Another Tutorial

A short lead description about this content page. It can be bold or italic and can be split over multiple paragraphs.

This is a placeholder page. Replace it with your own content.

CI GitHub release Go Report Card Artifact Hub License: MIT Last Commit doc.crds.dev

log2rbac-operator

Kubernetes operator that helps you to set up the RBAC rules for your application. If requested, it scans the application’s log files for authorization errors and adds them as exceptions/rights to the associated Role. It is like having a sudo command for your service accounts. However, with great power comes great responsibility. The goal of the tool is to find the minimum set of rights that is needed for your workload to run instead of using the cluster admin for everything.

User has to allow this process by creating a RbacNegotiation custom resource where they need to specify the app* and Role. Role can be either existing one or operator can create a new one for you and bind it to the service account that’s configured with the deployment. Again if the service account is not there, it will be created by the operator.

* App can be one of the following:

  • Deployment
  • StatefulSet
  • DaemonSet
  • Service
  • ReplicaSet
  • or key-value pair specifying the pod selector

By creating RbacNegotiation we start the “recording” regime for our workload in which the operator will be building this allow list of verbs, resources (rbac entries) for the given role.

Watch the full asciicast (pauseable demo)

This project is conceptually very similar to audit2rbac. The main distinction here is that log2rbac is based on the controller pattern and on the output from the workloads, while audit2rbac uses the k8s’ audit log and it’s a “one-pass” CLI tool.

Quick Start

# clone repo and
make deploy

alternatively install it using all-in-one yaml

kubectl apply -f http://bit.do/log2rbac

or helm:

helm repo add log2rbac https://jkremser.github.io/log2rbac-operator
helm repo update
helm -n log2rbac upgrade -i log2rbac log2rbac/log2rbac-operator --wait --create-namespace

Now when the operator was installed, let’s deploy something that needs the special RBAC. Prometheus Operator is a good example.

kubectl create ns monitoring
kubectl apply -n monitoring -f https://github.com/prometheus-operator/kube-prometheus/raw/v0.10.0/manifests/prometheusOperator-deployment.yaml

This deployment will fail to start because of the missing rights to do its stuff. Let’s request the RBAC negotiation process.

# create RbacNegotiation for Prometheus operator
cat <<CustomResource | kubectl apply -f -
apiVersion: kremser.dev/v1
kind: RbacNegotiation
metadata:
  name: for-prom
spec:
  for:
    namespace: monitoring
    kind: Deployment
    name: prometheus-operator
  role:
    name: foo
    isClusterRole: true
    createIfNotExist: true
CustomResource
# After some time, the Prometheus Operator should start and we should see.

k describe clusterrole foo
Name:         foo
Labels:       <none>
Annotations:  app.kubernetes.io/created-by=log2rbac
PolicyRule:
  Resources                                  Non-Resource URLs  Resource Names  Verbs
  ---------                                  -----------------  --------------  -----
  configmaps                                 []                 []              [list watch]
  namespaces                                 []                 []              [list watch]
  secrets                                    []                 []              [list watch]
  statefulsets.apps                          []                 []              [list watch]
  alertmanagerconfigs.monitoring.coreos.com  []                 []              [list]
  alertmanagers.monitoring.coreos.com        []                 []              [list]
  ...

Note: This set of rights was necessary only for the Prometheus operator to start. Once we start interacting with the subject of our RBAC negotiation process, the new code paths will be executed and possibly new rights will be requested. It might be a good idea to run e2e tests or at least have some script that calls the Prometheus operator’s functionality (CRUDing all its CRDs). Last but not least, once we are happy with the resulting role and the rights it got, we should delete the RBAC negotiation custom resource to reduce the attack surface on our cluster.

You may want to capture the role in yaml format and store it as part of your infrastructure code in git:

k get clusterrole foo -o yaml | k neat

Kubectl Plugin

Installation:

kubectl krew install log2rbac

It can help with creating those RbacNegotiation custom resources by interactive TUI api.

It’s located in this repo

Observability

Operator’s code has been instrumented by ~OpenTracing~ OpenTelemetry calls so that one can export the spans to Jaeger or Zipkin and connect the dots. There is an example deployment using open telemetry collector running as a side-car container that exports the traces to Jaeger that runs in its own deployment. To deploy this scenario, issue:

make deploy-otel

To check the Jaeger web UI for log2rbac traces, the easiest way is to

kubectl port-forward svc/jaeger-collector 16686
open http://localhost:16686

Jaeger + log2rbac screenshot

Configuration

Following options are available as env variables for the operator:

Variable name Description default value
COLORS whether the colorful output in logs should be used true
NO_BANNER whether no ascii art should be printed during start false
SYNC_INTERVAL_AFTER_NO_RBAC_ENTRY_MINUTES if no rbac related entry was found in logs, how long to wait for the next check 5
SYNC_INTERVAL_AFTER_NO_LOGS_SECONDS if it was not possible to get the logs, how long to wait for the next check 30
SYNC_INTERVAL_AFTER_POD_RESTART_SECONDS how long to wait after rbac entry was added and pod was restarted by the operator 20
SHOULD_RESTART_APP_PODS whether the operator should be restarting the pods after modifying the role true
TRACING_ENABLED if the application should be sending the traces to OTLP collector false
OTEL_EXPORTER_OTLP_ENDPOINT host:port where the spans (traces) should be sent localhost:4318
TRACING_SAMPLING_RATIO float64 representing the ratio how often the span should be kept/dropped AlwaysSample ~ 1.0

Last modified October 25, 2022: web (8c0fc88)