그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그 그

  6 minute read  

NetworkPolicys In Garden, Seed, Shoot Clusters

This document describes which Kubernetes NetworkPolicys deployed by Gardener into the various clusters.

Garden Cluster

(via gardener-operator and gardener-resource-manager)

The gardener-operator runs a NetworkPolicy controller which is responsible for the following namespaces:

  • garden
  • istio-system
  • *istio-ingress-*
  • shoot-*
  • extension-* (in case the garden cluster is a seed cluster at the same time)

It deploys the following so-called “general NetworkPolicys”:

NamePurpose
deny-allDenies all ingress and egress traffic for all pods in this namespace. Hence, all traffic must be explicitly allowed.
allow-to-dnsAllows egress traffic from pods labeled with networking.gardener.cloud/to-dns=allowed to DNS pods running in the kube-sytem namespace. In practice, most of the pods performing network egress traffic need this label.
allow-to-runtime-apiserverAllows egress traffic from pods labeled with networking.gardener.cloud/to-runtime-apiserver=allowed to the API server of the runtime cluster.
allow-to-blocked-cidrsAllows egress traffic from pods labeled with networking.gardener.cloud/to-blocked-cidrs=allowed to explicitly blocked addresses configured by human operators (configured via .spec.networking.blockedCIDRs in the Seed). For instance, this can be used to block the cloud provider’s metadata service.
allow-to-public-networksAllows egress traffic from pods labeled with networking.gardener.cloud/to-public-networks=allowed to all public network IPs, except for private networks (RFC1918), carrier-grade NAT (RFC6598), and explicitly blocked addresses configured by human operators for all pods labeled with networking.gardener.cloud/to-public-networks=allowed. In practice, this blocks egress traffic to all networks in the cluster and only allows egress traffic to public IPv4 addresses.
allow-to-private-networksAllows egress traffic from pods labeled with networking.gardener.cloud/to-private-networks=allowed to the private networks (RFC1918) and carrier-grade NAT (RFC6598) except for cluster-specific networks (configured via .spec.networks in the Seed).
allow-to-shoot-networksAllows egress traffic from pods labeled with networking.gardener.cloud/to-shoot-networks=allowed to IPv4 blocks belonging to the shoot networks (configured via .spec.networking in the Shoot). In practice, this should be used by components which use VPN tunnel to communicate to pods in the shoot cluster. Note that this policy only exists in shoot-* namespaces.

Apart from those, the gardener-operator also enables the NetworkPolicy controller of gardener-resource-manager. Please find more information in the linked document. In summary, most of the pods that initiate connections with other pods will have labels with networking.resources.gardener.cloud/ prefixes. This way, they leverage the automatically created NetworkPolicys by the controller. As a result, in most cases no special/custom-crafted NetworkPolicys must be created anymore.

Seed Cluster

(via gardenlet and gardener-resource-manager)

In seed clusters it works the same way as in the garden cluster managed by gardener-operator. When a seed cluster is the garden cluster at the same time, gardenlet does not enable the NetworkPolicy controller (since gardener-operator already runs it). Otherwise, it uses the exact same controller and code like gardener-operator, resulting in the same behaviour in both garden and seed clusters.

Logging & Monitoring

Seed System Namespaces

As part of the seed reconciliation flow, the gardenlet deploys various Prometheus instances into the garden namespace. See also this document for more information. Each pod that should be scraped for metrics by these instances must have a Service which is annotated with

annotations:
  networking.resources.gardener.cloud/from-all-seed-scrape-targets-allowed-ports: '[{"port":<metrics-port-on-pod>,"protocol":"<protocol, typically TCP>"}]'

If the respective pod is not running in the garden namespace, the Service needs these annotations in addition:

annotations:
  networking.resources.gardener.cloud/namespace-selectors: '[{"matchLabels":{"kubernetes.io/metadata.name":"garden"}}]'

If the respective pod is running in an extension-* namespace, the Service needs this annotation in addition:

annotations:
  networking.resources.gardener.cloud/pod-label-selector-namespace-alias: extensions

This automatically allows the needed network traffic from the respective Prometheus pods.

Shoot Namespaces

As part of the shoot reconciliation flow, the gardenlet deploys a shoot-specific Prometheus into the shoot namespace. Each pod that should be scraped for metrics must have a Service which is annotated with

annotations:
  networking.resources.gardener.cloud/from-all-scrape-targets-allowed-ports: '[{"port":<metrics-port-on-pod>,"protocol":"<protocol, typically TCP>"}]'

This automatically allows the network traffic from the Prometheus pod.

Webhook Servers

Components serving webhook handlers that must be reached by kube-apiservers of the virtual garden cluster or shoot clusters just need to annotate their Service as follows:

annotations:
  networking.resources.gardener.cloud/from-all-webhook-targets-allowed-ports: '[{"port":<server-port-on-pod>,"protocol":"<protocol, typically TCP>"}]'

This automatically allows the network traffic from the API server pods.

In case the servers run in a different namespace than the kube-apiservers, the following annotations are needed:

annotations:
  networking.resources.gardener.cloud/from-all-webhook-targets-allowed-ports: '[{"port":<server-port-on-pod>,"protocol":"<protocol, typically TCP>"}]'
  networking.resources.gardener.cloud/pod-label-selector-namespace-alias: extensions
  # for the virtual garden cluster:
  networking.resources.gardener.cloud/namespace-selectors: '[{"matchLabels":{"kubernetes.io/metadata.name":"garden"}}]'
  # for shoot clusters:
  networking.resources.gardener.cloud/namespace-selectors: '[{"matchLabels":{"gardener.cloud/role":"shoot"}}]'

Additional Namespace Coverage in Garden/Seed Cluster

In some cases, garden or seed clusters might run components in dedicated namespaces which are not covered by the controller by default (see list above). Still, it might(/should) be desired to also include such “custom namespaces” into the control of the NetworkPolicy controllers.

In order to do so, human operators can adapt the component configs of gardener-operator or gardenlet by providing label selectors for additional namespaces:

controllers:
  networkPolicy:
    additionalNamespaceSelectors:
    - matchLabels:
        foo: bar

Communication With kube-apiserver For Components In Custom Namespaces

Egress Traffic

Component running in such custom namespaces might need to initiate the communication with the kube-apiservers of the virtual garden cluster or a shoot cluster. In order to achieve this, their custom namespace must be labeled with networking.gardener.cloud/access-target-apiserver=allowed. This will make the NetworkPolicy controllers automatically provisioning the required policies into their namespace.

As a result, the respective component pods just need to be labeled with

  • networking.resources.gardener.cloud/to-garden-virtual-garden-kube-apiserver-tcp-443=allowed (virtual garden cluster)
  • networking.resources.gardener.cloud/to-all-shoots-kube-apiserver-tcp-443=allowed (shoot clusters)

Ingress Traffic

Components running in such custom namespaces might serve webhook handlers that must be reached by the kube-apiservers of the virtual garden cluster or a shoot cluster. In order to achieve this, their Service must be annotated. Please refer to this section for more information.

Shoot Cluster

(via gardenlet)

For shoot clusters, the concepts mentioned above don’t apply and are not enabled. Instead, gardenlet only deploys a few “custom” NetworkPolicys for the shoot system components running in the kube-system namespace. All other namespaces in the shoot cluster do not contain network policies deployed by gardenlet.

As a best practice, every pod deployed into the kube-system namespace should use appropriate NetworkPolicy in order to only allow required network traffic. Therefore, pods should have labels matching to the selectors of the available network policies.

gardenlet deploys the following NetworkPolicys:

NAME                                       POD-SELECTOR
gardener.cloud--allow-dns                  k8s-app in (kube-dns)
gardener.cloud--allow-from-seed            networking.gardener.cloud/from-seed=allowed
gardener.cloud--allow-to-dns               networking.gardener.cloud/to-dns=allowed
gardener.cloud--allow-to-apiserver         networking.gardener.cloud/to-apiserver=allowed
gardener.cloud--allow-to-from-nginx        app=nginx-ingress
gardener.cloud--allow-to-kubelet           networking.gardener.cloud/to-kubelet=allowed
gardener.cloud--allow-to-public-networks   networking.gardener.cloud/to-public-networks=allowed
gardener.cloud--allow-vpn                  app=vpn-shoot

Note that a deny-all policy will not be created by gardenlet. Shoot owners can create it manually if needed/desired. Above listed NetworkPolicys ensure that the traffic for the shoot system components is allowed in case such deny-all policies is created.

Implications for Gardener Extensions

Gardener extensions sometimes need to deploy additional components into the shoot namespace in the seed cluster hosting the control plane. For example, the gardener-extension-provider-aws deploys the cloud-controller-manager into the shoot namespace. In most cases, such pods require network policy labels to allow the traffic they are initiating.

For components deployed in the kube-system namespace of the shoots (e.g., CNI plugins or CSI drivers, etc.), custom NetworkPolicys might be required to ensure the respective components can still communicate in case the user creates a deny-all policy.