During some of our testing on openshift, we had performed several workflows which did the following:

1. Forked a new process with a new kubeconfig
2. Performed several commands using kubectl (~10-20)
3. Gather forks.
4. repeat.

What this showed was that the sum of connection management and tls overhead was quite large.
Ideally we could reduce the overhead by reusing the connection which is done in many of the e2e tests.

Some ideas here include:

• The notion of a repl, or shell, for kubectl
• Several different tools for client side mgmt
• ...

/cc @jeremyeder @kubernetes/sig-scalability @smarterclayton @liggitt