lib: Attach stdout to child only if --log=stdout and stdout FD is a tty

[odivlad]

We've faced with a regression after we moved from frr 8.5 to a newer versions (9.0, 9.1, 10.0, 10.1).

In our automation we start frr from python code using subprocess like this:

subprocess.run(
    ["/usr/lib/frr/frrinit.sh", "start"],
    stdout=subprocess.PIPE,
    stderr=subprocess.PIPE,
)

This call hangs in poll() for stdout FD indefinitely and frrinit.sh hangs in zombie state unless we kill subprocess.run() call.
If we remove stdout capturing from subprocess.run() call (make associated to stdout FD a tty FD), things go well and process terminates correctly with no hangs and zombies.
This commit fixes this situation where stdout of a process started in a daemon mode was attached to a calling process regardless of presence --log option.

If this patch is acceptable, it's worth backporting down to 9.0 branch, where the initial commit 4d28aea appeared.

[ton31337]

Could you fix the styling?

[odivlad]

@ton31337, sure. Done.

[ton31337]

https://github.com/FRRouting/frr/pull/16738/checks?check_run_id=29656115900

[odivlad]

@ton31337, done.
It was not obviously that git clang-format should be called twice.
Additionally removed a period symbol from the end of commit summary.

[ton31337]

LGTM

[odivlad]

@ton31337, this change seems to be qualified as a bugfix, so would you consider to backport it down to supported branches?

[odivlad]

@ton31337, it's just gentle ping about backport.

[ton31337]

@Mergifyio backport stable/10.1 stable/10.0 stable/9.1

[mergify]

backport stable/10.1 stable/10.0 stable/9.1

✅ Backports have been created

• #16964 lib: Attach stdout to child only if --log=stdout and stdout FD is a tty (backport #16738) has been created for branch stable/10.1
• #16965 lib: Attach stdout to child only if --log=stdout and stdout FD is a tty (backport #16738) has been created for branch stable/10.0
• #16966 lib: Attach stdout to child only if --log=stdout and stdout FD is a tty (backport #16738) has been created for branch stable/9.1

[odivlad]

Thanks, @ton31337 and @donaldsharp!

[donaldsharp]

This commit breaks topotest recording of memory leaked on shutdown. From my perspective this is a pretty big regression. I'd like to discuss what we can do to fix this else I am going to want to back this out.

[odivlad]

Can you please point out how to reproduce the issue, so I can look on it?

[donaldsharp]

This code change assumes that you are running as a daemon and if you are not you should not just be closing files. Additionally it breaks things like --log=stdout | grep <something> or --log=stdout > /tmp/bar

In any event if you run topotests pre and post your change you can see that it has stopped reporting on memory leaks.

[odivlad]

Okay, thanks. I'll try to look at this tomorrow.