Configure transparent proxying

In order to automatically intercept traffic from and to a service through a kuma-dp data plane proxy instance, Kuma utilizes a transparent proxying using iptables.

Transparent proxying helps with a smoother rollout of a Service Mesh to a current deployment by preserving existing service naming and as the result - avoid changes to the application code.

Kubernetes

On Kubernetes kuma-dp leverages transparent proxying automatically via iptables installed with kuma-init container or CNI. All incoming and outgoing traffic is automatically intercepted by kuma-dp without having to change the application code.

Kuma integrates with a service naming provided by Kubernetes DNS as well as providing its own Kuma DNS for multizone service naming.

Universal

On Universal kuma-dp leverages the [data plane proxy specification] (/docs/2.3.x/production/dp-config/dpp-on-universal#dataplane-configuration) associated to it for receiving incoming requests on a pre-defined port.

There are several advantages for using transparent proxying in universal mode:

  • Simpler Dataplane resource, as the outbound section becomes obsolete and can be skipped.
  • Universal service naming with .mesh DNS domain instead of explicit outbound like https://localhost:10001.
  • Support for hostnames of your choice using VirtualOutbounds that lets you preserve existing service naming.
  • Better service manageability (security, tracing).

Setting up the service host

Prerequisites:

  • kuma-dp, envoy, and coredns must run on the worker node – that is, the node that runs your service mesh workload.
  • coredns must be in the PATH so that kuma-dp can access it.
    • You can also set the location with the --dns-coredns-path flag to kuma-dp.

Kuma comes with kumactl executable which can help us to prepare the host. Due to the wide variety of Linux setup options, these steps may vary and may need to be adjusted for the specifics of the particular deployment. The host that will run the kuma-dp process in transparent proxying mode needs to be prepared with the following steps executed as root:

  1. Create a new dedicated user on the machine.

    1. useradd -u 5678 -U kuma-dp

  2. Redirect all the relevant inbound, outbound and DNS traffic to the Kuma data plane proxy.

    1. kumactl install transparent-proxy \
    2.   --kuma-dp-user kuma-dp \
    3.   --skip-resolv-conf \
    4.   --redirect-dns

Please note that this command will change the host iptables rules.

The changes will persist over restarts, so this command is needed only once. Reverting to the original state of the host can be done by issuing kumactl uninstall transparent-proxy.

Data plane proxy resource

In transparent proxying mode, the Dataplane resource should omit the networking.outbound section and use networking.transparentProxying section instead.

  1. type: Dataplane
  2. mesh: default
  3. name: 
  4. networking:
  5.   address: 
  6.   inbound:
  7.   - port: 
  8.     tags:
  9.       kuma.io/service: demo-client
  10.   transparentProxying:
  11.     redirectPortInbound: 15006
  12.     redirectPortOutbound: 15001

The ports illustrated above are the default ones that kumactl install transparent-proxy will set. These can be changed using the relevant flags to that command.

Invoking the Kuma data plane

It is important that the kuma-dp process runs with the same system user that was passed to kumactl install transparent-proxy --kuma-dp-user.

When systemd is used, this can be done with an entry User=kuma-dp in the [Service] section of the service file.

When starting kuma-dp with a script or some other automation instead, we can use runuser with the aforementioned yaml resource as follows:

  1. runuser -u kuma-dp -- \
  2.   /usr/bin/kuma-dp run \
  3.     --cp-address=https://<IP or hostname of CP>:5678 \
  4.     --dataplane-token-file=/kuma/token-demo \
  5.     --dataplane-file=/kuma/dpyaml-demo \
  6.     --dataplane-var name=dp-demo \
  7.     --dataplane-var address=<IP of VM> \
  8.     --dataplane-var port=<Port of the service>  \
  9.     --binary-path /usr/local/bin/envoy

You can now reach the service on the same IP and port as before installing transparent proxy, but now the traffic goes through Envoy. At the same time, you can now connect to services using Kuma DNS.

firewalld support

If you run firewalld to manage firewalls and wrap iptables, add the --store-firewalld flag to kumactl install transparent-proxy. This persists the relevant rules across host restarts. The changes are stored in /etc/firewalld/direct.xml. There is no uninstall command for this feature.

Upgrades

Before upgrading to the next version of Kuma, make sure to run kumactl uninstall transparent-proxy and only then replace the kumactl binary. This will ensure smooth upgrade and no leftovers from the previous installations.

Configuration

Intercepted traffic

By default, all the traffic is intercepted by Envoy. You can exclude which ports are intercepted by Envoy with the following annotations placed on the Pod

  1. apiVersion: apps/v1
  2. kind: Deployment
  3. metadata:
  4.   name: example-app
  5.   namespace: kuma-example
  6. spec:
  7.   ...
  8.   template:
  9.     metadata:
  10.       ...
  11.       annotations:
  12.         # all incomming connections on ports 1234 won't be intercepted by Envoy
  13.         traffic.kuma.io/exclude-inbound-ports: "1234"
  14.         # all outgoing connections on ports 5678, 8900 won't be intercepted by Envoy
  15.         traffic.kuma.io/exclude-outbound-ports: "5678,8900"
  16.     spec:
  17.       containers:
  18.         ...

You can also control this value on whole Kuma deployment with the following Kuma CP configuration

  1. KUMA_RUNTIME_KUBERNETES_SIDECAR_TRAFFIC_EXCLUDE_INBOUND_PORTS=1234
  2. KUMA_RUNTIME_KUBERNETES_SIDECAR_TRAFFIC_EXCLUDE_OUTBOUND_PORTS=5678,8900

The default settings will exclude the SSH port 22 from the redirection, thus allowing the remote access to the host to be preserved. If the host is set up to use other remote management mechanisms, use --exclude-inbound-ports to provide a comma separated list of the TCP ports that will be excluded from the redirection.

Execute kumactl install transparent-proxy --help to see available options.

Reachable Services

By default, every data plane proxy in the mesh follows every other data plane proxy. This may lead to performance problems in larger deployments of the mesh. It is highly recommended to define a list of services that your service connects to.

  1. apiVersion: apps/v1
  2. kind: Deployment
  3. metadata:
  4.   name: example-app
  5.   namespace: kuma-example
  6. spec:
  7.   ...
  8.   template:
  9.     metadata:
  10.       ...
  11.       annotations:
  12.         # a comma separated list of kuma.io/service values
  13.         kuma.io/transparent-proxying-reachable-services: "redis_kuma-demo_svc_6379,elastic_kuma-demo_svc_9200"
  14.     spec:
  15.       containers:
  16.         ...
  1. type: Dataplane
  2. mesh: default
  3. name: 
  4. networking:
  5.   address: 
  6.   inbound:
  7.   - port: 
  8.     tags:
  9.       kuma.io/service: demo-client
  10.   transparentProxying:
  11.     redirectPortInbound: 15006
  12.     redirectPortOutbound: 15001
  13.     reachableServices:
  14.       - redis_kuma-demo_svc_6379
  15.       - elastic_kuma-demo_svc_9200

Transparent Proxy with eBPF (experimental)

Starting from Kuma 2.0 you can setup transparent proxy to use eBPF instead of iptables.

To use Transparent Proxy with eBPF your environment has to use Kernel >= 5.7 and have cgroup2 available

  1. kumactl install control-plane \
  2.   --set "experimental.ebpf.enabled=true" | kubectl apply -f-
  1. kumactl install transparent-proxy \
  2.   --experimental-transparent-proxy-engine \
  3.   --ebpf-enabled \
  4.   --ebpf-instance-ip <IP_ADDRESS> \
  5.   --ebpf-programs-source-path <PATH>

If your environment contains more than one non-loopback network interface, and you want to specify explicitly which one should be used for transparent proxying you should provide it using--ebpf-tc-attach-iface <IFACE_NAME> flag, during transparent proxy installation.