Monitor your Pachyderm cluster with Prometheus¶

Pachyderm's deployment manifest exposes Prometheus metrics, allowing an easy set up of the monitoring of your cluster. Only available for self-managed deployments today.

Prometheus installation and Service Monitor creation¶

1. Helm install kube-prometheus-stack, Prometheus' Kubernetes cluster monitoring using the Prometheus Operator:

   • Get Repo Info

     helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
     helm repo update
     

   • Install the Prometheus-operator helm chart

     helm install <a-release-name> prometheus-community/kube-prometheus-stack
     

2. Create a ServiceMonitor for Pachyderm in Kubernetes:

   • Create a myprometheusservice.yaml

     apiVersion: monitoring.coreos.com/v1
     kind: ServiceMonitor
     metadata:
       name: pachyderm-scraper
       labels:
         release: <a-release-name>
     spec:
         selector:
             matchLabels:
               suite: pachyderm
         namespaceSelector:
             matchNames:
             - default
         endpoints:
         - port: prom-metrics
           interval: 30s
     

   • Create a ServiceMonitor looking to scrape metrics from suite: pachyderm:

     kubectl create -f myprometheusservice.yaml
     

     The prometheus-operator will search for the pods based on the label selector <a-release-name> and creates a prometheus target so prometheus will scrape the metrics endpoint prom-metrics.

     In this case, it looks for anything with the label suite: pachyderm - which is by default associated with all Pachyderm resources.

   Note

   Our Service Monitor pachyderm-scraper above maps the endpoint port prom-metrics to a corresponding prom-metrics port described in Pachyderm's deployment manifest. Let's take a quick look at this file:

   kubectl -o json get service/pachd
   

   In the json file, find:

       {
       "name": "prom-metrics",
       "port": 1656,
       "protocol": "TCP",
       "targetPort": "prom-metrics"
       }
   

Port-Forward¶

One last step before you can collect your metrics: If you followed the instruction above, you can connect to Prometheus by using kubectl port-forward.

  kubectl port-forward pod/prometheus-<a-release-name>-kube-prometheus-stack-prometheus-0 9090

Browse¶

You can now browse your targets (http://localhost:9090/targets). Run a pipeline of your choice. The pachyderm-scraper should be visible:

In the ClassicUI tab, you should be able to see the new pachyderm metrics.

References¶

• Kube Prometheus Stack documentation.
• Prometheus documentation.