Onboarding VMs

Bare metal servers

In this guide we only call out virtual machines (VMs). If you want to onboard a workload running on a bare metal server, simply replace VM for bare metal. There is no difference in handling between them.

Problem definition

Istio and the underlying Kubernetes platform create a sealed ecosystem, where control plane and data plane components are tightly integrated. For example, control plane components running on each node create a mutually trusted relationship. When new pods are scheduled to run on a node, the node is a trusted entity and its critical resources, like iptables, are modified.

When a virtual machine (VM) is brought into that ecosystem, it is an outsider. To successfully extend an Istio/Kubernetes cluster with a VM, the following steps must be taken:

• Authentication. The VM must establish an authenticated encrypted session to the control plane. The VM must prove that it is allowed to join the cluster.
• Routing. The VM must be aware of services defined in the Kubernetes cluster and vice-versa. If the VM runs a service, it must be visible to pods running inside the cluster.

Overview

Onboarding a Virtual Machine (VM) into a TSB managed Istio service mesh can be broken down into the following steps:

• Registering the VM workload with the Istio control plane (WorkloadEntry)
• Obtaining a bootstrap security token and seed configuration for the Istio Proxy that will run on the VM
• Transfering the bootstrap security token and seed configuration to the VM
• Starting Istio Proxy on the VM

To improve the user experience with VM onboarding, TSB comes with tctl CLI that automates most of these tasks.

At a high level, tctl aims to streamline VM onboarding flow down to a single command:

tctl x sidecar-bootstrap

The tctl sidecar bootstrap logic, as well as the registration of the VM workload with the service mesh is driven by the configuration inside a WorkloadEntry resource. tctl sidecar bootstrap allows you to onboard VMs in various network and deployment scenarios to your service mesh on Kubernetes. The tctl sidecar bootstrap also allows VM onboarding to be reproduced at any point from any context, by a developer machine, or a CI/CD pipeline.

Requirements

Before you get started make sure that you have:

✓ TSB version 0.9 or above
✓ A Kubernetes cluster onboarded into TSB
✓ The relevant application deployed on that Kubernetes cluster in the appropriate namespace
✓ A Virtual Machine spun up and ready to go
✓ The most recent kubectl ready
✓ The most recent Tetrate Service Bridge CLI (tctl) ready

Differences between environments

This set-up guide provides the common steps you need to take to get a VM onboarded. Since you have to deal with your specific combination of cloud providers, networks, firewalls, workloads, and operating systems, you will need to adapt the steps so that they will work for your situation. In this guide, we will use the example of onboarding the Ratings service from the Istio Bookinfo example on an Ubuntu VM.

Walkthrough

Cluster Mesh expansion

To allow workloads from outside the Kubernetes environment to become part of the service mesh, you need to enable mesh expansion in the cluster. Edit the ControlPlane resource to include the meshExpansion property as shown below.

apiVersion: install.tetrate.io/v1alpha1
kind: ControlPlane
metadata:
  name: controlplane
  namespace: istio-system
spec:
  meshExpansion: {}

To edit the resource, run the following command:

kubectl patch ControlPlane controlplane -n istio-system \
    --patch '{"spec":{"meshExpansion":{}}}' \
    --type merge

Once you have completed this step you can onboard as many VM workloads to this cluster as needed. If you have multiple clusters, repeat this step for each cluster where you need to onboard VMs.

VM Preparation

To prepare your VM for onboarding you will need to have SSH access to a privileged user on the VM, because you must add a user account and install additional software.

Example environment

As an example this guide will show how to prepare an Ubuntu 18.04 LTS virtual machine.

First, ensure that Docker is installed on the VM. You will be installing Istio Proxy later on, which will run inside a Docker container. Using Docker allows you to keep the Proxy dependencies isolated from your operating system installation, as well as provide a homogeneous environment for the Proxy to run on.

To install Docker on the VM, run:

sudo apt-get update
sudo apt-get -y install docker.io

To allow tctl to onboard your VM workload, create and configure a dedicated user account. This user account will need permissions to interact with the Docker daemon as well as have SSH access. To bootstrap the onboarding process, the tctl tool will connect to your VM using SSH.

To set up and configure the user account, run the following commands:

# create dedicated user account “istio-proxy” for VM onboarding
sudo useradd --create-home istio-proxy

# sudo into the dedicated user
sudo su - istio-proxy

# configure SSH access for the new user account
mkdir -p $HOME/.ssh
chmod 700 $HOME/.ssh
touch $HOME/.ssh/authorized_keys
chmod 600 $HOME/.ssh/authorized_keys

#
# Add your SSH public key to $HOME/.ssh/authorized_keys
#

# go back to the privileged user
exit

To give the new user account permissions to interact with Docker daemon, you must add the account to the docker user group:

sudo usermod -aG docker istio-proxy

To store the onboarding configuration, you must set up a directory for it. If you wish to use a different path, make sure it is reflected in the WorkloadEntry resource which you will be configuring later.

sudo mkdir -p /etc/istio-proxy
sudo chmod  775 /etc/istio-proxy
sudo chown istio-proxy:istio-proxy /etc/istio-proxy

If your workload is not running yet, start it now. In our example we will run the Ratings service from the Istio Bookinfo example. This example will be running inside Docker but this is not required. Your workload can run from the operating system as a regular process.

sudo docker run -d \
    --name ratings \
    -p 127.0.0.1:9080:9080 \
    docker.io/istio/examples-bookinfo-ratings-v1:1.16.2

Configuring firewalls

To allow a VM to join a service mesh, there must be IP (L3) connectivity between the VM and the Kubernetes cluster. You may need to configure firewalls at the Kubernetes and the VM network ends, to allow traffic between the two on the various TCP ports used for traffic.

Kubernetes and VM on the same network (or peered networks)

Since all workloads have direct IP connectivity, traffic between the VM and Pod IPs will not use the VM gateway.

In this scenario you must:

✓ allow ingress traffic from the VM IP to the entire range of TCP ports on the Pod IPs
✓ allow ingress traffic from Pod IPs to a relevant set of TCP ports on the VM IP

Kubernetes and VM on different networks

When workloads that span Kubernetes and VM do not have direct IP connectivity, traffic must flow through the VM Gateway.

In this scenario where the networks are segregated, the following TCP ports on the VM Gateway in Kubernetes must be accessible by the VM:

• 15012 (control plane xDS traffic)
• 15443 (data plane ingress traffic)
• 9411 (sidecar tracing data ingress)
• 11800 (sidecar access logs ingress)

GKE and EKS

On both GKE and EKS these ports will allow incoming traffic automatically.

The ports you need to open for traffic to your VM workload from the Kubernetes cluster is dependent on the proxy listening ports that you will configure in the Sidecar resource. In the example, we use port 9080. In this case, we need to allow TCP traffic from Kubernetes to VM on port 9080.

Create a WorkloadEntry

A WorkloadEntry resource captures information about a workload running on a VM which will allow you to properly onboard the VM with a verifiable identity that is recognized by the TSB service mesh.

The configuration of a WorkloadEntry deals with the following information:

• Details needed by the service mesh to register the VM workload
• Annotations to provide tctl with the right information to bootstrap the VM onboarding
• Labels for TSB observability that hold the logical service identity of the workload

A template example of a WorkloadEntry resource can be seen below and highlights properties that must be configured based on the specifics of your environment.

apiVersion: networking.istio.io/v1beta1
kind: WorkloadEntry
metadata:
  name: ratings-vm
  namespace: bookinfo
  annotations:
    sidecar-bootstrap.istio.io/ssh-host: <ssh-host>
    sidecar-bootstrap.istio.io/ssh-user: istio-proxy
    sidecar-bootstrap.istio.io/proxy-config-dir: /etc/istio-proxy
    sidecar-bootstrap.istio.io/proxy-image-hub: docker.io/tetrate
    sidecar-bootstrap.istio.io/proxy-instance-ip: <proxy-instance-ip>
spec:
  address: <address>
  labels:
    class: vm
    app: ratings   # mandatory label for observability through TSB
    version: v3    # mandatory label for observability through TSB
  serviceAccount: bookinfo-ratings
  network: <vm-network-name>

network: <vm-network-name>

The service mesh in your Kubernetes cluster needs to know if your VM resides in a network that can directly reach the pod IPs. As explained in the firewall section, this will determine if traffic should be routed through the VM gateway or not. By adding the network property and providing a name for the VM network, the service mesh will enable VM gateway routing. If you omit the network property, the service mesh will assume the VM to run on a network with direct IP connectivity.

address: <address>

Address must hold the destination IP of the VM workload that can be directly connected to by the pods. In the same network scenario this is the VM IP address that a pod can directly connect to. In a segregated network scenario this is the VM IP address the pods can reach. As an example, if you have different VPCs for Kubernetes and VM this can be a private IP address as long as VPC routing/peering is set up correctly. In case of different cloud providers this typically is a public IP address on which the VM is reachable.

proxy-instance-ip: <proxy-instance-ip>

If this annotation is provided, it must hold the IP address the Istio Proxy sidecar on the VM can bind its listener to. This typically is the IP address of the interface that will receive incoming traffic originating from outside. If the VM has an interface that is configured with a public IP address and this is the same IP as the address property, this annotation can be omitted. Most cloud providers the VMs do not have an interface that listens directly on a public IP address, but on a private IP. In this case, you must configure the internal IP address of the VM to which external incoming traffic is routed.

ssh-host: <ssh-host>

When you execute the tctl bootstrap command, tctl tries to connect to the VM it needs to onboard. The default behavior is for tctl to use the IP address as found in the address property. If the machine you run tctl on does not have direct IP connectivity to that address, for instance in the case of address holding a private IP address, you can set this optional ssh-host annotation. In this case, provide the IP address or hostname that allows tctl to connect over SSH to the VM.

In the ratings VM example we will assume the following:

• Kubernetes Cluster at cloud provider and VM on-prem
• External VM IP address is not directly bound on the machine
• Istio Proxy on VM Workload will listen on TCP port 9080
• tctl and Kubernetes can both reach the VM over the same external IP.
• VM internal IP: 10.128.0.2
• VM external IP: 35.194.38.142

Example WorkloadEntry using these assumptions looks like this:

apiVersion: networking.istio.io/v1beta1
kind: WorkloadEntry
metadata:
  name: ratings-vm
  namespace: bookinfo
  annotations:
    sidecar-bootstrap.istio.io/ssh-user: istio-proxy
    sidecar-bootstrap.istio.io/proxy-config-dir: /etc/istio-proxy
    sidecar-bootstrap.istio.io/proxy-image-hub: docker.io/tetrate
    sidecar-bootstrap.istio.io/proxy-instance-ip: 10.128.0.2
spec:
  address: 35.194.38.142
  labels:
    class: vm
    app: ratings
    version: v3
  serviceAccount: bookinfo-ratings
  network: vm-us-central1-a

Save as ratings-workloadentry.yaml. You can add this file to source control or apply directly to your cluster using kubectl:

kubectl apply -f ratings-workloadentry.yaml

Create a Sidecar

Now that we have configured our WorkloadEntry to provide information for bootstrapping the onboarding process of the VM and IP connectivity, we need to configure the VMs Istio Proxy sidecar. The Sidecar resource gives you control over the configuration of the Istio Proxy. In this example the Sidecar configuration allows you to avoid using IPtables on the VM for redirecting traffic.

The Sidecar example below shows the configuration for the ratings VM example, listening on TCP port 9080.

apiVersion: networking.istio.io/v1beta1
kind: Sidecar
metadata:
  name: bookinfo-ratings-no-iptables
  namespace: bookinfo
spec:
  egress:
  - bind: 127.0.0.2
    hosts:
    - ./*
  ingress:
  - defaultEndpoint: 127.0.0.1:9080
    port:
      name: http
      number: 9080
      protocol: HTTP
  workloadSelector:
    labels:
      app: ratings
      class: vm

Save as ratings-sidecar.yaml. You can add this file to source control or apply directly to your cluster using kubectl:

kubectl apply -f ratings-sidecar.yaml

Onboard the VM

With both WorkloadEntry and Sidecar configured and applied to your Kubernetes cluster, the VM workload is now registered with your service mesh. With your VM and service mesh prepared, we can use tctl to complete the actual onboarding process.

The tctl CLI will:

• Obtain a bootstrap security token and seed configuration from the service mesh
• Transfer this bootstrap security token and seed configuration to the VM
• Start Istio Proxy using the bootstrap security token and seed configuration

Since this onboarding process is complex, tctl implements a dry run feature. It will allow you to inspect the process flow without actual execution. To see what tctl is planning to do, run:

tctl x sidecar-bootstrap ratings-vm.bookinfo \
    --start-istio-proxy \
    --dry-run

You’ll see an output similar to below:

[SSH client] going to connect to istio-proxy@35.194.38.142:22

[SSH client] going to execute a command remotely: mkdir -p /etc/istio-proxy

[SSH client] going to copy into a remote file: /etc/istio-proxy/sidecar.env
JWT_POLICY=third-party-jwt
PROV_CERT=/var/run/secrets/istio
OUTPUT_CERTS=/var/run/secrets/istio
PILOT_CERT_PROVIDER=istiod
...

The dry run will output both generated configuration and commands that will be run over SSH on the VM.

Once you’re satisfied with the test, you can remove the --dry-run argument and start the actual onboarding process like this:

tctl x sidecar-bootstrap ratings-vm.bookinfo \
    --start-istio-proxy

An output similar to below will appear:

[SSH client] connecting to istio-proxy@35.194.38.142:22
[SSH client] executing a command remotely: mkdir -p /etc/istio-proxy
[SSH client] copying into a remote file: /etc/istio-proxy/sidecar.env
[SSH client] copying into a remote file: /etc/istio-proxy/k8s-ca.pem
[SSH client] copying into a remote file: /etc/istio-proxy/istio-ca.pem
[SSH client] copying into a remote file: /etc/istio-proxy/istio-token
[SSH client] executing a command remotely: docker rm --force istio-proxy
istio-proxy
[SSH client] executing a command remotely: docker run -d --name istio-proxy --restart unless-stopped --network host ...
[SSH client] closing connection

When done, you should see the generated configuration copied on the VM and a Docker container with Istio Proxy started.

VM Workload calling mesh services

The VM workload in this example does not need to initiate requests to services inside the service mesh. In case your VM workload does, special considerations need to be made with respect to egress routing. The Istio Proxy instance on your VM is dynamically provided with a list of available services in the service mesh. Since our example does not use iptables to automatically redirect traffic to the Istio Proxy you will need to update your /etc/hosts file to include the FQDN of each of the services your Workload needs to initiate requests to and point them to the bind address of the egress listener.

Example of additions to /etc/hosts:

# direct the following services to the egress listener address
127.0.0.2 reviews.bookinfo.svc
127.0.0.2 www.example.org

For more information on this topic, please consult the IstioEgressListener reference.

Testing workload traffic

To test if onboarding the VM succeeded and traffic is flowing between workloads on VM and Kubernetes, you have multiple options. Next to inspecting logs of your services to see if activity occurs, you can also use the TSB UI to look at the various service metrics and the topology map. The values for the app and version labels you provided in the WorkloadEntry will be reflected in the topology map and service metrics.