Insights Architecture

Fairwinds Insights consists of two main components:

• Client - this consists of the users' cluster, where the Insights Agent runs each of the enabled reports
• Server - this consists of the Insights backend, database, and file storage

Report Architectures

Typically each report runs as a CronJob on a configurable schedule (usually once/hour by default). The report will analyze resources in the cluster then send a JSON report with its findings to insights.fairwinds.com.

While each report has its own requirements, there are some baseline requirements needed by the Insights Agent:

• Sufficient RBAC permissions to create new Jobs in the cluster
• The view ClusterRole
• Network access to insights.fairwinds.com

Additional architecture notes and RBAC requirements for different reports are listed below. Note that the Insights Agent only requires egress from the cluster. You will not need to open up access for any kind of network ingress.

Any exceptions or additions to this flow are listed below.

Goldilocks

In addition to the typical cronjob, Goldilocks will run a long lived Deployment for controlling VPA objects. Goldilocks will add a VPA (in recommend mode) to any incoming Deployment in order to provide resource recommendations.

Goldilocks needs access to raw.githubusercontent.com in order to access the CRDs for VPA objects which will be installed in the cluster alongside the Insights Agent.

Goldilocks also requires a metrics-server installation.

kube-bench

kube-bench can run in one of two modes:

• cronjob (default) - this runs the report for a single node
• daemonset - this runs the report on all nodes

If your nodes are homogenous (i.e. you only have a single node pool), the cronjob mode is sufficient. But if your nodes have different configurations, you'll likely want to run kube-bench as a DaemonSet.

kube-bench needs access to some resources on the node in order to perform its audit, including:

• /var/lib
• /etc/systemd
• /etc/kubernetes
• /usr/local/mount-from-host/bin
• root access

kube-hunter

kube-hunter runs in pod mode (opens new window) in order to discover what running a malicious container can discover and do on your cluster.

Nova

Nova needs access to Secrets which contain metadata about installed Helm charts.

Nova will also need network access to any third-party Helm repositories from which Helm charts have been installed (e.g. https://kubernetes-charts.storage.googleapis.com).

Nova will find all Helm charts installed in the cluster then cross-check those charts with the upstream repositories to check for new versions.

OPA

The OPA report will reach out to the Insights API to pull down any Rego policies that have been added to your organization. It will need view access to the cluster in order to run those policies against existing resources.

Pluto

Pluto needs view access to the cluster in order to check for deprecated API versions.

Polaris

Polaris needs view access to the cluster in order to analyze workload configurations. It also needs access to Secrets in order to examine Helm charts.

RBAC Reporter

RBAC Reporter needs view access to the cluster to aggregate information about RBAC roles and bindings.

Trivy

Trivy will download a list of known CVEs from GitHub (opens new window) then cross-check these against images in your cluster.

You may need to grant Trivy access to private images if your nodes don't automatically have access to your private Docker registries.

Workloads

The Workloads report needs view access to your cluster.