Leaky tests hang on Windows

[chipsenkbeil]

This seems to only happen on Windows, but nextest hangs after all tests have run. I have a lot of leaky tests because I'm spawning a cli that runs for the entire test run and I cannot kill it until I do so manually as part of my ci.

Is there something I can do to determine why it's hanging only on Windows? Anything I can do to prevent the hanging?

---

.config/nextest.toml

[profile.ci]
fail-fast = false
retries = 2
slow-timeout = { period = "60s", terminate-after = 3 }
status-level = "fail"
final-status-level = "fail"

[sunshowers]

Hi there, thanks for the report!

Honestly I'm not sure what's going on there. Based on your screenshot it seems like all the tests have completed running and the test run has ended, yet nextest hasn't quit yet. This seems to suggest that one of the futures that was spawned hasn't exited yet (but I don't know which one).

Do you have a minimal example you can share, that I can try and debug? This will also help me see what sort of logging I can add in the future.

[chipsenkbeil]

@sunshowers I finally got an example working. I had to set up a Windows pipe server that receives connections from a client.

The example repository is https://github.com/chipsenkbeil/nextest-hang-example

Run using cargo nextest run --profile ci --release --all-features.

The test stands up a singleton server that persists, and regardless of success or failure the nextest runner hangs at the end. I've also verified that even when I create an instance of a server per test that should be killed when dropped, if the test fails and leaves the process around, nextest still hangs. So my guess is that this is something about the Window pipe logic that tokio has implemented.

A video illustrating hanging with nextest is this:

nextest-hanging-2.mp4

[sunshowers]

Thanks! Going to look at this soon (trying out tokio-console to debug this: #665).

[sunshowers]

OK, figured it out. Will try and publish a fix in the next couple of days.

[sunshowers]

This is now out as part of cargo-nextest 0.9.44. Thanks for reporting it and for the fantastic minimal example!