I’ve been investigating running ingress controllers on Kubernetes recently, especially ones on a private IP address so that I can set up a public load balancer in front and control access that way. Testing that the controller from the internet without any access into the network proved to be an interesting challenge.

Obviously one way is to create a new deployment with some image that I could then run kubectl exec into and get access that way. What I didn’t know, was that instead of actually creating a deployment that I would have to delete afterwards, I could use kubectl run instead, passing in similar arguments to when you want to shell into a docker container.

I found in my usual goto Kubernetes help doc, the Kubernetes Cheat Sheet (if you don’t have this bookmarked you should) that using kubectl run you could start a container with the image of your choice and then run whatever commands you wanted to, then when you left it, the pod would be deleted. Like it was never there.

So what’s this magic command?

kubectl run -it --rm {pod name} --image={image name} --namespace {namespace to run pod in}

So if you wanted to run a pod called ingress-test with an image called debian in a namespace called ingress you’d run the following command:

kubectl run -it --rm ingress-test --image=debian -n ingress

and then you have a Debian shell, in the appropriate namespace that will be deleted by the cluster when you’re done.