Allow to avoid the --mount option

[mgoltzsche]

Currently when a function needs to access the project's file system the --mount option must be provided to kpt fn run.
This is very project-specific and seems unnecessary. It would be great if kpt could be used in a project-agnostic way (at least in most common cases).

I imagine two ways of supporting that:
a) Always mount the project directory at a fixed path into the function container - actually the directory in which the function is specified should be mounted to be independent from the current project/working directory.
b) Wrap all non-yaml files in resources with a special kind (without applying them) and provide them to the function via stdin so that the function container can e.g. write them to its file system (I did that with kustomizr).

[prachirp]

cc @monopole feature request for kpt fn run

[mgoltzsche]

I was also asking myself how to run kpt in a build pipeline within k8s. There the usual problem occurs that docker is required (while in most cases functions may not be run inside k8s but on a developer's machine I imagine they need to be run in a pull request pipeline just to verify that everything has been generated consistently).

Besides solving this with podman (I didn't try it yet but it should do) it would be great to have a simple, k8s-native solution to run functions without the need to mount the docker.sock or run a privileged container.

I figure one could translate docker run calls to kubectl run but this would only be a simple solution if there are no volume mounts involved.

[wstrange]

I was also asking myself how to run kpt in a build pipeline within k8s. There the usual problem occurs that docker is required.

I have the same question. We use Tekton and Kaniko today to build images in k8s without requiring access to the docker socket.

kpt functions feel like they are tricky to integrate with "K8S native"

[mgoltzsche]

Potentially this could be solved with #1151.