Attach to pod feature

Is your enhancement related to a problem? Please describe

For our current project, we need the ability to attach to the container in a pod to send stdin and read stdout. If I understood correctly, currently there is no way to do it.

Describe the solution you’d like

Implement kubectl attach equivalent the same way as kubectl exec is implemented. The main concern regarding the implementation of such feature is that its’ DSL would clash the existing one for exec (e.g. redirectingInput().redirectingOutput().exec("...")). In my point of view, the DSL should be extended instead of copying, i.e.:

ExecWatch execResult = client.pods().withName("...").inContainer("session").redirectingInput().exec("ls");
AttachWatch attachResult = client.pods().withName("...").inContainer("session").redirectingInput().attach("ls");

Describe alternatives you’ve considered

The alternative way to support such case would be to give a user the ability to use WebSockets directly as it is implemented in official kubernetes-client (though, just to mention, official kubernetes-client has the attach feature). I don’t like such approach, but it may be easier to implement.

Additional context

I’m ready to try to implement this feature by myself after syntax & requirements will be clear.

[shawkins]

I’m trying to attach to the main process (that is actually a shell) of the pod. In that case I can reconnect to the same process (with same previously defined env vars, for example).

Sounds good.

I’m ready to try to implement this feature by myself after syntax & requirements will be clear.

Try adding ExecWatch attach() as a method on https://github.com/fabric8io/kubernetes-client/blob/master/kubernetes-client-api/src/main/java/io/fabric8/kubernetes/client/dsl/Execable.java

Then add the attach implementation on PodOperationsImpl - you should be able to reuse most of the exec logic.

[kuzznya]

do you want a new process, or are you trying to attach to the main process of the pod?

I’m trying to attach to the main process (that is actually a shell) of the pod. In that case I can reconnect to the same process (with same previously defined env vars, for example).

App failure or rescheduling results in a new process

Sorry, I didn’t clarify that in my previous message - I meant rescheduling or failure of my app with k8s client, not of a pod with a shell.