Sandbox vs Executor to securely run generated code

[maeste]

Hi,
I'm new to the project but love it and would be keen to contribute. I've started looking to secure executors and worked on some draft code to support the podman Executor. I see there is already work underway to support Docker for this, but podman is deamonless, and it could have a few advantages for this purpose (at least it could be an alternative)

That said, I've seen PR #747 removing Executor from the documentation in favour of using SandBoxes for E2B and for Docker (new introduction in that documentation). Having those sandboxes makes sense, and I can work on something similar based on podman. I'd still like to know if you plan to get rid of the executors and keep only SaandBoxes entirely or if the two concepts are alternatives for different use cases.
That said, do you plan to include the sandboxes in the project or just document how to use them?

My original idea was to generalize as much as possible the code for Executor (i.e. abstract class) to make it possible and easy to maintain different implementations like Docker and Podman and evaluate if it would make sense to have a cloud executor on Kubernetes, maybe even Knative. Of course, I'm happy to discuss this idea (I'm not so sure it would make any sense); it was mostly an excuse to get my hands in the jam and contribute to the project, but having a goal that makes sense.

Oh, and my final question is: Is GitHub discussion the right way to discuss that kind of thing? I haven't found any reference to Discord or any other live chat, and I went for it because I had many questions, but please let me know if you prefer any other approach (maybe creating issues and discussing them over there).

Best Regards,
Stefano

[maeste]

Well I decided to went on and show some code. PR #776 provide 2 examples for Podman and Kubernetes (minikube) sandbox. They are pretty raw, miming the Docker example in the documentation recently merged (#747). They are fine as an example...of course the can and should be improved with better error handling, telemetry etc etc, but I just want to show some code as an example. I'm not even sure if there is any interest here on having multiple local container (Docker and Sandbox) and if running the agents in a pod or general in a cluster is something you are planning or desire to have.

Just my 2C
S.

[maeste]

Just a quick follow-up here: I've opened also PR #816 Adding podman executor. I haven't created one for kubernetes, for which I think it make sense only the full sandbox provided in #776, but I'd happy to discuss it.
I've also moved this thread to "Ideas" which seems more appropriate category for this kind of discussion

S.