Sonobuoy Plugins

Table of Contents

• Sonobuoy Plugins
  • Overview
  • Plugin Definition
    • Writing your own plugin
      • The plugin definition file
      • Contract
  • Available Plugins

Overview

In addition to querying API objects, Sonobuoy also supports a plugin model. In this model, worker pods are dispatched into the cluster to collect data from each node, and use an aggregation URL to submit their results back to a waiting aggregation pod. See the diagram below:

Two main components specify plugin behavior:

1. Plugin Selection: A section in the main config (config.json) that declares which plugins to use in the Sonobuoy run. This can be generated or passed in with the --config flag to sonobuoy run or sonobuoy gen.

   These configs are defined by the end user.

2. Plugin Definition: A YAML document that defines metadata and a pod to produce a result.

   This YAML is defined by the plugin developer, and can be taken as a given by the end user.

Plugin Definition

• /etc/sonobuoy/plugins.d
• $HOME/.sonobuoy/plugins.d
• ./plugins.d

This search path can be overridden by the PluginSearchPath value of the Sonobuoy config.

Writing your own plugin

The plugin definition file

---
sonobuoy-config:
  driver: Job        # Job or DaemonSet. Job runs once per run, Daemonset runs on every node per run.
  plugin-name: e2e   # The name of the plugin
  result-type: e2e   # The name of the "result type." Usually the name of the plugin.
spec:                # A kubernetes container spec
  env:
  - name: E2E_FOCUS
    value: Pods should be submitted and removed
  image: gcr.io/heptio-images/kube-conformance:latest
  imagePullPolicy: Always
  name: e2e
  volumeMounts:
  - mountPath: /tmp/results
    name: results
    readOnly: false
  - mountPath: /var/log/test
    name: test-volume
extra-volumes:
- name: test-volume
  hostPath:
    # directory location on host
    path: /data

Contract

A definition file defines a container that runs the tests. This container can be anything you want, but must fulfil a contract.

After your container completes its work, it needs to signal to Sonobuoy that it's done. This is done by writing out a filename to a results file. The default value is /tmp/results/done, which you can configure with the ResultsDir value in the Sonobuoy config.

Sonobuoy waits for the done file to be present, then transmits the indicated file back to the aggregator. The results file is opaque to Sonobuoy, and is made available in the Sonobuoy results tarball in its original form.

If you need additional mounts besides the default results mount that Sonobuoy always provides, you can define them in the extra-volumes field.

Choosing which plugins to run

All of the plugin definition files get mounted as files on the aggregator pod which runs them.

The aggregator loads all the plugins it finds, but a separate list controls which plugins actually get run. There is a separate config.json which gets mounted on the aggregator which sets the configuration options for the aggregator. It has a field Plugins which is an array of plugin names. The default value includes both the e2e and systemd-logs plugin:

"Plugins":[{"name":"e2e"},{"name":"systemd-logs"}]

If you want to prevent one of those plugins from being run, simply remove that item from the list. Likewise, if you'd like to run your own custom plugin, you need to add it to this list (in addition to adding its definition file to the plugin configmap):

"Plugins":[{"name":"custom-plugin"},{"name":"systemd-logs"}]

In either case, you use the sonobuoy gen flow to edit the YAML and start the run with kubectl.

Available Plugins

The default Sonobuoy plugins are available in the examples/plugins.d directory in this repository. Here's the current list: