While trying to locally reproduce intermittent failures of main.named_pipe which happened a number of times on "windows-1809" CI worker, I found that if main.named_pipe is called twice in a row, second test instance hangs. If tests are executed in parallel with N workers, it's possible to execute N instances of the main.named_pipe without failure. Then N+1 instance is scheduled on a worker that executed an instance of main.named_pipe before, and it hangs.

git state: 92014bd1c611fd27c487bcc47a7521baa435214c