Install Multiple Istio Control Planes in a Single Cluster
The following information describes an experimental feature, which is intended for evaluation purposes only.
This guide walks you through the process of installing multiple Istio control planes within a single cluster and then a way to scope workloads to specific control planes. This deployment model has a single Kubernetes control plane with multiple Istio control planes and meshes. The separation between the meshes is provided by Kubernetes namespaces and RBAC.
Multiple meshes in a single cluster
Using discoverySelectors
, you can scope Kubernetes resources in a cluster to specific namespaces managed by an Istio control plane. This includes the Istio custom resources (e.g., Gateway, VirtualService, DestinationRule, etc.) used to configure the mesh. Furthermore, discoverySelectors
can be used to configure which namespaces should include the istio-ca-root-cert
config map for a particular Istio control plane. Together, these functions allow mesh operators to specify the namespaces for a given control plane, enabling soft multi-tenancy for multiple meshes based on the boundary of one or more namespaces. This guide uses discoverySelectors
, along with the revisions capability of Istio, to demonstrate how two meshes can be deployed on a single cluster, each working with a properly scoped subset of the cluster’s resources.
Before you begin
This guide requires that you have a Kubernetes cluster with any of the supported Kubernetes versions: 1.24, 1.25, 1.26, 1.27.
This cluster will host two control planes installed in two different system namespaces. The mesh application workloads will run in multiple application-specific namespaces, each namespace associated with one or the other control plane based on revision and discovery selector configurations.
Cluster configuration
Deploying multiple control planes
Deploying multiple Istio control planes on a single cluster can be achieved by using different system namespaces for each control plane. Istio revisions and discoverySelectors
are then used to scope the resources and workloads that are managed by each control plane.
By default, Istio only uses discoverySelectors
to scope workload endpoints. To enable full resource scoping, including configuration resources, the feature flag ENABLE_ENHANCED_RESOURCE_SCOPING
must be set to true.
Create the first system namespace,
usergroup-1
, and deploy istiod in it:$ kubectl create ns usergroup-1
$ kubectl label ns usergroup-1 usergroup=usergroup-1
$ istioctl install -y -f - <<EOF
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
metadata:
namespace: usergroup-1
spec:
profile: minimal
revision: usergroup-1
meshConfig:
discoverySelectors:
- matchLabels:
usergroup: usergroup-1
values:
global:
istioNamespace: usergroup-1
pilot:
env:
ENABLE_ENHANCED_RESOURCE_SCOPING: true
EOF
Create the second system namespace,
usergroup-2
, and deploy istiod in it:$ kubectl create ns usergroup-2
$ kubectl label ns usergroup-2 usergroup=usergroup-2
$ istioctl install -y -f - <<EOF
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
metadata:
namespace: usergroup-2
spec:
profile: minimal
revision: usergroup-2
meshConfig:
discoverySelectors:
- matchLabels:
usergroup: usergroup-2
values:
global:
istioNamespace: usergroup-2
pilot:
env:
ENABLE_ENHANCED_RESOURCE_SCOPING: true
EOF
Deploy a policy for workloads in the
usergroup-1
namespace to only accept mutual TLS traffic:$ kubectl apply -f - <<EOF
apiVersion: security.istio.io/v1beta1
kind: PeerAuthentication
metadata:
name: "usergroup-1-peerauth"
namespace: "usergroup-1"
spec:
mtls:
mode: STRICT
EOF
Deploy a policy for workloads in the
usergroup-2
namespace to only accept mutual TLS traffic:$ kubectl apply -f - <<EOF
apiVersion: security.istio.io/v1beta1
kind: PeerAuthentication
metadata:
name: "usergroup-2-peerauth"
namespace: "usergroup-2"
spec:
mtls:
mode: STRICT
EOF
Verify the multiple control plane creation
Check the labels on the system namespaces for each control plane:
$ kubectl get ns usergroup-1 usergroup2 --show-labels
NAME STATUS AGE LABELS
usergroup-1 Active 13m kubernetes.io/metadata.name=usergroup-1,usergroup=usergroup-1
usergroup-2 Active 12m kubernetes.io/metadata.name=usergroup-2,usergroup=usergroup-2
Verify the control planes are deployed and running:
$ kubectl get pods -n usergroup-1
NAMESPACE NAME READY STATUS RESTARTS AGE
usergroup-1 istiod-usergroup-1-5ccc849b5f-wnqd6 1/1 Running 0 12m
$ kubectl get pods -n usergroup-2
NAMESPACE NAME READY STATUS RESTARTS AGE
usergroup-2 istiod-usergroup-2-658d6458f7-slpd9 1/1 Running 0 12m
You will notice that one istiod deployment per usergroup is created in the specified namespaces.
Run the following commands to list the installed webhooks:
$ kubectl get validatingwebhookconfiguration
NAME WEBHOOKS AGE
istio-validator-usergroup-1-usergroup-1 1 18m
istio-validator-usergroup-2-usergroup-2 1 18m
istiod-default-validator 1 18m
$ kubectl get mutatingwebhookconfiguration
NAME WEBHOOKS AGE
istio-revision-tag-default-usergroup-1 4 18m
istio-sidecar-injector-usergroup-1-usergroup-1 2 19m
istio-sidecar-injector-usergroup-2-usergroup-2 2 18m
Note that the output includes
istiod-default-validator
andistio-revision-tag-default-usergroup-1
, which are the default webhook configurations used for handling requests coming from resources which are not associated with any revision. In a fully scoped environment where every control plane is associated with its resources through proper namespace labeling, there is no need for these default webhook configurations. They should never be invoked.
Deploy application workloads per usergroup
Create three application namespaces:
$ kubectl create ns app-ns-1
$ kubectl create ns app-ns-2
$ kubectl create ns app-ns-3
Label each namespace to associate them with their respective control planes:
$ kubectl label ns app-ns-1 usergroup=usergroup-1 istio.io/rev=usergroup-1
$ kubectl label ns app-ns-2 usergroup=usergroup-2 istio.io/rev=usergroup-2
$ kubectl label ns app-ns-3 usergroup=usergroup-2 istio.io/rev=usergroup-2
Deploy one
sleep
andhttpbin
application per namespace:$ kubectl -n app-ns-1 apply -f samples/sleep/sleep.yaml
$ kubectl -n app-ns-1 apply -f samples/httpbin/httpbin.yaml
$ kubectl -n app-ns-2 apply -f samples/sleep/sleep.yaml
$ kubectl -n app-ns-2 apply -f samples/httpbin/httpbin.yaml
$ kubectl -n app-ns-3 apply -f samples/sleep/sleep.yaml
$ kubectl -n app-ns-3 apply -f samples/httpbin/httpbin.yaml
Wait a few seconds for the
httpbin
andsleep
pods to be running with sidecars injected:$ kubectl get pods -n app-ns-1
NAME READY STATUS RESTARTS AGE
httpbin-9dbd644c7-zc2v4 2/2 Running 0 115m
sleep-78ff5975c6-fml7c 2/2 Running 0 115m
$ kubectl get pods -n app-ns-2
NAME READY STATUS RESTARTS AGE
httpbin-9dbd644c7-sd9ln 2/2 Running 0 115m
sleep-78ff5975c6-sz728 2/2 Running 0 115m
$ kubectl get pods -n app-ns-3
NAME READY STATUS RESTARTS AGE
httpbin-9dbd644c7-8ll27 2/2 Running 0 115m
sleep-78ff5975c6-sg4tq 2/2 Running 0 115m
Verify the application to control plane mapping
Now that the applications are deployed, you can use the istioctl ps
command to confirm that the application workloads are managed by their respective control plane, i.e., app-ns-1
is managed by usergroup-1
, app-ns-2
and app-ns-3
are managed by usergroup-2
:
$ istioctl ps -i usergroup-1
NAME CLUSTER CDS LDS EDS RDS ECDS ISTIOD VERSION
httpbin-9dbd644c7-hccpf.app-ns-1 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-1-5ccc849b5f-wnqd6 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
sleep-78ff5975c6-9zb77.app-ns-1 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-1-5ccc849b5f-wnqd6 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
$ istioctl ps -i usergroup-2
NAME CLUSTER CDS LDS EDS RDS ECDS ISTIOD VERSION
httpbin-9dbd644c7-vvcqj.app-ns-3 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-2-658d6458f7-slpd9 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
httpbin-9dbd644c7-xzgfm.app-ns-2 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-2-658d6458f7-slpd9 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
sleep-78ff5975c6-fthmt.app-ns-2 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-2-658d6458f7-slpd9 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
sleep-78ff5975c6-nxtth.app-ns-3 Kubernetes SYNCED SYNCED SYNCED SYNCED NOT SENT istiod-usergroup-2-658d6458f7-slpd9 1.17-alpha.f5212a6f7df61fd8156f3585154bed2f003c4117
Verify the application connectivity is ONLY within the respective usergroup
Send a request from the
sleep
pod inapp-ns-1
inusergroup-1
to thehttpbin
service inapp-ns-2
inusergroup-2
. The communication should fail:$ kubectl -n app-ns-1 exec "$(kubectl -n app-ns-1 get pod -l app=sleep -o jsonpath={.items..metadata.name})" -c sleep -- curl -sIL http://httpbin.app-ns-2.svc.cluster.local:8000
HTTP/1.1 503 Service Unavailable
content-length: 95
content-type: text/plain
date: Sat, 24 Dec 2022 06:54:54 GMT
server: envoy
Send a request from the
sleep
pod inapp-ns-2
inusergroup-2
to thehttpbin
service inapp-ns-3
inusergroup-2
. The communication should work:$ kubectl -n app-ns-2 exec "$(kubectl -n app-ns-2 get pod -l app=sleep -o jsonpath={.items..metadata.name})" -c sleep -- curl -sIL http://httpbin.app-ns-3.svc.cluster.local:8000
HTTP/1.1 200 OK
server: envoy
date: Thu, 22 Dec 2022 15:01:36 GMT
content-type: text/html; charset=utf-8
content-length: 9593
access-control-allow-origin: *
access-control-allow-credentials: true
x-envoy-upstream-service-time: 3
Cleanup
Clean up the first usergroup:
$ istioctl uninstall --revision usergroup-1
$ kubectl delete ns app-ns-1 usergroup-1
Clean up the second usergroup:
$ istioctl uninstall --revision usergroup-2
$ kubectl delete ns app-ns-2 app-ns-3 usergroup-2
A Cluster Administrator must make sure that Mesh Administrators DO NOT have permission to invoke the global istioctl uninstall --purge
command, because that would uninstall all control planes in the cluster.