{% hint style="info" title="feature differences" %} network policies and istio authorization is only applied in GCP clusters. {% endhint %}
Access policies express which applications and services you are able to communicate with, both inbound and outbound. The default policy is to deny all incoming and outgoing traffic for your application, meaning you must be conscious of which services/application you consume, and who your consumers are.
Inbound rules specifies what other applications in the same cluster your application receives traffic from.
For app app-a
to be able to receive incoming requests from app-b
in the same cluster and the same namespace, this specification is needed for app-a
:
apiVersion: "nais.io/v1alpha1"
kind: "Application"
metadata:
name: app-a
...
spec:
...
accessPolicy:
inbound:
rules:
- application: app-b
For app app-a
to be able to receive incoming requests from app-b
in the same cluster but another namespace (othernamespace
), this specification is needed for app-a
:
apiVersion: "nais.io/v1alpha1"
kind: "Application"
metadata:
name: app-a
...
spec:
...
accessPolicy:
inbound:
rules:
- application: app-b
namespace: othernamespace
Inbound rules specifies what other applications your application receives traffic from. spec.accessPolicy.outbound.rules
specifies which applications in the same cluster to open for. To open for external applications, use the field spec.accessPolicy.outbound.external
.
For app app-a
to be able to send requests to app-b
in the same cluster and the same namespace, this specification is needed for app-a
:
apiVersion: "nais.io/v1alpha1"
kind: "Application"
metadata:
name: app-a
...
spec:
...
accessPolicy:
outbound:
rules:
- application: app-b
For app app-a
to be able to send requests requests to app-b
in the same cluster but in another namespace (othernamespace
), this specification is needed for app-a
:
apiVersion: "nais.io/v1alpha1"
kind: "Application"
metadata:
name: app-a
...
spec:
...
accessPolicy:
outbound:
rules:
- application: app-b
namespace: othernamespace
In order to send requests to services outside of the cluster, external.host
is needed:
apiVersion: "nais.io/v1alpha1"
kind: "Application"
metadata:
name: app-a
...
spec:
...
accessPolicy:
outbound:
external:
- host: www.external-application.com
The previous application manifest examples will create both Kubernetes Network Policies and Istio resources.
These requirements must be met for access policies to be working as expected:
- The cluster must have Istio installed
- The cluster must have network policies enabled
- Naiserator must run with the flag
--access-policies=true
Every app created will have this default network policy that allows traffic from Istio pilot and mixer, as well as kube-dns. This policy will be created for every app, also those who don't have any access policies specified.
apiVersion: extensions/v1beta1
kind: NetworkPolicy
metadata:
labels:
app: appname
team: teamname
name: appname
namespace: teamname
spec:
egress:
- to:
- namespaceSelector:
matchLabels:
name: istio-system
podSelector:
matchLabels:
istio: pilot
- namespaceSelector:
matchLabels:
name: istio-system
podSelector:
matchLabels:
istio: mixer
- namespaceSelector: {}
podSelector:
matchLabels:
k8s-app: kube-dns
- ipBlock:
cidr: 0.0.0.0/0
except:
- 10.0.0.0/8
- 172.16.0.0/12
- 192.168.0.0/16
podSelector:
matchLabels:
app: appname
policyTypes:
- Egress
The applications specified in spec.accessPolicy.inbound.rules
and spec.accessPolicy.outbound.rules
will append these fields to the default Network Policy:
apiVersion: extensions/v1beta1
kind: NetworkPolicy
...
spec:
egress:
- to:
...
- namespaceSelector:
matchLabels:
name: othernamespace
podSelector:
matchLabels:
app: app-b
- podSelector:
matchLabels:
app: app-b
- from:
- namespaceSelector:
matchLabels:
name: othernamespace
podSelector:
matchLabels:
app: app-b
- podSelector:
matchLabels:
app: app-b
podSelector:
matchLabels:
app: appname
policyTypes:
- Egress
- Ingress
{% hint style="info" %}
Note that for namespace match labels to work, the namespaces must be labeled with name: namespacename
.
kube-system
should be labeled accordingly for the default rule that allows traffic to kube-dns
, but in GCP, the label is removed by some job in regular intervals...
{% endhint %}
The policies from spec.accessPolicy
will in addition create these Istio-resources:
For Istio to allow request from app-b
in the same namespace and in othernamespace
, these resources will be created:
apiVersion: rbac.istio.io/v1alpha1
kind: ServiceRole
metadata:
labels:
app: app-a
team: my-team
name: app-a
namespace: my-team
spec:
rules:
- methods:
- '*'
paths:
- '*'
services:
- app-a.my-team.svc.cluster.local
apiVersion: rbac.istio.io/v1alpha1
kind: ServiceRoleBinding
metadata:
labels:
app: app-a
team: my-team
name: app-a
namespace: my-team
spec:
roleRef:
kind: ServiceRole
name: app-a
subjects:
- user: cluster.local/ns/my-team/sa/app-b
- user: cluster.local/ns/othernamespace/sa/app-b
spec.accessRules.outbound.external
will create ServiceEntry:
apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
labels:
app: app-a
team: my-team
name: app-a
namespace: my-team
spec:
hosts:
- www.external-application.com
location: MESH_EXTERNAL
ports:
- name: https
number: 443
protocol: HTTPS
resolution: DNS
In the cloud spec.ingresses
will create VirtualService instead of Ingress objects:
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
labels:
app: app-a
team: my-team
name: app-a-app-a-dev-gcp-nais-io
namespace: my-team
spec:
gateways:
- istio-system/ingress-gateway-nais-io
hosts:
- my-app.dev-gcp.nais.io
http:
- route:
- destination:
host: app-a
port:
number: 80
subset: ""
weight: 100