New to KubeDB? Please start here.

PgBouncer Grafana Dashboard

KubeDB exposes PgBouncer metrics through a sidecar exporter, and its own view of each resource (status, phase, version) through Panopticon. Once Prometheus is scraping both, you can visualize them in Grafana using a pre-built KubeDB dashboard. This tutorial walks through the full setup: deploying the monitoring stack, enabling monitoring on a PgBouncer instance, and importing the Grafana dashboard.

Before You Begin

• You need a Kubernetes cluster with kubectl configured. If you do not already have a cluster, you can create one by using kind.

• KubeDB must be installed in your cluster with kubedb-metrics enabled. Follow the setup guide here and make sure to include the flag below during installation:

  --set kubedb-metrics.enabled=true
  

  kubedb-metrics creates MetricsConfiguration objects for each database type, which Panopticon (see Configuration) uses to expose metrics to Prometheus.

• PgBouncer sits in front of a PostgreSQL server. Prepare a KubeDB Postgres instance (for example ha-postgres in the demo namespace) following the streaming replication guide, or use any externally managed Postgres.

• To keep monitoring resources isolated, we use a separate monitoring namespace and deploy the database in the demo namespace.

  $ kubectl create ns monitoring
  namespace/monitoring created
  
  $ kubectl create ns demo
  namespace/demo created
  

• Before proceeding, complete the Configuration steps to deploy kube-prometheus-stack and Panopticon.

Note: YAML files used in this tutorial are stored in docs/examples/pgbouncer/monitoring folder in GitHub repository kubedb/docs.

Setup

Step 1: Deploy PgBouncer

Below is the PgBouncer object pointing at the ha-postgres backend, with monitoring configured to use Prometheus Operator.

apiVersion: kubedb.com/v1
kind: PgBouncer
metadata:
  name: pb-grafana-demo
  namespace: demo
spec:
  replicas: 1
  version: "1.24.0"
  database:
    syncUsers: true
    databaseName: "postgres"
    databaseRef:
      name: "ha-postgres"
      namespace: demo
  connectionPool:
    poolMode: session
    port: 5432
  deletionPolicy: WipeOut
  monitor:
    agent: prometheus.io/operator
    prometheus:
      serviceMonitor:
        labels:
          release: prometheus
        interval: 10s

Here,

• database.databaseRef points at the backend Postgres instance PgBouncer pools connections for.
• monitor.agent: prometheus.io/operator tells KubeDB to create a ServiceMonitor for this instance.
• monitor.prometheus.serviceMonitor.labels must match the serviceMonitorSelector label of your Prometheus (release: prometheus).
• monitor.prometheus.serviceMonitor.interval sets the scrape interval to 10 seconds.

Create the PgBouncer instance:

$ kubectl create -f https://github.com/kubedb/docs/raw/v2026.7.10/docs/examples/pgbouncer/monitoring/pb-grafana-demo.yaml
pgbouncer.kubedb.com/pb-grafana-demo created

Wait for it to be Ready:

$ kubectl get pb -n demo pb-grafana-demo
NAME             TYPE           VERSION   STATUS   AGE
pb-grafana-demo  kubedb.com/v1  1.24.0    Ready    65s

KubeDB creates a stats service named {pgbouncer-name}-stats for the exporter:

$ kubectl get svc -n demo --selector="app.kubernetes.io/instance=pb-grafana-demo"
NAME                  TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)             AGE
pb-grafana-demo       ClusterIP   10.96.201.180   <none>        5432/TCP            2m
pb-grafana-demo-pods  ClusterIP   None            <none>        5432/TCP            2m
pb-grafana-demo-stats ClusterIP   10.96.73.22     <none>        9719/TCP            2m

KubeDB also creates a ServiceMonitor in the demo namespace:

$ kubectl get servicemonitor -n demo
NAME                  AGE
pb-grafana-demo-stats 2m

Verify it carries the correct label:

$ kubectl get servicemonitor -n demo pb-grafana-demo-stats -o jsonpath='{.metadata.labels}'
{"release":"prometheus", ...}

Step 2: Verify Prometheus is Scraping

Port-forward the Prometheus pod:

$ kubectl port-forward -n monitoring \
  prometheus-prometheus-kube-prometheus-prometheus-0 9090
Forwarding from 127.0.0.1:9090 -> 9090
Forwarding from [::1]:9090 -> 9090

Open http://localhost:9090/targets in your browser. Look for an entry whose service label matches pb-grafana-demo-stats. Its state should be UP.

If the target is missing, check that the ServiceMonitor label (release: prometheus) matches the Prometheus serviceMonitorSelector.

Step 3: Access Grafana

Port-forward the Grafana service:

$ kubectl port-forward -n monitoring svc/prometheus-grafana 3000:80
Forwarding from 127.0.0.1:3000 -> 80

Open http://localhost:3000. The username is admin. Retrieve the auto-generated password from the secret:

$ kubectl get secret -n monitoring prometheus-grafana \
  -o jsonpath='{.data.admin-password}' | base64 -d

After a successful login you will see the Grafana home page:

Step 4: Configure Prometheus as a Data Source

If you installed Grafana via kube-prometheus-stack, Prometheus is already configured as the default data source — skip to Step 5.

If you’re using a different Grafana instance than the one installed in the Configuration prerequisite (linked in “Before You Begin” above), add Prometheus as a data source manually:

1. Go to Connections → Data sources → Add new data source.

2. Select Prometheus.

3. Set the URL to your Prometheus service:

   http://prometheus-operated.monitoring.svc:9090
   

4. Click Save & test. You should see Data source is working.

Step 5: Import Dashboard — Option A: Automatic (chart)

Rather than downloading and uploading each JSON file by hand (Option B below), KubeDB ships a chart that creates all matching dashboards for you as GrafanaDashboard custom resources. A separate controller, grafana-operator, watches these resources and pushes the actual dashboard JSON into your Grafana instance — both pieces are required.

1. Install grafana-operator (skip if it’s already running in your cluster):

$ helm repo add appscode https://charts.appscode.com/stable/
$ helm repo update

$ helm upgrade --install grafana-operator appscode/grafana-operator \
    --version v2026.6.12 \
    --namespace kubeops --create-namespace

2. Register your Grafana instance as an AppBinding (skip if you’ve already done this in another guide on this cluster). grafana-operator needs to know where to push dashboards and how to authenticate — it reads this from an AppBinding object, not from the chart install command itself. Since Grafana here came bundled with kube-prometheus-stack, reuse its existing admin credentials:

$ kubectl port-forward -n monitoring svc/prometheus-grafana 3000:80 &

$ curl -s -X POST -H "Content-Type: application/json" \
    -u admin:<grafana_password> \
    http://localhost:3000/api/auth/keys \
    -d '{"name":"kubedb-dashboards","role":"Admin"}'
# Note the returned "key"

$ kill %1

$ kubectl create secret generic grafana-admin-token -n monitoring \
    --from-literal=token='<key-from-above>'

$ cat <<EOF | kubectl apply -f -
apiVersion: appcatalog.appscode.com/v1alpha1
kind: AppBinding
metadata:
  name: grafana
  namespace: monitoring
spec:
  type: monitoring.appscode.com/grafana
  clientConfig:
    url: http://prometheus-grafana.monitoring.svc:80
  secret:
    name: grafana-admin-token
EOF

3. Install the dashboards:

The chart packages dashboard JSON for every database it supports, so install it from a copy trimmed down to just PgBouncer’s dashboards (this also keeps grafana-operator from creating dashboards for databases you don’t run):

$ helm pull appscode/kubedb-grafana-dashboards --version v2026.7.10 --untar

$ cd kubedb-grafana-dashboards/dashboards
$ ls | grep -v '^pgbouncer$' | xargs rm -rf   # keep only dashboards/pgbouncer
$ cd ../..

$ helm package kubedb-grafana-dashboards
Successfully packaged chart and saved it to: kubedb-grafana-dashboards-v2026.7.10.tgz

$ helm upgrade -i kubedb-grafana-dashboards-pgbouncer ./kubedb-grafana-dashboards-v2026.7.10.tgz \
    -n kubeops --create-namespace \
    --set grafana.name=grafana \
    --set grafana.namespace=monitoring

Use a release name unique to this database (kubedb-grafana-dashboards-pgbouncer), not the plain kubedb-grafana-dashboards name — if you also follow another DB’s Grafana Dashboard guide on this same cluster, each helm upgrade -i under a shared release name would prune the previous DB’s dashboards (Helm removes anything not in the new release’s manifest). A per-DB release name lets them coexist.

grafana.name/grafana.namespace point the chart’s GrafanaDashboard resources at the AppBinding created in step 2 (omitting them falls back to whichever AppBinding in your cluster is labeled as the cluster-default Grafana, if any — explicit is safer on a shared cluster). No need to touch featureGates — with every other database’s dashboards/ folder removed, their gates simply match zero files and render nothing, regardless of being true by default.

This creates every dashboard the chart ships for PgBouncer — KubeDB / PgBouncer / Summary, KubeDB / PgBouncer / Pod, KubeDB / PgBouncer / Database — which grafana-operator then pushes into your Grafana instance automatically. No manual JSON download or upload needed.

Verify they landed:

$ kubectl get grafanadashboards.openviz.dev -n kubeops | grep -i pgbouncer
NAME                                 TITLE                            SYNCED    AGE
grafana-kubedb-pgbouncer-summary     KubeDB / PgBouncer / Summary     Current   30s
grafana-kubedb-pgbouncer-pod         KubeDB / PgBouncer / Pod         Current   30s
grafana-kubedb-pgbouncer-database    KubeDB / PgBouncer / Database    Current   30s

SYNCED: Current confirms grafana-operator successfully pushed each dashboard into Grafana. Open Grafana — the dashboard are already there under Dashboards, fully wired to your Prometheus data source, ready to explore in Step 6 below.

If the SYNCED column is missing entirely from your output (not just showing a non-Current value), grafana-operator most likely never processed the resource at all. Check that the operator pod is actually running (kubectl get pods -n kubeops -l app.kubernetes.io/name=grafana-operator), inspect its logs for AppBinding/auth errors (kubectl logs -n kubeops deploy/grafana-operator), and check kubectl get grafanadashboards.openviz.dev -n kubeops <name> -o yaml for status.conditions — the CR existing only means kubectl accepted it, not that it reached Grafana.

After importing them, they will appear under Dashboards in the left sidebar as well:

Step 5: Import Dashboard — Option B: Manual (JSON upload)

If you’d rather not run grafana-operator, or want fine-grained control over exactly which dashboards get imported, upload the same dashboard JSON files by hand instead.

The KubeDB PgBouncer dashboards are distributed as JSON files. Each JSON file is a complete dashboard definition — panels, queries, variables, and layout — that Grafana loads in one shot. Without importing, you would have to build every panel and write every PromQL query by hand. Importing lets you skip that entirely.

Three dashboards are available. Download all three JSON files from the appscode/grafana-dashboards repository (pgbouncer/ folder):

Import steps (repeat for each of the three files):

1. In Grafana, click the + icon in the left sidebar.
2. Select Import from the menu.
3. Click Upload JSON file and select one of the downloaded .json files.
4. In the Prometheus dropdown that appears, select your Prometheus data source.
5. Click Import.

The import page looks like this — click Upload dashboard JSON file to select the file:

After importing all three files, they will appear under Dashboards in the left sidebar.

Step 6: Explore the Dashboard

After opening a dashboard, you will see dropdown filters at the top. These control which data is shown across all panels — change them to focus on a specific instance without editing any queries.

Once you set these, all panels update automatically. Below is what each dashboard shows:

KubeDB / PgBouncer / Summary — start here for a pool-wide overview:

• Total / Client / Server Connections — connections across all pools
• Active Pools — number of connection pools to the backend
• Queries per Second — query throughput routed through PgBouncer
• Waiting Clients — clients queued waiting for a server connection (sustained non-zero indicates pool exhaustion)
• Avg Query Time — average time queries spend waiting + executing
• CPU / Memory / Storage — resource consumption vs. requests and limits

KubeDB / PgBouncer / Pod — drill into a specific pod:

• Client / Server Connections — connections held on this pod
• Active Pools — pools managed by this pod
• Queries per Second — per-pod query throughput
• Waiting Clients — queued clients on this pod
• CPU / Memory — per-pod resource usage

KubeDB / PgBouncer / Database — per-database pool metrics:

• Pool State — cl_active, cl_waiting, sv_active, sv_idle per database
• Active / Idle Server Connections — backend connection usage per database
• Max Client Connections — configured limit per database
• Query Rate — queries per second per database
• Total Xact / Query Count — transaction and query counters per database

Cleaning up

# Remove the PgBouncer instance
kubectl delete pb -n demo pb-grafana-demo

# Remove the backend Postgres instance
kubectl delete pg -n demo ha-postgres

# Remove namespaces
kubectl delete ns demo

# Uninstall the Grafana dashboards chart, if you used Option A
helm uninstall kubedb-grafana-dashboards-pgbouncer -n kubeops

# Uninstall grafana-operator (optional — skip if other DB guides on this cluster still use it)
helm uninstall grafana-operator -n kubeops

# Uninstall monitoring stack (optional)
helm uninstall prometheus -n monitoring
helm uninstall panopticon -n kubeops
kubectl delete ns monitoring kubeops

Next Steps

• Monitor your PgBouncer database with KubeDB using built-in Prometheus.
• Monitor your PgBouncer database with KubeDB using Prometheus Operator.
• Want to hack on KubeDB? Check our contribution guidelines.