[FEATURE] Expose ports AFTER k3d creation

[deminngi]

I had a problem how to expose the tiller ingress port AFTER k3d creation, to be able to use a local helm client and set the HELM_HOST variable to HELM_HOST=localhost:44134

Here my solution:
Forward the internal ingress port over a docker container using socat

Here my procedure:

• Get the IP address of the k3d-default-server

$ kubectl get nodes

• Get the k3d network

$ sudo docker network ls

• Run a detached docker socat container

 $ docker run \
    -d \
    -p <local-port>:<local-port> \
    --name=<k3d-host>-<k3d-port>-link \
    --network <k3d-namespace> \
    alpine/socat \
      TCP4-LISTEN:<local-port>,fork,reuseaddr \
      TCP4:<k3d-host>:<k3d-port>

Example:

 $ docker run \
    -d \
    -p 44134:44134 \
    --name=k3d-default-server-44134-link \
    --network k3d-default \
    alpine/socat \
      TCP4-LISTEN:44134,fork,reuseaddr \
      TCP4:172.19.0.2:44134

Question:

1. Is there a way to implement this as a command in k3d?

$ k3d expose -l <local-port> -h <k3d-host> -r <k3d-port> -n <k3d-namespace>|default=k3d-default>

2. May be adding this to FAQ?

[mmacfadden]

This seems like a highly desirable feature. We don't always know exactly what is going to be deployed in k3s when creating the cluster.

[jaceksan]

+1

[iwilltry42]

Hi there, apparently I didn't ever comment on this, but I remember that we already had discussions about this @giminni .
First of all: Thanks for opening this issue and providing a workaround!

This is a limitation introduced by how Docker works and nothing that we can't easily fix unfortunately.
What we could do however is: re-creating the containers (keeping the state in the volume) and adding the additional port mappings. This could surely work, but would introduce downtimes 🤔

[jaceksan]

Personally I am now focused on stateful apps, esp. database clusters, which are managed by operators. Recreating PODs would be feasible, one by one, handled by the operator.

[LordChunk]

I understand this might not work for everyone, but if you're running a single node on your machine setting the network to the host's network, you also won't have to deal with opening up all required ports.
https://k3d.io/internals/networking/#host-network

[iwilltry42]

The use cases mentioned here can be handled by the new node edit and cluster edit features in k3d v5.0.0 (landing next month) ✔️

[keyiis]

@iwilltry42 how to expose k3d-k3s-default-server port to host,not loadbalancer

root@9b2af6447dbc:~# k3d node edit k3d-k3s-default-server-0 --port-add 1883:1883
FATA[0000] Currently only the loadbalancer can be updated!

[iwilltry42]

@keyiis , this is not supported, as we have to restart (or rather re-create) the container to add ports (docker won't let you add ports to running containers). This only works properly for the loadbalancer, as it doesn't hold any state.
The K3s containers (k3d nodes) however, hold quite a lot of state, so re-creating them would break the cluster.