server_key_algorithms test flaky?

[cpu]

Noticed a failure on main that was resolved after a retry:

thread 'server_key_algorithms' panicked at tests/runner.rs:807:9:
assertion `left == right` failed: stdout EOF -- read so far []
  left: 0
 right: 1
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Archived build log: logs_25438992097.zip

[cpu]

I'm able to reproduce this locally running the test in a loop (example script), but it's fairly rare. My most recent failure happened after 450 test runs. After staring at the code for a while I'm pretty stumped.

It seems like when the test fails it's because reading the pipe created by Command::new(...).stdout(Stdio::piped()) to run target/server is returning 0 on the very first iteration of wait_for_stdout, in the first call to stream.stdout().as_mut().unwrap().read(&mut input).unwrap(). I'm not sure how that could happen 🤔 Maybe dumping stderr when it does would give some hint?