaml ServiceMonitor setup, then compare the chart templates around that resource. Add the proposed helm/templates/external-servicemonitor.yaml integration so externalPrometheus settings control namespace, labels, and service selectors while the default behavior remains unchanged.