Forward all TCP and UDP ports from load balancer to nginx ingress on Azure Kubernetes Service

Cluster information:

Kubernetes version: v1.13.10
Cloud being used: AKS
Installation method:
Host OS:
CNI and version:
CRI and version:

I am trying to implement a TCP/UDP gateway using kubernetes and I want to dynamically open and close a lot of ports.

Here is the detailed process:

  • We have a running container (containerA) that accepts incoming TCP connection on port 8080
  • We have a load balancer with ip 1.1.1.1, port 9091 is pointed to nginx ingress
  • Nginx Ingress will manage the connection between loadbalancer and containerA using TCP configmap
  • Loadbalancer 1.1.1.1:9091 -> nginx tcp stream 9091 -> backend containerA port 8080
  • When a new client comes, we will provision a new container (containerB) but with same port 8080
  • We will add a new port to the load balancer (port 9092)
  • Loadbalancer 1.1.1.1:9092 -> nginx tcp stream 9092 -> backend containerB port 8080
  • Repeat adding ports for new clients

The nginx ingress configmap for TCP connections looks like this:

apiVersion: v1
data:
  "9091": default/php-apache1:8080
  "9092": default/php-apache2:8080
  "9093": default/php-apache3:8080
  "9094": default/php-apache4:8080
kind: ConfigMap

Excerpt from Nginx ingress deployment yaml:

ports:
- containerPort: 9091
  hostPort: 9091
  name: 9091-tcp
  protocol: TCP
- containerPort: 9092
  hostPort: 9092
  name: 9092-tcp
  protocol: TCP

I was able to open specific TCP/UDP ports and everything works fine but right now I have 2 dilemmas:

  • Adding all the ports one by one on the yaml file is inefficient and hard to manage
  • Adding a new port (ex TCP/9091) by modifying the deployment yaml file causes the existing pods to restart. this behavior is undesirable when new ports are added every now and then

Based on my observation, when adding a new port to the nginx tcp configmap, the changes are reloaded successfully and ports are opened without needing a restart. The problem is, the ports are not yet routed properly unless you modify and add the port to the deployment yaml, which in turn causes the pod to restart.

My question is

  1. Is it possible to add the routing rules only so that the nginx pod doesn’t have to restart?
  2. Is it possible to route all ports coming from the load balancer directly to NGINX ingress under Azure Kubernetes Service
1 Like

Do you have a small example repo that shows these things working in its current state?