From 9a88ffd08c64a6e83a4a2852f7efe5d23f26494f Mon Sep 17 00:00:00 2001
From: Eero Tamminen <eero.t.tamminen@intel.com>
Date: Thu, 3 Jul 2025 18:59:57 +0300
Subject: [PATCH] Automated Grafana dashboard + Prometheus support for KubeAI

- Install vLLM dashboards if Grafana pod is found
- Install vLLM podMonitor if deployed Prometheus Helm release is found
- More descriptive KubeAI Helm release name: "opea" -> "opea-kubeai"

Signed-off-by: Eero Tamminen <eero.t.tamminen@intel.com>
---
 kubeai/README.md  | 16 +++++++++-------
 kubeai/install.sh | 44 ++++++++++++++++++++++++++++++++++++--------
 2 files changed, 45 insertions(+), 15 deletions(-)

diff --git a/kubeai/README.md b/kubeai/README.md
index 49a7a7b22..3be889f09 100644
--- a/kubeai/README.md
+++ b/kubeai/README.md
@@ -300,26 +300,28 @@ on the configuration options.
 
 # Observability
 
-With [Prometheus](../helm-charts/monitoring.md) running, install script can enable monitoring of the vLLM inference engine instances.
+With [kube-prometheus-stack](../helm-charts/monitoring.md) Helm chart already deployed, install script will automatically enable monitoring for the vLLM inference engine pods.
 
-Script requires Prometheus Helm chart release name for that, e.g.
+If script did not detect it, one can specify Prometheus Helm chart release manually:
 
 ```
 release=prometheus-stack
 ./install.sh $release
 ```
 
-Port-forward Grafana.
+If script finds also a (running) Grafana instance, it will install "vLLM scaling" and "vLLM details" dashboards for it.
+
+But they can be installed also manually afterwards:
 
 ```
-kubectl port-forward -n $ns svc/$release-grafana 3000:80
+ns=monitoring # Grafana namespace
+kubectl apply -n $ns -f grafana/vllm-scaling.yaml -f grafana/vllm-details.yaml
 ```
 
-Install "vLLM scaling" and "vLLM details" dashboards, to the same namespace as Grafana.
+Then port-forward Grafana.
 
 ```
-ns=monitoring
-kubectl apply -n $ns -f grafana/vllm-scaling.yaml -f grafana/vllm-details.yaml
+kubectl port-forward -n $ns svc/$release-grafana 3000:80
 ```
 
 Open web-browser to `http://localhost:3000` with `admin` / `prom-operator` given as the username / password for login, to view the dashboards.
diff --git a/kubeai/install.sh b/kubeai/install.sh
index 8bfde03ef..55804c37f 100755
--- a/kubeai/install.sh
+++ b/kubeai/install.sh
@@ -8,30 +8,32 @@ DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 error_exit () {
 	name=${0##*/}
 	cat <<EOF
+
 Usage: [HF_TOKEN=<token>] $name [HF token file] [Prometheus release name]
 
 HuggingFace token for accessing models can be given either as an
 environment variable, or as a name of a file containing the token
 value.
 
-If Prometheus Helm release name is given, Prometheus monitoring is
-enabled for the inference engine instances, and vLLM dashboard
-(configMap) is installed for Grafana.
+If script finds deployed "kube-prometheus-stack" Helm chart release,
+it enables Prometheus monitoring for the vLLM inference engine pods.
+vLLM dashboards (configMaps) are installed if also a running Grafana
+instance is detected.
+
+Prometheus release name can also be given as an argument,
 
 ERROR: $1!
 EOF
     exit 1
 }
 
-metrics=""
+release=""
 for arg in "$@"; do
 	if [ -f "$arg" ]; then
 		echo "Using HF token from '$arg' file."
 		HF_TOKEN=$(cat "$arg")
 	else
-		echo "Enabling vLLM inference pod monitoring for '$arg' Prometheus Helm install."
-		metrics="--set metrics.prometheusOperator.vLLMPodMonitor.labels.release=$arg"
-		metrics="$metrics --set metrics.prometheusOperator.vLLMPodMonitor.enabled=true"
+		release="$arg"
 	fi
 done
 
@@ -39,7 +41,33 @@ if [ -z "$HF_TOKEN" ]; then
 	error_exit "HF token missing"
 fi
 
-helm upgrade --install opea -n kubeai kubeai/kubeai \
+if [ -z "$release" ]; then
+	if [ -z "$(which jq)" ]; then
+		error_exit "please install 'jq' to parse Helm releases info"
+	fi
+	# check whether cluster has deployed Prometheus Helm chart release, if none specified
+	release=$(helm list -A -o json | jq '.[] | select(.chart|match("^kube-prometheus-stack")) | select(.status=="deployed") | .name' | tr -d '"')
+fi
+
+metrics=""
+if [ -n "$release" ]; then
+	running="status.phase=Running"
+	grafana="app.kubernetes.io/name=grafana"
+	jsonpath="{.items[0].metadata.namespace}"
+
+	# check for Grafana namespace
+	ns=$(kubectl get -A pod --field-selector="$running" --selector="$grafana" -o jsonpath="$jsonpath")
+	if [ -n "$ns" ]; then
+		echo "Grafana available, installing vLLM dashboards to '$ns' namespace."
+		kubectl apply -n $ns -f $DIR/grafana/vllm-scaling.yaml -f $DIR/grafana/vllm-details.yaml
+	fi
+
+	echo "Enabling vLLM pod monitoring for '$release' Prometheus Helm install."
+	metrics="--set metrics.prometheusOperator.vLLMPodMonitor.labels.release=$release"
+	metrics="$metrics --set metrics.prometheusOperator.vLLMPodMonitor.enabled=true"
+fi
+
+helm upgrade --install opea-kubeai -n kubeai kubeai/kubeai \
     --create-namespace \
     --set secrets.huggingface.token="$HF_TOKEN" \
     -f $DIR/opea-values.yaml $metrics