Aporeto integration with Kubernetes Network Policies
TL;DR? Jump to the Getting Started section.
Trireme-Kubernetes is a simple, straightforward implementation of the Kubernetes Network Policies specifications. It is independent from the used networking backend and works in any Kubernetes cluster - even in managed Kubernetes clusters like Google Kubernetes Engine (GKE) or Azure Container Service (AKS).
One of its powerful features is that you can deploy it to multiple Kubernetes clusters and secure network traffic between specific pods of the different clusters (to secure e.g. MySQL replication or a MongoDB replicaset).
Trireme-Kubernetes builds upon a powerful concept of identity based on standard Kubernetes tags.
It is based on the Trireme Zero-Trust library.
More on Kubernetes network policies:
The architecture of Trireme-Kubernetes is the following:
Trireme-Kubernetes consists of several components - not all of them are required:
Trireme-Kubernetes: the enforcement service which polices network connections (a.k.a "flows" in Trireme terminology) based on standard
NetworkPoliciesdefined on the Kubernetes API
Trireme-CSR (optional): an identity service (basically a CA) that is used to automatically sign certificates and generate asymmetric KeyPairs for each Trireme-Kubernetes instance. Note that this is deployed by default. However, you can exchange it to a simple pre-shared key deployment (PSK) if you really wish to do so.
Trireme-Statistics (optional): the monitoring and statistics bundle that currently implements the trireme-lib collector interface for InfluxDB. Flows and Container events can be displayed in either Grafana, Chronograf or Trireme-Graph - which shows a generated graph specifically for Kubernetes network flows between pods. Depending on your use-case, some or all of those frontend tools can be deployed.
IPTableswith access to the
ipsetutility to be installed
DaemonSetmodel (default and recommended), Trireme requires access to the in-cluster Kubernetes service API Token of its pod. Access to the Kubernetes Namespaces/Pods/NetworkPolicies must be available as read-only. NOTE: the default deployment takes care of this.
Trireme-Kubernetes is focused on being simple and straight forward to deploy. NOTE: for any serious deployment, the extensive deployment guide should be followed.
This section provides a quick and easy way to try Trireme-Kubernetes in your existing cluster.
If you are using GKE or another system on which you don't have admin access (for RBAC / ABAC), make sure you can configure additional ABAC / RBAC rules. Specifically on GKE you have to ensure that you have full cluster admin rights through RBAC. You can ensure that you do, by running the following command (replace with your account email address):
kubectl create clusterrolebinding cluster-admin-binding --clusterrole=cluster-admin [email protected]
1) Checkout the deployment files:
git clone https://github.com/aporeto-inc/trireme-kubernetes.git cd trireme-kubernetes/deployment
2) Create the
ConfigMapfrom this configuration file: (keeping everything by default should be fine)
kubectl create -f trireme-config-cm.yaml
3) Optionally, deploy the Trireme-Statistics bundle now (this will deploy all possible frontend options):
kubectl create -f statistics/
4) Create a dummy self-signed Certificate Authority (CA) for Trireme-CSR (the identity service) and add it as a Kubernetes secret (requires the tg utility - quick install:
go get -u github.com/aporeto-inc/tg):
5) Finally, deploy Trireme-CSR and Trireme-Kubernetes:
kubectl create -f trireme/
At this point, the whole framework is up and running and you can access the Services in order to display your NetworkPolicy metrics:
$ kubectl --namespace=kube-system get services
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE chronograf ClusterIP 10.43.241.132 8888/TCP 20h chronograf-public LoadBalancer 10.43.254.222 188.8.131.52 80:32153/TCP 20h grafana ClusterIP 10.43.241.104 3000/TCP 20h grafana-public LoadBalancer 10.43.241.153 184.108.40.206 80:30716/TCP 20h graph ClusterIP 10.43.248.120 8080/TCP 20h graph-public LoadBalancer 10.43.254.146 220.127.116.11 80:31709/TCP 20h influxdb ClusterIP 10.43.245.190 8086/TCP 20h
You can test your setup with NetworkPolicies by using an example two-tier application such as apobeer
git clone https://github.com/aporeto-inc/apobeer cd apobeer/kubernetes kubectl create -f .
The deployed NetworkPolicy allows traffic from
backend, but not from
As a result, streaming your logs on any frontend pod should give you a stream of Beers:
$ kubectl logs frontend-mffv7 -n beer The beer of the day is: "Cantillon Blåbær Lambik" The beer of the day is: "Rochefort Trappistes 10" [...]
And as defined by the policy, only
frontendis able to connect.
externallogs show that it was unable to connect to
$ kubectl logs external-bww23 -n beer
Kubernetes does not enforce natively NetworkPolicies and requires a third party solution/controller to do so.
Unlike most of the traditional solutions, Trireme is not tight together with a complex networking solution. It therefore gives you the freedom to use one Networking implementation if needed and another NetworkPolicy provider. It acts as the controller to enforce the defined Kubernetes network policies.
Trireme-kubernetes does not rely on any distributed control-plane or setup (no need to plug into
etcd). Enforcement is performed directly on every node without any shared state propagation (more info at Trireme )
Trireme-Kubernetes can be deployed as:
DaemonSet. (recommended deployment)