Egress Gateway

The egress gateway feature routes all IPv4 connections originating from pods and destined to specific cluster-external CIDRs through particular nodes, from now on called “gateway nodes”.

When the egress gateway feature is enabled and egress gateway policies are in place, matching packets that leave the cluster are masqueraded with selected, predictable IPs associated with the gateway nodes. As an example, this feature can be used in combination with legacy firewalls to allow traffic to legacy infrastructure only from specific pods within a given namespace. The pods typically have ever-changing IP addresses, and even if masquerading was to be used as a way to mitigate this, the IP addresses of nodes can also change frequently over time.

This document explains how to enable the egress gateway feature and how to configure egress gateway policies to route and SNAT the egress traffic for a specific workload.


This guide assumes that Cilium has been correctly installed in your Kubernetes cluster. Please see Cilium Quick Installation for more information. If unsure, run cilium status and validate that Cilium is up and running.


For more insights on Cilium’s Egress Gateway, check out eCHO episode 76: Cilium Egress Gateway.

Preliminary Considerations

Cilium must make use of network-facing interfaces and IP addresses present on the designated gateway nodes. These interfaces and IP addresses must be provisioned and configured by the operator based on their networking environment. The process is highly-dependent on said networking environment. For example, in AWS/EKS, and depending on the requirements, this may mean creating one or more Elastic Network Interfaces with one or more IP addresses and attaching them to instances that serve as gateway nodes so that AWS can adequately route traffic flowing from and to the instances. Other cloud providers have similar networking requirements and constructs.

Additionally, the enablement of the egress gateway feature requires that both BPF masquerading and the kube-proxy replacement are enabled, which may not be possible in all environments (due to, e.g., incompatible kernel versions).

Delay for enforcement of egress policies on new pods

When new pods are started, there is a delay before egress gateway policies are applied for those pods. That means traffic from those pods may leave the cluster with a source IP address (pod IP or node IP) that doesn’t match the egress gateway IP. That egressing traffic will also not be redirected through the gateway node.

Incompatibility with other features

Because egress gateway isn’t compatible with identity allocation mode kvstore, you must use Kubernetes as Cilium’s identity store (identityAllocationMode set to crd). This is the default setting for new installations.

Egress gateway is not compatible with the Cluster Mesh feature. The gateway selected by an egress gateway policy must be in the same cluster as the selected pods.

Egress gateway is not compatible with the CiliumEndpointSlice feature (see GitHub issue 24833 for details).

Egress gateway is not supported for IPv6 traffic.

Enable egress gateway

The egress gateway feature and all the requirements can be enabled as follow:

$ helm upgrade cilium ./cilium \
   --namespace kube-system \
   --reuse-values \
   --set egressGateway.enabled=true \
   --set bpf.masquerade=true \
   --set kubeProxyReplacement=true \
   --set l7Proxy=false

Rollout both the agent pods and the operator pods to make the changes effective:

$ kubectl rollout restart ds cilium -n kube-system
$ kubectl rollout restart deploy cilium-operator -n kube-system

Writing egress gateway policies

The API provided by Cilium to drive the egress gateway feature is the CiliumEgressGatewayPolicy resource.


CiliumEgressGatewayPolicy is a cluster-scoped custom resource definition, so a .metadata.namespace field should not be specified.

kind: CiliumEgressGatewayPolicy
  name: example-policy

To target pods belonging to a given namespace only labels/expressions should be used instead (as described below).

Selecting source pods

The selectors field of a CiliumEgressGatewayPolicy resource is used to select source pods via a label selector. This can be done using matchLabels:

- podSelector:
      labelKey: labelVal

It can also be done using matchExpressions:

- podSelector:
    - {key: testKey, operator: In, values: [testVal]}
    - {key: testKey2, operator: NotIn, values: [testVal2]}

Moreover, multiple podSelector can be specified:

- podSelector:
- podSelector:

To select pods belonging to a given namespace, the special io.kubernetes.pod.namespace label should be used.


Only security identities will be taken into account. See Limiting Identity-Relevant Labels for more information.

Selecting the destination

One or more IPv4 destination CIDRs can be specified with destinationCIDRs:

- "a.b.c.d/32"
- "e.f.g.0/24"


Any IP belonging to these ranges which is also an internal cluster IP (e.g. pods, nodes, Kubernetes API server) will be excluded from the egress gateway SNAT logic.

It’s possible to specify exceptions to the destinationCIDRs list with excludedCIDRs:

- "a.b.0.0/16"
- "a.b.c.0/24"

In this case traffic destined to the a.b.0.0/16 CIDR, except for the a.b.c.0/24 destination, will go through egress gateway and leave the cluster with the designated egress IP.

Selecting and configuring the gateway node

The node that should act as gateway node for a given policy can be configured with the egressGateway field. The node is matched based on its labels, with the nodeSelector field:

      testLabel: testVal


In case multiple nodes are a match for the given set of labels, the first node in lexical ordering based on their name will be selected.


If there is no match for the given set of labels, Cilium drops the traffic that matches the destination CIDR(s).

The IP address that should be used to SNAT traffic must also be configured. There are 3 different ways this can be achieved:

  1. By specifying the interface:

          testLabel: testVal
      interface: ethX

    In this case the first IPv4 address assigned to the ethX interface will be used.

  2. By explicitly specifying the egress IP:

          testLabel: testVal
      egressIP: a.b.c.d


    The egress IP must be assigned to a network device on the node.

  3. By omitting both egressIP and interface properties, which will make the agent use the first IPv4 assigned to the interface for the default route.

          testLabel: testVal

Regardless of which way the egress IP is configured, the user must ensure that Cilium is running on the device that has the egress IP assigned to it, by setting the --devices agent option accordingly.


The egressIP and interface properties cannot both be specified in the egressGateway spec. Egress Gateway Policies that contain both of these properties will be ignored by Cilium.

Example policy

Below is an example of a CiliumEgressGatewayPolicy resource that conforms to the specification above:

kind: CiliumEgressGatewayPolicy
  name: egress-sample
  # Specify which pods should be subject to the current policy.
  # Multiple pod selectors can be specified.
  - podSelector:
        org: empire
        class: mediabot
        # The following label selects default namespace
        io.kubernetes.pod.namespace: default

  # Specify which destination CIDR(s) this policy applies to.
  # Multiple CIDRs can be specified.
  - ""

  # Configure the gateway node.
    # Specify which node should act as gateway for this policy.
      matchLabels: a-specific-node

    # Specify the IP address used to SNAT traffic matched by the policy.
    # It must exist as an IP associated with a network interface on the instance.

    # Alternatively it's possible to specify the interface to be used for egress traffic.
    # In this case the first IPv4 assigned to that interface will be used as egress IP.
    # interface: enp0s8

Creating the CiliumEgressGatewayPolicy resource above would cause all traffic originating from pods with the org: empire and class: mediabot labels in the default namespace and destined to (i.e. all traffic leaving the cluster) to be routed through the gateway node with the a-specific-node label, which will then SNAT said traffic with the egress IP.

Selection of the egress network interface

For gateway nodes with multiple network interfaces, Cilium selects the egress network interface based on the node’s routing setup (ip route get <externalIP> from <egressIP>).


Redirecting to the correct egress network interface can fail under certain conditions when using a pre-5.10 kernel. In this case Cilium falls back to the current (== default) network interface.

For environments that strictly require traffic to leave through the correct egress interface (for example EKS in ENI mode), it is recommended to use a 5.10 kernel or newer.

Testing the egress gateway feature

In this section we are going to show the necessary steps to test the feature. First we deploy a pod that connects to a cluster-external service. Then we apply a CiliumEgressGatewayPolicy and observe that the pod’s connection gets redirected through the Gateway node. We assume a 2-node cluster with IPs (node1) and (node2). The client pod gets deployed to node1, and the CEGP selects node2 as Gateway node.

Create an external service (optional)

If you don’t have an external service to experiment with, you can use Nginx, as the server access logs will show from which IP address the request is coming.

Create an nginx service on a Linux node that is external to the existing Kubernetes cluster, and use it as the destination of the egress traffic:

$ # Install and start nginx
$ sudo apt install nginx
$ sudo systemctl start nginx

In this example, the IP associated with the host running the Nginx instance will be

Deploy client pods

Deploy a client pod that will be used to connect to the Nginx instance:

$ kubectl create -f
$ kubectl get pods
NAME                             READY   STATUS    RESTARTS   AGE
pod/mediabot                     1/1     Running   0          14s

$ kubectl exec mediabot -- curl

Verify from the Nginx access log (or other external services) that the request is coming from one of the nodes in the Kubernetes cluster. In this example the access logs should contain something like:

$ tail /var/log/nginx/access.log
[...] - - [04/Apr/2021:22:06:57 +0000] "GET / HTTP/1.1" 200 612 "-" "curl/7.52.1"

since the client pod is running on the node it is expected that, without any Cilium egress gateway policy in place, traffic will leave the cluster with the IP of the node.

Apply egress gateway policy

Download the egress-sample Egress Gateway Policy yaml:

$ wget

Modify the destinationCIDRs to include the IP of the host where your designated external service is running on.

Specifying an IP address in the egressIP field is optional. To make things easier in this example, it is possible to comment out that line. This way, the agent will use the first IPv4 assigned to the interface for the default route.

To let the policy select the node designated to be the Egress Gateway, apply the label egress-node: true to it:

$ kubectl label nodes <egress-gateway-node> egress-node=true

Note that the Egress Gateway node should be a different node from the one where the mediabot pod is running on.

Apply the egress-sample egress gateway Policy, which will cause all traffic from the mediabot pod to leave the cluster with the IP of the Egress Gateway node:

$ kubectl apply -f egress-gateway-policy.yaml

Verify the setup

We can now verify with the client pod that the policy is working correctly:

$ kubectl exec mediabot -- curl
<HTML><HEAD><meta http-equiv="content-type" content="text/html;charset=utf-8">

The access log from Nginx should show that the request is coming from the selected Egress IP rather than the one of the node where the pod is running:

$ tail /var/log/nginx/access.log
[...] - - [04/Apr/2021:22:06:57 +0000] "GET / HTTP/1.1" 200 612 "-" "curl/7.52.1"


To troubleshoot a policy that is not behaving as expected, you can view the egress configuration in a cilium agent (the configuration is propagated to all agents, so it shouldn’t matter which one you pick).

$ kubectl -n kube-system exec ds/cilium -- cilium-dbg bpf egress list
Defaulted container "cilium-agent" out of: cilium-agent, config (init), mount-cgroup (init), apply-sysctl-overwrites (init), mount-bpf-fs (init), wait-for-node-init (init), clean-cilium-state (init)
Source IP    Destination CIDR    Egress IP   Gateway IP

The Source IP address matches the IP address of each pod that matches the policy’s podSelector. The Gateway IP address matches the (internal) IP address of the egress node that matches the policy’s nodeSelector. The Egress IP is on all agents except for the one running on the egress gateway node, where you should see the Egress IP address being used for this traffic (which will be the egressIP from the policy, if specified).

If the egress list shown does not contain entries as expected to match your policy, check that the pod(s) and egress node are labeled correctly to match the policy selectors.