Configure a custom CNI chain
Container Network Interface (CNI) is the standard interface for Kubernetes network plugins. A CNI chain runs multiple plugins in sequence, each handling a specific networking task such as IP address allocation or routing, to extend cluster networking without replacing Terway.
Prerequisites
-
An ACK managed cluster that uses the Terway network plugin is created. For more information, see Create an ACK managed cluster.
-
The Terway version is v1.5.6 or later, and the cluster does not use exclusive elastic network interface (ENI) mode.
Container Service for Kubernetes (ACK) does not guarantee that multiple CNI plugins can work together. Configuring a custom CNI chain is a high-risk operation. Configure the chain only after you fully understand the capabilities of each CNI plugin, and proceed with caution to avoid business interruption caused by configuration errors.
Configure a custom CNI chain
To use a CNI chain, add the required CNI plugins to the Terway configuration file.
Use cni_chain
Modify the Terway configuration file eni-config.
kubectl edit cm -nkube-system eni-config
|
Parameter |
Description |
|
10-terway.conf |
The Terway CNI configuration. Do not modify its contents. Important
Do not modify the original configuration. |
|
cni_chain |
The additional CNI rules. These rules are appended after Terway's existing rules. This parameter supports per-node configuration. Important
|
Configuration file example
apiVersion: v1
kind: ConfigMap
metadata:
name: eni-config
namespace: kube-system
data:
10-terway.conf: |
# Keep unchanged if no configuration is required
cni_chain: |
- type: tuning
sysctl:
net.ipv6.conf.all.disable_ipv6: "1"
Run kubectl rollout restart -n kube-system daemonset.apps/terway-eniip to recreate the Terway pods.
If the configuration is correct, the custom plugin settings appear in the /etc/cni/net.d/10-terway.conflist configuration file on the node.
Use 10-terway.conflist (not recommended)
-
Modify the Terway configuration file
eni-config.kubectl edit cm -nkube-system eni-configParameter
Description
10-terway.conf
The Terway CNI configuration. Do not modify its contents.
ImportantDo not modify the original configuration.
10-terway.conflist
The custom CNI chain configuration.
pluginsshould list the configuration in10-terway.confas the first CNI configuration.ImportantThe configurations in this topic are examples only. Do not copy them directly, because doing so may cause configuration errors.
The configuration block should be in JSON format. Ensure that the format is correct.
Configuration file example
kind: ConfigMap apiVersion: v1 metadata: name: eni-config namespace: kube-system data: 10-terway.conflist: | { "plugins": [ { "cniVersion": "0.4.0", "name": "terway", "type": "terway", "capabilities": {"bandwidth": true} }, { "type": "portmap", "capabilities": {"portMappings": true}, "externalSetMarkChain":"KUBE-MARK-MASQ" } ] } 10-terway.conf: | { "cniVersion": "0.4.0", "name": "terway", "type": "terway", "capabilities": {"bandwidth": true} } -
Run
kubectl rollout restart -n kube-system daemonset.apps/terway-eniipto recreate the Terway pods.If the configuration is correct, the custom plugin settings appear in the
/etc/cni/net.d/10-terway.conflistconfiguration file on the node.
Use cases
Configure portmap
The portmap plugin primarily maps a pod's internal ports to ports on the host, allowing external networks to access specific services in the pod.
If you need public network access, ensure that the corresponding ports are open for inbound traffic in the security group.
The policy rules generated by portmap may be incompatible with Network Policies. Test them thoroughly before use.
Configuration example
-
Access with the private IP address and port of a node:
apiVersion: v1 kind: ConfigMap metadata: name: eni-config namespace: kube-system data: 10-terway.conf: | { "cniVersion": "0.4.0", "name": "terway", "type": "terway", "capabilities": {"bandwidth": true} } cni_chain: | - type: portmap capabilities: portMappings: true externalSetMarkChain: KUBE-MARK-MASQ -
Access with the public IP address and port:
Public network access requires the following configuration:
Parameter
Non-Datapath V2
Datapath V2
masqAll
Not required
Required
symmetric_routing
Required
Required
-
masqAll: a portmap plugin parameter. Requires portmap v1.7.1 or later.
-
symmetric_routing: a Terway plugin parameter. Requires Terway v1.15.0 or later.
-
Datapath V2 clusters with kube-proxy replacement support port mapping without the portmap plugin.
apiVersion: v1 kind: ConfigMap metadata: name: eni-config namespace: kube-system data: 10-terway.conf: | { "cniVersion": "0.4.0", "name": "terway", "type": "terway", "symmetric_routing": true, "capabilities": { "bandwidth": true } } cni_chain: | - type: portmap capabilities: portMappings: true masqAll: false -
Disable IPv6 link-local addresses for containers
Inside a container, even if your cluster does not have IPv6 dual-stack enabled, the network interface is still automatically assigned an IPv6 link-local address after it is created. The address belongs to the fe80::/64 range. This is the default behavior of the operating system kernel. It is usually harmless and does not interfere with normal operations.
However, if your application handles network addresses incorrectly, it may mistake this IPv6 link-local address for the pod IP address and try to use it for communication. But link-local addresses are intended for communication between devices on the same link and cannot be used for network communication, which may cause your application to behave abnormally. If your application mistakes this IPv6 link-local address for the pod IP address, the application code has a defect. Report the issue to the application vendor first and seek a solution.
If you still need to disable IPv6 link-local addresses in containers, configure the tuning plugin.
Configuration example
kind: ConfigMap
apiVersion: v1
metadata:
name: eni-config
namespace: kube-system
data:
10-terway.conf: |
{
"cniVersion": "0.4.0",
"name": "terway",
"type": "terway",
"capabilities": {"bandwidth": true}
}
cni_chain: |
- type: tuning
sysctl:
net.ipv6.conf.all.disable_ipv6: "1"