Problems with AddressResolver on ipv6 stack

[ecejnj42]

Version

5.0.0

Context

There is a problem in KubeServiceState when a WebSocket is created to a Kubernetes API with an ipv6 stack.
In particular, the WebSocket fails to establish connection after the TLS handshake is correctly performed.
After some tries, I discovered that if environment variable "KUBERNETES_SERVICE_HOST" is enclosed in square brackets:
"fd00:eccd:22:ffff::5555:1" -> "[fd00:eccd:22:ffff::5555:1]"
then connection is correctly performed and KubeServiceState is able to get notified about changes in "endpoints" configuration.

Therefore there should be some code within the WebSocket framework that is not able to correctly handle "ipv6:port" schema
(i.e. To avoid problems with the ":" in the ipv6 for an uri schema, just enclose it with square brackets)

Steps to reproduce

To reproduce just deploy a Kubernetes cluster with ipv6 stack.
In particular service for api server should be something like:
`kubectl get svc -n default

NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP fd00:eccd:22:ffff::5555:1 443/TCP 224d
`

Do you have a reproducer?

No response

[vietj]

do you mean we are not properly parsing the "ip" address in the JSON payload ?

[ecejnj42]

I do not know exactly where the problem is, but I guess it is not in the json payload.
The problem exactly comes when in class KubeServiceState, in method connectWebSocket:

String requestURI = "/api/v1/namespaces/" + this.resolver.namespace + "/endpoints?watch=true&allowWatchBookmarks=true&resourceVersion=" + this.lastResourceVersion; WebSocketConnectOptions connectOptions = new WebSocketConnectOptions(); connectOptions.setServer(this.resolver.server); connectOptions.setURI(requestURI);

Setting the server "this.resolver.server" caused that the later connection to fail:
.connect(connectOptions).onComplete((ar) -> { if (ar.succeeded()) { WebSocket ws = (WebSocket)ar.result(); if (this.disposed) { ws.close(); } else { this.ws = ws; } } else if (!this.disposed) { this.vertx.setTimer(500L, (id) -> this.connectWebSocket()); }

The connection just fails. I activated the TLS debug option (-Djavax.net.debug=ssl,handshake,data,plaintext) and I saw the TLS is established but after that, the processing is stuck.
Just changing the environment variable KUBERNETES_SERVICE_HOST, which is used to connect the WebSocket to Kubernetes API , and enclosing in square brackets "[..]", just solves the problem.
I assume that somewhere within the "connect" an http uri is composed as a String using the "host:port", which is not valid if "host" is an ipv6. Anyway, I was not able to trace that out.

[vietj]

So it might be just a matter of documenting it

[ecejnj42]

I do not think so, because the variable value KUBERNETES_SERVICE_HOST is generated automatically from kubernetes. I just did some tricks to enter the container and change it for my manual test (I do not think this shall be documented as a general procedure)
I am not completly sure because I did not debug the whole application, but there is some code in Http1xClientConnection, method toWebSocket, that is recreating the URI, using ":":
if (!wsuri.isAbsolute()) { // Netty requires an absolute url wsuri = new URI((ssl ? "https:" : "http:") + "//" + server.host() + ":" + server.port() + requestURI); }
Perhaps the issue is related with that or something similar.