A small DPDK telemetry exporter

The recommended way when running locally. Remember to mount the dpdk run dir as a volume and add extra options to the command line, e.g.:

$ docker run --rm --name exporter -p 8000:8000 -v /var/run/dpdk/:/var/run/dpdk/ nfvri/dpdk-telemetry-exporter:0.1 dpdkTelemetryExporter -vvv -T 5

To run as a sidecar, add the exporter container to your Deployment/Statefulset/Daemonset definition with mount access to the dpdk run directory (usually /var/run/dpdk) as follows:

apiVersion: apps/v1
kind: Deployment
...
spec:
  template:
    spec:
      containers:
      - name: telemetry-exporter
        imagePullPolicy: Always
        image: nfvri/dpdk-telemetry-exporter:0.1
        command: ["dpdkTelemetryExporter"]
        args: ["-vvv"]
        volumeMounts:
           - mountPath: /var/run/dpdk/
             name: dpdkrun-volume
        resources:
          requests:
            memory: "1Gi"
            cpu: "1000m"
          limits:
            memory: "1Gi"
            cpu: "1000m"
        ports:
          - containerPort: 8000
...

Then assuming you have a Prometheus-operator deployment, use a Service and ServiceMonitor to specify a target to the exporter (be careful to match the appropriate labels/namespaces for your case):

---
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: dpdk-deployment-monitor
  namespace: monitoring
  labels:
    app: dpdk
    release: k8s-prom
spec:
  endpoints:
  - port: metrics
    path: /
    interval: "5s"
    scrapeTimeout: "5s"
  namespaceSelector:
    matchNames:
      - dpdk
  selector:
    matchLabels:
      app: dpdk

---
apiVersion: v1
kind: Service
metadata:
  name: dpdk-deployment-svc
  namespace: dpdk
  labels:
    app: dpdk
spec:
  ports:
  - name: metrics
    port: 8000
    protocol: TCP
  selector:
    app: dpdk

Please prefer to run from the docker image. If local installation is absolutely necessary, you can install the exporter with:

$ sudo apt-get update && sudo apt-get install -y python3 python3-pip

$ python3 setup.py install

You can then run it with:

$ dpdkTelemetryExporter -h
usage: DPDKTelemetryExporter [-h] [-t THREADS] [-p PORT] [-T TIMEOUT] [-v]

optional arguments:
  -h, --help            show this help message and exit
  -t THREADS, --threads THREADS
                        DPDKTelemetryExporter parallel threads (default: 8)
  -p PORT, --port PORT  DPDKTelemetryExporter port (default: 8000)
  -T TIMEOUT, --timeout TIMEOUT
                        The update interval in seconds (default 5)
  -v, --verbose         Set output verbosity (default: -vv = INFO)

If you have set the dpdk run directory to an "odd" (i.e. not /var/run/dpdk) location, you can specify it by setting the DPDK_RUN_DIR environment variable to make the telemetry socket discoverable by the exporter.

The exporter understands (and re-wraps to proper Prometheus types) the following metrics (if available):

A description of the meaning of label names follows. All label values as per the Prometheus spec are strings.

In the example above for Kubernetes pod sidecar run, the Prometheus target is set automatically by prometheus-operator. If you have to manually create a target, locate your prometheus.yml file and add a scrape config for the exporter target, using the proper ip and port where Prometheus can contact the exporter:

  - job_name: 'dpdk-telemetry-exporter'
    scrape_interval: 5s
    scrape_timeout: 5s
    static_configs:
      - targets: ['192.168.123.1:8000']

A sample grafana dashboard is provided at grafana_dashboard.json. This is primarily meant for a Kubernetes Prometheus, so it assumes that there is a pod label available in the panel queries. Feel free to change it to the instance label if not running Prometheus in Kubernetes.