asteval VS piston

There is asteval project. It makes more poweful (ans slower) virtual python interpreter which you can use to run many commands in shared context. My benchmark (100 000 runs of simple python expressions):

If compare to asteval (which is actually has much more features), evalidate is much faster in my benchmarks (benchmark code in repo): 0.017s vs 1.232s

But, literals are not good enough very often, and one may want to do "len(foo)+bar[i]" , so I went searching trying to find a good implementation. I dug and found asteval, but this issue I raised breaks it for me, apart from the fact that I try not to rely on too many modules beyond the standard library + it has a bunch of peculiarities that make it not a real drop-in replacement to eval().

and public reviews (as well as reviews on similar approaches [1]), the approach does not seem to satisfy my requirements. After some additional searching, I found a possible dockerized solution:

https://github.com/engineer-man/piston

I want to ask the HN community if anybody has experience in this problem space and what solutions they would suggest. Is the Piston's dockerized approach secure enough to be used in production systems?

I would really appreciate any insights anyone could provide.

Anything in the milliseconds is pretty large for fast languages, so I think it could be something to do with the way that they sandbox submissions. Ultimately I dunno, but it is interesting to speculate how they do it, maybe some cheesed Linux containers like piston / https://github.com/engineer-man/piston or something. Heavily altered runtime could swing a few ms here or there, so that's the logic for not relying on Leetcode for accurate assumptions, I guess

If you're deadset on the absurd madness you can attempt to use a community "punching bag" container, like https://github.com/engineer-man/piston, something like that or a honeypot that's tuned to resist abuses of the infra. But that way lies pain, lots of pain and dogecoin miners, rats, and trojans. Regular precautions won't be enough, and even a locked-down container/VM is likely only a matter of time. Decent hackers' bots are gonna run down a giant list of stuff from metasploit and there's probably something in open-enough userland that can be abused for escape.

I don't know if there is any resource availible to read specifically for this problem. But here is what I found on a quick google search: https://github.com/engineer-man/piston

I found some open-source projects like piston and Judge0. But there are some limitations like the number of requests per second, etc.

You should consider using something like piston. This way they can run code in a sandboxed way and you don’t have to worry. You can host it yourself and let them use that which probably yields the best results but there is also a free hosted version you can use. You could even make a simple website wrapper for the students to use to run their code.

The code runner is https://github.com/engineer-man/piston

Biggest one: https://github.com/engineer-man/piston Made with Go lang: https://github.com/ranna-go/ranna Something here: https://github.com/jakhax/sandman

why not use a secure code execution engine like python-discord/snekbox or engineer-man/piston though?