Skip to content

Monitoring

Two different methods to install and configure Prometheus and Grafana are described in this doc. * Prometheus and Grafana installation using Pod Annotations. This installs Prometheus and Grafana in the same namespace as NGINX Ingress * Prometheus and Grafana installation using Service Monitors. This installs Prometheus and Grafana in two different namespaces. This is the preferred method, and helm charts supports this by default.

Prometheus and Grafana installation using Pod Annotations

This tutorial will show you how to install Prometheus and Grafana for scraping the metrics of the Ingress-Nginx Controller.

Important

This example uses emptyDir volumes for Prometheus and Grafana. This means once the pod gets terminated you will lose all the data.

Before You Begin

  • The Ingress-Nginx Controller should already be deployed according to the deployment instructions here.

  • The controller should be configured for exporting metrics. This requires 3 configurations to the controller. These configurations are :

  • controller.metrics.enabled=true
  • controller.podAnnotations."prometheus.io/scrape"="true"
  • controller.podAnnotations."prometheus.io/port"="10254"

  • The easiest way to configure the controller for metrics is via helm upgrade. Assuming you have installed the ingress-nginx controller as a helm release named ingress-nginx, then you can simply type the command shown below :

    helm upgrade ingress-nginx ingress-nginx \
    --repo https://kubernetes.github.io/ingress-nginx \
    --namespace ingress-nginx \
    --set controller.metrics.enabled=true \
    --set-string controller.podAnnotations."prometheus\.io/scrape"="true" \
    --set-string controller.podAnnotations."prometheus\.io/port"="10254"
    

  • You can validate that the controller is configured for metrics by looking at the values of the installed release, like this:
    helm get values ingress-nginx --namespace ingress-nginx
    
  • You should be able to see the values shown below:
    ..
    controller:
      metrics:
        enabled: true
      podAnnotations:
        prometheus.io/port: "10254"
        prometheus.io/scrape: "true"
    ..
    
  • If you are not using helm, you will have to edit your manifests like this:
    • Service manifest:
      apiVersion: v1
      kind: Service
      ..
      spec:
        ports:
          - name: prometheus
            port: 10254
            targetPort: prometheus
            ..
      
    • Deployment manifest:
      apiVersion: v1
      kind: Deployment
      ..
      spec:
        template:
          metadata:
            annotations:
              prometheus.io/scrape: "true"
              prometheus.io/port: "10254"
          spec:
            containers:
              - name: controller
                ports:
                  - name: prometheus
                    containerPort: 10254
                  ..
      

Deploy and configure Prometheus Server

Note that the kustomize bases used in this tutorial are stored in the deploy folder of the GitHub repository kubernetes/ingress-nginx.

  • The Prometheus server must be configured so that it can discover endpoints of services. If a Prometheus server is already running in the cluster and if it is configured in a way that it can find the ingress controller pods, no extra configuration is needed.

  • If there is no existing Prometheus server running, the rest of this tutorial will guide you through the steps needed to deploy a properly configured Prometheus server.

  • Running the following command deploys prometheus in Kubernetes:

kubectl apply --kustomize github.com/kubernetes/ingress-nginx/deploy/prometheus/

Prometheus Dashboard

  • Open Prometheus dashboard in a web browser:
kubectl get svc -n ingress-nginx
NAME                   TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)                                      AGE
default-http-backend   ClusterIP   10.103.59.201   <none>        80/TCP                                       3d
ingress-nginx          NodePort    10.97.44.72     <none>        80:30100/TCP,443:30154/TCP,10254:32049/TCP   5h
prometheus-server      NodePort    10.98.233.86    <none>        9090:32630/TCP                               1m
  • Obtain the IP address of the nodes in the running cluster:
kubectl get nodes -o wide
  • In some cases where the node only have internal IP addresses we need to execute:
kubectl get nodes --selector=kubernetes.io/role!=master -o jsonpath={.items[*].status.addresses[?\(@.type==\"InternalIP\"\)].address}
10.192.0.2 10.192.0.3 10.192.0.4
  • Open your browser and visit the following URL: http://{node IP address}:{prometheus-svc-nodeport} to load the Prometheus Dashboard.

  • According to the above example, this URL will be http://10.192.0.3:32630

Prometheus Dashboard

Grafana

  • Install grafana using the below command
    kubectl apply --kustomize github.com/kubernetes/ingress-nginx/deploy/grafana/
    
  • Look at the services

    kubectl get svc -n ingress-nginx
    NAME                   TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)                                      AGE
    default-http-backend   ClusterIP   10.103.59.201   <none>        80/TCP                                       3d
    ingress-nginx          NodePort    10.97.44.72     <none>        80:30100/TCP,443:30154/TCP,10254:32049/TCP   5h
    prometheus-server      NodePort    10.98.233.86    <none>        9090:32630/TCP                               10m
    grafana                NodePort    10.98.233.87    <none>        3000:31086/TCP                               10m
    

  • Open your browser and visit the following URL: http://{node IP address}:{grafana-svc-nodeport} to load the Grafana Dashboard. According to the above example, this URL will be http://10.192.0.3:31086

The username and password is admin

  • After the login you can import the Grafana dashboard from official dashboards, by following steps given below :

    • Navigate to lefthand panel of grafana
    • Hover on the gearwheel icon for Configuration and click "Data Sources"
    • Click "Add data source"
    • Select "Prometheus"
    • Enter the details (note: I used http://CLUSTER_IP_PROMETHEUS_SVC:9090)
    • Left menu (hover over +) -> Dashboard
    • Click "Import"
    • Enter the copy pasted json from https://raw.githubusercontent.com/kubernetes/ingress-nginx/main/deploy/grafana/dashboards/nginx.json
    • Click Import JSON
    • Select the Prometheus data source
    • Click "Import"

Grafana Dashboard

Caveats

Wildcard ingresses

  • By default request metrics are labeled with the hostname. When you have a wildcard domain ingress, then there will be no metrics for that ingress (to prevent the metrics from exploding in cardinality). To get metrics in this case you need to run the ingress controller with --metrics-per-host=false (you will lose labeling by hostname, but still have labeling by ingress).

Grafana dashboard using ingress resource

  • If you want to expose the dashboard for grafana using an ingress resource, then you can :
    • change the service type of the prometheus-server service and the grafana service to "ClusterIP" like this :
      kubectl -n ingress-nginx edit svc grafana
      
    • This will open the currently deployed service grafana in the default editor configured in your shell (vi/nvim/nano/other)
    • scroll down to line 34 that looks like "type: NodePort"
    • change it to look like "type: ClusterIP". Save and exit.
    • create an ingress resource with backend as "grafana" and port as "3000"
  • Similarly, you can edit the service "prometheus-server" and add an ingress resource.

Prometheus and Grafana installation using Service Monitors

This document assumes you're using helm and using the kube-prometheus-stack package to install Prometheus and Grafana.

Verify Ingress-Nginx Controller is installed

  • The Ingress-Nginx Controller should already be deployed according to the deployment instructions here.

  • To check if Ingress controller is deployed,

    kubectl get pods -n ingress-nginx
    

  • The result should look something like: NAME READY STATUS RESTARTS AGE ingress-nginx-controller-7c489dc7b7-ccrf6 1/1 Running 0 19h

Verify Prometheus is installed

  • To check if Prometheus is already deployed, run the following command:

helm ls -A
NAME          NAMESPACE       REVISION    UPDATED                                 STATUS      CHART                           APP VERSION
ingress-nginx ingress-nginx   10          2022-01-20 18:08:55.267373 -0800 PST    deployed    ingress-nginx-4.0.16            1.1.1
prometheus    prometheus      1           2022-01-20 16:07:25.086828 -0800 PST    deployed    kube-prometheus-stack-30.1.0    0.53.1
- Notice that prometheus is installed in a differenet namespace than ingress-nginx

  • If prometheus is not installed, then you can install from here

Re-configure Ingress-Nginx Controller

  • The Ingress NGINX controller needs to be reconfigured for exporting metrics. This requires 3 additional configurations to the controller. These configurations are :
    controller.metrics.enabled=true
    controller.metrics.serviceMonitor.enabled=true
    controller.metrics.serviceMonitor.additionalLabels.release="prometheus"
    
  • The easiest way of doing this is to helm upgrade
    helm upgrade ingress-nginx ingress-nginx/ingress-nginx \
    --namespace ingress-nginx \
    --set controller.metrics.enabled=true \
    --set controller.metrics.serviceMonitor.enabled=true \
    --set controller.metrics.serviceMonitor.additionalLabels.release="prometheus"
    
  • Here controller.metrics.serviceMonitor.additionalLabels.release="prometheus" should match the name of the helm release of the kube-prometheus-stack

  • You can validate that the controller has been successfully reconfigured to export metrics by looking at the values of the installed release, like this:

    helm get values ingress-nginx --namespace ingress-nginx
    
    controller:
      metrics:
        enabled: true
        serviceMonitor:
          additionalLabels:
            release: prometheus
          enabled: true
    

Configure Prometheus

  • Since Prometheus is running in a different namespace and not in the ingress-nginx namespace, it would not be able to discover ServiceMonitors in other namespaces when installed. Reconfigure your kube-prometheus-stack Helm installation to set serviceMonitorSelectorNilUsesHelmValues flag to false. By default, Prometheus only discovers PodMonitors within its own namespace. This should be disabled by setting podMonitorSelectorNilUsesHelmValues to false
  • The configurations required are:
    prometheus.prometheusSpec.podMonitorSelectorNilUsesHelmValues=false
    prometheus.prometheusSpec.serviceMonitorSelectorNilUsesHelmValues=false
    
  • The easiest way of doing this is to use helm upgrade ...
    helm upgrade prometheus prometheus-community/kube-prometheus-stack \
    --namespace prometheus  \
    --set prometheus.prometheusSpec.podMonitorSelectorNilUsesHelmValues=false \
    --set prometheus.prometheusSpec.serviceMonitorSelectorNilUsesHelmValues=false
    
  • You can validate that Prometheus has been reconfigured by looking at the values of the installed release, like this:
    helm get values prometheus --namespace prometheus
    
  • You should be able to see the values shown below:
    prometheus:
      prometheusSpec:
        podMonitorSelectorNilUsesHelmValues: false
        serviceMonitorSelectorNilUsesHelmValues: false
    

Connect and view Prometheus dashboard

  • Port forward to Prometheus service. Find out the name of the prometheus service by using the following command:
    kubectl get svc -n prometheus
    

The result of this command would look like:

NAME                                      TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)                      AGE
alertmanager-operated                     ClusterIP   None             <none>        9093/TCP,9094/TCP,9094/UDP   7h46m
prometheus-grafana                        ClusterIP   10.106.28.162    <none>        80/TCP                       7h46m
prometheus-kube-prometheus-alertmanager   ClusterIP   10.108.125.245   <none>        9093/TCP                     7h46m
prometheus-kube-prometheus-operator       ClusterIP   10.110.220.1     <none>        443/TCP                      7h46m
prometheus-kube-prometheus-prometheus     ClusterIP   10.102.72.134    <none>        9090/TCP                     7h46m
prometheus-kube-state-metrics             ClusterIP   10.104.231.181   <none>        8080/TCP                     7h46m
prometheus-operated                       ClusterIP   None             <none>        9090/TCP                     7h46m
prometheus-prometheus-node-exporter       ClusterIP   10.96.247.128    <none>        9100/TCP                     7h46m
prometheus-kube-prometheus-prometheus is the service we want to port forward to. We can do so using the following command:
kubectl port-forward svc/prometheus-kube-prometheus-prometheus -n prometheus 9090:9090
When you run the above command, you should see something like:
Forwarding from 127.0.0.1:9090 -> 9090
Forwarding from [::1]:9090 -> 9090
- Open your browser and visit the following URL http://localhost:{port-forwarded-port} according to the above example it would be, http://localhost:9090

Prometheus Dashboard

Connect and view Grafana dashboard

  • Port forward to Grafana service. Find out the name of the Grafana service by using the following command:
    kubectl get svc -n prometheus
    

The result of this command would look like:

NAME                                      TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)                      AGE
alertmanager-operated                     ClusterIP   None             <none>        9093/TCP,9094/TCP,9094/UDP   7h46m
prometheus-grafana                        ClusterIP   10.106.28.162    <none>        80/TCP                       7h46m
prometheus-kube-prometheus-alertmanager   ClusterIP   10.108.125.245   <none>        9093/TCP                     7h46m
prometheus-kube-prometheus-operator       ClusterIP   10.110.220.1     <none>        443/TCP                      7h46m
prometheus-kube-prometheus-prometheus     ClusterIP   10.102.72.134    <none>        9090/TCP                     7h46m
prometheus-kube-state-metrics             ClusterIP   10.104.231.181   <none>        8080/TCP                     7h46m
prometheus-operated                       ClusterIP   None             <none>        9090/TCP                     7h46m
prometheus-prometheus-node-exporter       ClusterIP   10.96.247.128    <none>        9100/TCP                     7h46m
prometheus-grafana is the service we want to port forward to. We can do so using the following command:
kubectl port-forward svc/prometheus-grafana  3000:80 -n prometheus
When you run the above command, you should see something like:
Forwarding from 127.0.0.1:3000 -> 3000
Forwarding from [::1]:3000 -> 3000
- Open your browser and visit the following URL http://localhost:{port-forwarded-port} according to the above example it would be, http://localhost:3000 The default username/ password is admin/prom-operator - After the login you can import the Grafana dashboard from official dashboards, by following steps given below :

  • Navigate to lefthand panel of grafana
  • Hover on the gearwheel icon for Configuration and click "Data Sources"
  • Click "Add data source"
  • Select "Prometheus"
  • Enter the details (note: I used http://10.102.72.134:9090 which is the CLUSTER-IP for Prometheus service)
  • Left menu (hover over +) -> Dashboard
  • Click "Import"
  • Enter the copy pasted json from https://raw.githubusercontent.com/kubernetes/ingress-nginx/main/deploy/grafana/dashboards/nginx.json
  • Click Import JSON
  • Select the Prometheus data source
  • Click "Import"

Grafana Dashboard

Exposed metrics

Prometheus metrics are exposed on port 10254.

Request metrics

  • nginx_ingress_controller_request_duration_seconds Histogram\ The request processing (time elapsed between the first bytes were read from the client and the log write after the last bytes were sent to the client) time in seconds (affected by client speed).\ nginx var: request_time

  • nginx_ingress_controller_response_duration_seconds Histogram\ The time spent on receiving the response from the upstream server in seconds (affected by client speed when the response is bigger than proxy buffers).\ Note: can be up to several millis bigger than the nginx_ingress_controller_request_duration_seconds because of the different measuring method. nginx var: upstream_response_time

  • nginx_ingress_controller_header_duration_seconds Histogram\ The time spent on receiving first header from the upstream server\ nginx var: upstream_header_time

  • nginx_ingress_controller_connect_duration_seconds Histogram\ The time spent on establishing a connection with the upstream server\ nginx var: upstream_connect_time

  • nginx_ingress_controller_response_size Histogram\ The response length (including request line, header, and request body)\ nginx var: bytes_sent

  • nginx_ingress_controller_request_size Histogram\ The request length (including request line, header, and request body)\ nginx var: request_length

  • nginx_ingress_controller_requests Counter\ The total number of client requests

  • nginx_ingress_controller_bytes_sent Histogram\ The number of bytes sent to a client. Deprecated, use nginx_ingress_controller_response_size\ nginx var: bytes_sent

  • nginx_ingress_controller_ingress_upstream_latency_seconds Summary\ Upstream service latency per Ingress. Deprecated, use nginx_ingress_controller_connect_duration_seconds\ nginx var: upstream_connect_time

# HELP nginx_ingress_controller_bytes_sent The number of bytes sent to a client. DEPRECATED! Use nginx_ingress_controller_response_size
# TYPE nginx_ingress_controller_bytes_sent histogram
# HELP nginx_ingress_controller_connect_duration_seconds The time spent on establishing a connection with the upstream server
# TYPE nginx_ingress_controller_connect_duration_seconds nginx_ingress_controller_connect_duration_seconds
* HELP nginx_ingress_controller_header_duration_seconds The time spent on receiving first header from the upstream server
# TYPE nginx_ingress_controller_header_duration_seconds histogram
# HELP nginx_ingress_controller_ingress_upstream_latency_seconds Upstream service latency per Ingress DEPRECATED! Use nginx_ingress_controller_connect_duration_seconds
# TYPE nginx_ingress_controller_ingress_upstream_latency_seconds summary
# HELP nginx_ingress_controller_request_duration_seconds The request processing time in milliseconds
# TYPE nginx_ingress_controller_request_duration_seconds histogram
# HELP nginx_ingress_controller_request_size The request length (including request line, header, and request body)
# TYPE nginx_ingress_controller_request_size histogram
# HELP nginx_ingress_controller_requests The total number of client requests.
# TYPE nginx_ingress_controller_requests counter
# HELP nginx_ingress_controller_response_duration_seconds The time spent on receiving the response from the upstream server
# TYPE nginx_ingress_controller_response_duration_seconds histogram
# HELP nginx_ingress_controller_response_size The response length (including request line, header, and request body)
# TYPE nginx_ingress_controller_response_size histogram

Nginx process metrics

# HELP nginx_ingress_controller_nginx_process_connections current number of client connections with state {active, reading, writing, waiting}
# TYPE nginx_ingress_controller_nginx_process_connections gauge
# HELP nginx_ingress_controller_nginx_process_connections_total total number of connections with state {accepted, handled}
# TYPE nginx_ingress_controller_nginx_process_connections_total counter
# HELP nginx_ingress_controller_nginx_process_cpu_seconds_total Cpu usage in seconds
# TYPE nginx_ingress_controller_nginx_process_cpu_seconds_total counter
# HELP nginx_ingress_controller_nginx_process_num_procs number of processes
# TYPE nginx_ingress_controller_nginx_process_num_procs gauge
# HELP nginx_ingress_controller_nginx_process_oldest_start_time_seconds start time in seconds since 1970/01/01
# TYPE nginx_ingress_controller_nginx_process_oldest_start_time_seconds gauge
# HELP nginx_ingress_controller_nginx_process_read_bytes_total number of bytes read
# TYPE nginx_ingress_controller_nginx_process_read_bytes_total counter
# HELP nginx_ingress_controller_nginx_process_requests_total total number of client requests
# TYPE nginx_ingress_controller_nginx_process_requests_total counter
# HELP nginx_ingress_controller_nginx_process_resident_memory_bytes number of bytes of memory in use
# TYPE nginx_ingress_controller_nginx_process_resident_memory_bytes gauge
# HELP nginx_ingress_controller_nginx_process_virtual_memory_bytes number of bytes of memory in use
# TYPE nginx_ingress_controller_nginx_process_virtual_memory_bytes gauge
# HELP nginx_ingress_controller_nginx_process_write_bytes_total number of bytes written
# TYPE nginx_ingress_controller_nginx_process_write_bytes_total counter

Controller metrics

# HELP nginx_ingress_controller_build_info A metric with a constant '1' labeled with information about the build.
# TYPE nginx_ingress_controller_build_info gauge
# HELP nginx_ingress_controller_check_success Cumulative number of Ingress controller syntax check operations
# TYPE nginx_ingress_controller_check_success counter
# HELP nginx_ingress_controller_config_hash Running configuration hash actually running
# TYPE nginx_ingress_controller_config_hash gauge
# HELP nginx_ingress_controller_config_last_reload_successful Whether the last configuration reload attempt was successful
# TYPE nginx_ingress_controller_config_last_reload_successful gauge
# HELP nginx_ingress_controller_config_last_reload_successful_timestamp_seconds Timestamp of the last successful configuration reload.
# TYPE nginx_ingress_controller_config_last_reload_successful_timestamp_seconds gauge
# HELP nginx_ingress_controller_ssl_certificate_info Hold all labels associated to a certificate
# TYPE nginx_ingress_controller_ssl_certificate_info gauge
# HELP nginx_ingress_controller_success Cumulative number of Ingress controller reload operations
# TYPE nginx_ingress_controller_success counter
# HELP nginx_ingress_controller_orphan_ingress Gauge reporting status of ingress orphanity, 1 indicates orphaned ingress. 'namespace' is the string used to identify namespace of ingress, 'ingress' for ingress name and 'type' for 'no-service' or 'no-endpoint' of orphanity
# TYPE nginx_ingress_controller_orphan_ingress gauge

Admission metrics

# HELP nginx_ingress_controller_admission_config_size The size of the tested configuration
# TYPE nginx_ingress_controller_admission_config_size gauge
# HELP nginx_ingress_controller_admission_render_duration The processing duration of ingresses rendering by the admission controller (float seconds)
# TYPE nginx_ingress_controller_admission_render_duration gauge
# HELP nginx_ingress_controller_admission_render_ingresses The length of ingresses rendered by the admission controller
# TYPE nginx_ingress_controller_admission_render_ingresses gauge
# HELP nginx_ingress_controller_admission_roundtrip_duration The complete duration of the admission controller at the time to process a new event (float seconds)
# TYPE nginx_ingress_controller_admission_roundtrip_duration gauge
# HELP nginx_ingress_controller_admission_tested_duration The processing duration of the admission controller tests (float seconds)
# TYPE nginx_ingress_controller_admission_tested_duration gauge
# HELP nginx_ingress_controller_admission_tested_ingresses The length of ingresses processed by the admission controller
# TYPE nginx_ingress_controller_admission_tested_ingresses gauge

Histogram buckets

You can configure buckets for histogram metrics using these command line options (here are their default values): * --time-buckets=[0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10] * --length-buckets=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100] * --size-buckets=[10, 100, 1000, 10000, 100000, 1e+06, 1e+07]