kube-mgmt

Policy-based control for Kubernetes deployments.

About

kube-mgmt manages instances of the Open Policy Agent on top of Kubernetes. Use kube-mgmt to:

Load policies into OPA via Kubernetes (see Policies below.)
Replicate Kubernetes resources including CustomResourceDefinitions (CRDs) into OPA (see Caching below.)

Deployment Guide

Hello World

Create a new Namespace to deploy OPA into:
```
kubectl create namespace opa
```

Create a new Deployment that includes OPA and kube-mgmt (manifests/deployment.yml):

kubectl -n opa create -f https://raw.githubusercontent.com/open-policy-agent/kube-mgmt/master/manifests/deployment.yml

Define a simple policy (example.rego) with the following content:
```
package kubernetes

example = "Hello, Kubernetes!"
```

Create a ConfigMap containing the policy:

kubectl -n opa create configmap hello-world --from-file example.rego

Create a Service to expose OPA:

kubectl -n opa expose deployment opa --type=NodePort

Execute a policy query against OPA:

OPA_URL=$(minikube service -n opa opa --url)
curl $OPA_URL/v1/data/kubernetes/example

Policies

kube-mgmt automatically discovers policies stored in ConfigMaps in Kubernetes and loads them into OPA. kube-mgmt assumes a ConfigMap contains policies if the ConfigMap is:

Created in a namespace listed in the --policies option. If you specify --policies=* then kube-mgmt will look for policies in ALL namespaces.
Labelled with openpolicyagent.org/policy=rego.

When a policy has been successfully loaded into OPA, the openpolicyagent.org/policy-status annotation is set to {"status": "ok"}.

If loading fails for some reason (e.g., because of a parse error), the openpolicyagent.org/policy-status annotation is set to {"status": "error", "error": ...} where the error field contains details about the failure.

JSON Loading

kube-mgmt can be configured to load arbitrary JSON out of ConfigMaps into OPA's data namespace. This is useful for providing contextual information to your policies.

Enable data loading by specifying the --enable-data command-line flag to kube-mgmt. If data loading is enabled kube-mgmt will load JSON out of ConfigMaps labelled with openpolicyagent.org/data=opa.

The JSON data ConfigMaps must be in namespaces listed in the --policies option.

Data loaded out of ConfigMaps is laid out as follows:

<namespace>/<name>/<key>

For example, if the following ConfigMap was created:

kind: ConfigMap
apiVersion: v1
metadata:
  name: hello-data
  namespace: opa
  labels:
    openpolicyagent.org/data: opa
data:
  x.json: |
    {"a": [1,2,3,4]}

Note: "x.json" may be any key.

You could refer to the data inside your policies as follows:

data.opa["hello-data"]["x.json"].a[0]  # evaluates to 1

Note: "opa" is the namespace for the configMap. You may mock this in a test like other objects: with data.opa as my_mocked_object.

Caching

kube-mgmt can be configured to replicate Kubernetes resources into OPA so that you can express policies over an eventually consistent cache of Kubernetes state.

Replication is enabled with the following options:

# Replicate namespace-level resources. May be specified multiple times.
--replicate=<[group/]version/resource>

# Replicate cluster-level resources. May be specified multiple times.
--replicate-cluster=<[group/]version/resource>

Kubernetes resources replicated into OPA are laid out as follows:

<replicate-path>/<resource>/<namespace>/<name> # namespace scoped
<replicate-path>/<resource>/<name>             # cluster scoped

<replicate-path> is configurable (via --replicate-path) and defaults to kubernetes.
<resource> is the Kubernetes resource plural, e.g., nodes, pods, services, etc.
<namespace> is the namespace of the Kubernetes resource.
<name> is the name of the Kubernetes resource.

For example, to search for services with the label "foo" you could write:

some namespace, name
service := data.kubernetes.services[namespace][name]
service.metadata.labels["foo"]

An alternative way to visualize the layout is as single JSON document:

{
	"kubernetes": {
		"services": {
			"default": {
				"example-service": {...},
				"another-service": {...},
				...
			},
			...
		},
		...
}

The example below would replicate Deployments, Services, and Nodes into OPA:

--replicate=apps/v1beta/deployments
--replicate=v1/services
--replicate-cluster=v1/nodes

Custom Resource Definitions (CRDs)

kube-mgmt can also be configured to replicate Kubernetes Custom Resources using the --replicate and --replicate-cluster options. For an example of how OPA can be used to enforce admission control polices on Kubernetes custom resources see Admission Control For Custom Resources

Admission Control

To get started with admission control policy enforcement in Kubernetes 1.9 or later see the Kubernetes Admission Control tutorial. For older versions of Kubernetes, see Admission Control (1.7).

In the Kubernetes Admission Control tutorial, OPA is NOT running with an authorization policy configured and hence clients can read and write policies in OPA. When deploying OPA in an insecure environment, it is recommended to configure authentication and authorization on the OPA daemon. For an example of how OPA can be securely deployed as an admission controller see Admission Control Secure.

Development Guide

To run all of the tests and build the Docker image run make in this directory.

driosalido/kube-mgmt