[Tarantool-patches] [PATCH 6/7] popen: handle setsid os specifics

[Alexander Turenko]

On Tue, Mar 10, 2020 at 07:41:00PM +0300, Cyrill Gorcunov wrote:
> On Tue, Mar 10, 2020 at 07:36:46PM +0300, Alexander Turenko wrote:
> > > It seems that setsid() is used mainly to disassociate from a controlling
> > > terminal (to don't be hit by SIGHUP if it'll die). In this context
> > > setpgrp() would not be sufficient.
> > 
> > I just realized that there is another reason to use setsid(), where
> > setpgrp() is applicable too: move the child into its own process group
> > and kill the whole group (child and its childs if any) then. I mean, use
> > the corresponding flag (which I proposed to add in [1]), which will
> > change :kill() behaviour.
> > 
> > [1]: https://lists.tarantool.org/pipermail/tarantool-patches/2020-March/014608.html
> > 
> > So, it seems, we should do ioctl() + setpgrp() on Mac OS?
> 
> The child may generate subchildren with own groups. If I'm not missing
> something obvious we should provide only basic functionality whic would
> be enough to spawn new processes. The child may generate subchildren
> with own group, serisouly without pid namespace we simply do not control
> much. Thus I propose to leave it in the state it is right now.

We don't intend to offer some kind of guaranteed isolation. A child may
spawn a daemon, yep. But we usually know what we're run and whether it
will do something like this.

We should provide a tool to do the following:

* Allow to spawn a background process, which will not die even if a
  controlling terminal died (setsid() or ioctl() solve it).
* Allow to set a new process group for a process and its childs (say,
  for a non-inveractive shell) in order to be able to kill the entire
  group if we're going to free resources (setsid() or setpgrp() solve
  it).

Usual case for the latter bullet: spawn a pipeline using "sh -c 'foo |
bar'" and be able to kill it entirely. A non-interactive shell does not
perform job control, so even "bash -c 'foo & bar'" will be killed
entirely in the case.

After looking around discussions re other popen implementations, I guess
that we'll be asked for those abilities sooner or later. It seems
logical to implement it, since we know it is expected from a popen
implementation.

WBR, Alexander Turenko.