diff --git a/pai-management/bootstrap/prometheus/prometheus-configmap.yaml.template b/pai-management/bootstrap/prometheus/prometheus-configmap.yaml.template
index ff1677e0f..0f35b4d6c 100644
--- a/pai-management/bootstrap/prometheus/prometheus-configmap.yaml.template
+++ b/pai-management/bootstrap/prometheus/prometheus-configmap.yaml.template
@@ -65,6 +65,9 @@ data:
         - source_labels: [__meta_kubernetes_pod_name]
           action: replace
           target_label: k8s_pod_name
+        - source_labels: [__meta_kubernetes_pod_label_app]
+          action: replace
+          target_label: pai_service_name
 {% if clusterinfo['prometheusinfo']['alerting'] %}
     alerting:
       alertmanagers:
diff --git a/prometheus/doc/exporter-for-other-services.md b/prometheus/doc/exporter-for-other-services.md
index 0c156b6fe..532586779 100644
--- a/prometheus/doc/exporter-for-other-services.md
+++ b/prometheus/doc/exporter-for-other-services.md
@@ -148,4 +148,66 @@ If service decide to leverage node-exporter to expose metrics, service should mo
 located in `/datastorage/prometheus` and write file with extension `.prom`.
 
 For expose using port, service should annotate pod with `prometheus.io/scrape`, `prometheus.io/path`
-and `prometheus.io/port`.
+and `prometheus.io/port`. For service with label key `app`, prometheus will generate `pai_service_name`
+label for generated metrics.
+
+Prometheus will list all pods and get those pods with `prometheus.io/scrape` is true, and then,
+prometheus will issue request to `http://${pod_host_ip}:${prometheus.io/port}${prometheus.io/path}` to
+get metrics.
+
+For example, following deployment:
+
+```yaml
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: foo
+spec:
+  selector:
+    matchLabels:
+      app: foo
+  template:
+    metadata:
+      name: foo
+      labels:
+        app: foo
+      annotations:
+        prometheus.io/scrape: "true"
+        prometheus.io/path: "/"
+        prometheus.io/port: "80"
+    spec:
+      containers:
+      - name: nginx
+        image: nginx
+        ports:
+        - containerPort: 80
+        volumeMounts:
+        - name: workdir
+          mountPath: /usr/share/nginx/html
+      initContainers:
+        - name: install
+          image: busybox
+          command: ["sh", "-c", "echo pai_service_count 1 > /work-dir/index.html"]
+          volumeMounts:
+          - name: workdir
+            mountPath: "/work-dir"
+      volumes:
+        - name: workdir
+          emptyDir: {}
+      hostNetwork: true
+```
+
+will generate metrics like:
+
+`pai_service_count{instance="10.151.40.4:80",job="pai_serivce_exporter",k8s_pod_name="foo-69869cc4d6",pai_service_name="foo"} 1`
+
+metric labels `k8s_pod_name` and `pai_service_name` are generated from pod name and pod label value which has key `app`.
+
+In above example, we use nginx to accept requests from prometheus and expose metrics we write in
+initContainers. It is best for service owner to use
+[prometheus client](https://prometheus.io/docs/instrumenting/clientlibs/) to expose metrics.
+
+Note that, currently pai do not assume
+[kubernetes network model](https://kubernetes.io/docs/concepts/cluster-administration/networking/)
+enabled, so service has to use hostNetwork, this implies service owner should carefully choose port
+number to avoid port conflict, it is best to configure port through pai service configuration.