vLLM Kubernetes configuration reference

Reference tables for the values.yaml placeholders, collection interval settings, and troubleshooting steps used when monitoring vLLM on Kubernetes with Instana.

Configuration placeholder reference

The following table summarizes every placeholder in the values.yaml template and how to obtain its value:

Placeholder Where to find it Example
<vllm-namespace-name> Namespace: in kubectl describe pod text-generation
<component-label-value> Labels: component= in kubectl describe pod huggingface-vllm
<vllm-metrics-port> Port: under Containers in kubectl describe pod 8000
<cluster-name> kubectl config current-context prod-cluster-01
<cluster-distribution> Your cluster type openshift, eks, gke, aks, or kubernetes
<cluster-managed-by> Your platform team CNF, EKS, GKE, AKS
<cluster-short-name> Short label for the cluster prod-01
<cluster-full-name> Fully qualified cluster name prod-cluster-01.example.com
<cluster-version> kubectl version --short v1.29.0
<cluster-uid> kubectl get namespace kube-system -o jsonpath='{.metadata.uid}' 15a3639a-17b7-4f16-9c82-35967f73430d
<otlp-endpoint> Instana backend OTLP endpoint https://otlp-grpc-<tenant>-saas.instana.io:4318
<your-instana-agent-key> Instana UI: More > Agents > Install Agents abc123…

For Instana backend endpoints, see Self-hosted endpoints or SaaS endpoints.

Tuning collection intervals

Adjust the following fields in values.yaml to balance metric freshness against resource overhead:

Setting Path in values.yaml Recommended value
Prometheus scrape interval config.receivers.prometheus.config.scrape_configs[0].scrape_interval 10s (real-time), 30s (production)
Kubernetes cluster metrics config.receivers.k8s_cluster.collection_interval 10s (dev), 30s60s (production)
Kubelet stats config.receivers.kubeletstats.collection_interval 10s (dev), 30s60s (production)