Feature request: Output test maximum memory usage
Opened this issue · 1 comments
When running tests the time they take to execute is output to the console, it would also be valuable to see their maximum memory usage.
With large test sets you can start to hit OOM errors.
Thanks for the request!
It looks like the way to do this would be to use wait4 (available on most Unixes, e.g. Linux) to obtain, in particular, the maxrss. I'm not sure if something similar would work on Windows.
I did look briefly at the wait4 crate: https://github.com/lu-zero/wait4-rs. But it has limitations:
-
It doesn't support anything other than a few Unixes. This should be easy to fix.
-
Per lu-zero/wait4-rs#1 it looks like Windows support still needs a little shoring up. But the good news is that we do already create a JobObject within nextest, so fixing it is hopefully not very difficult.
-
We'll need to figure out a way to integrate the crate with async/Tokio. This would mean either working with upstream Tokio to add support for custom wait methods, or bypassing and either copy/pasting or rewriting most of Tokio's process code (e.g. the Unix code). We may be able to punt on this by using blocking calls, but that seems janky and I'm worried there are subtle cases (particularly around signal handling) where it breaks. The interaction between signal handling and process spawning is very fraught.
I think it'd be worth working with the Tokio folks to see what ideas they have here.
It's a somewhat involved project with a fair bit of coordination. I don't plan to work on this unless my employer happens to need it, but contributions welcome!