Prometheus Rules | DOCA Platform Framework

DPF Prometheus Rules

Introduction

This page introduces two sets of Prometheus rules, alert rules and recording rules, that complement the Grafana dashboards shipped with the dpf-operator Helm chart. DPF does not ship these rules as part of the Helm chart. They are reference PrometheusRule manifests living in deploy/helmfiles/prometheus-rules/, deployed alongside the kube-prometheus-stack install: the deploy/helmfiles/monitoring.yaml helmfile applies them automatically, or apply them to any Prometheus Operator deployment manually with kubectl apply --server-side -f deploy/helmfiles/prometheus-rules/.

Where to place rules

The manifests are PrometheusRule custom resources (monitoring.coreos.com/v1) that must land in a namespace your Prometheus instance watches. If you use the reference DPF install that means dpf-operator-system. A minimal skeleton looks like:

YAML

apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: <group-name>
  namespace: dpf-operator-system
  labels:
    app.kubernetes.io/part-of: doca-platform-framework
    # Required if your Prometheus selects rules by a specific label. The default
    # kube-prometheus-stack install uses `release=kube-prometheus-stack`.
    release: kube-prometheus-stack
spec:
  groups:
    - name: <group-name>
      rules: []

If the rules do not load, check your Prometheus instance's ruleSelector and ruleNamespaceSelector. The PrometheusRule labels must match the selector your Prometheus is configured with, otherwise the Prometheus Operator will silently skip the resource.

Naming conventions

Alert names: CamelCase, prefixed with DPF so the whole set is filterable by name (DPFDPUNotReady, DPFDPUClusterAPIServerLatencyHigh, DPFOperatorPodNotReady).
Recording-rule names: lowercase, using the Prometheus community <level>:<metric>:<operation> shape, namespaced under dpf: (dpf:dpu:ready, dpf:dpucluster:apiserver_request_duration:p99).
Labels: every rule carries severity: critical|warning and service: doca-platform-framework, so alerts and series can be filtered consistently in Prometheus, Grafana, and Alertmanager routes (ALERTS{service="doca-platform-framework"}, dpf:*{service="doca-platform-framework"}).

Severity policy

Severity	Use when
`critical`	Service is impacted, pages someone, requires immediate action.
`warning`	Degraded state or stuck reconcile, but no immediate user impact.

Documentation

Last updated: October 14, 2026