redirect fds rather than closing

[skshetry]

On inherited processes, the sys.{stdin/stderr/stdout} object were being set to None due to result of closing the standard stream fds (see closerange below). We redirect 0, 1 and 2 to dev null now.

Also rolled back to use subprocess for macOS as fork is not safe there.

I have also added a dvc daemon test command for use in a functional test.

[codecov]

Codecov Report

Attention: 11 lines in your changes are missing coverage. Please review.

Files	Coverage Δ
dvc/commands/daemon.py	72.72% <100.00%> (+6.06%)	⬆️
tests/func/test_daemon.py	86.66% <86.66%> (ø)
dvc/daemon.py	48.05% <18.18%> (ø)

... and 67 files with indirect coverage changes

📢 Thoughts on this report? Let us know!.

[dberenbaum]

It's working as expected for me

[efiop]

Suggested change

	def redirect_streams_to_null():
	def _redirect_streams_to_null():

[efiop]

Let's at least include a link to your explanation comment on github.

[efiop]

@skshetry From your research in #10026 (comment) , are there indications that fork might be a problem on linux as well?

[skshetry]

I haven't found any issues with fork (without exec) in Linux, except on multithreading cases. But it seems fork is not considered safe in Unix if you don't immediately exec and might cause deadlock. See https://www.evanjones.ca/fork-is-dangerous.html.

I am not sure if using fork only on Linux is worth it at this point, if we are not going to use it in Windows and in macOS.
I will create a PR to use subprocess instead.

[efiop]

@skshetry Sounds good. Do you want to fix windows test so we could merge this for now or would you like to wait for that PR?

[skshetry]

I will wait for the next PR, i need to add tests. Will keep this open as an alternative for now.

[skshetry]

Unfortunately, we'll have to stick with this PR. There is no way to detach a subprocess (without double-forking).