I reduced the startup time of my shell by one second. Here’s how:

What I do

I work a lot with the shell (or “Terminal” app on MacOS), mostly kubectl, git, terraform and docker. And of course I use the absolutely best shell of all, oh-my-zsh.

The Problem

I noticed that recently the startup time for a new shell (or new tab) has grown longer than a second. That’s annoying, because usually when I open a new tab, I want to quickly check on something, like the logs of a pod or if I have committed something in a different project.

I inserted some gdate +%s.%N commands in my .zshrc to find out, at which point the slowdown was occurring. (gdate is the gnu-version of date and can be installed with brew install coreutils.)

The problem lies in the kubectl plugin, which creates the completions-list via source <(kubectl completion zsh). This creates a huge list of instructions.

The Solution

  1. The kubectl plugin also defines the alias k=kubectl, which shaves seconds, if not minutes off my daily work.
  2. When I start a new shell, I expect it to be fast. When I check on something in the Kubernetes cluster I expect it to go through the network, i.e. to be slow.

So I delayed the initialization of the kubectl completion until I first call something:

function k() {
  source <(kubectl completion zsh)
  alias k=kubectl
  kubectl $@

This way, when I write k get pods for the first time, it creates the autocompletion list, rebinds what k does and then runs the command I wanted in the first place. All following k get pods go directly to kubectl.