posix_spawnp creates ghost processes

posix_spawnp creates ghost processes

[sten.kristian.ivarsson]

Hi all

We're having a rather complex application and have noticed a rather weird
behaviour that I cannot find any information about

We're using posix_spawnp and sometimes it creates extra "ghost-process(es)"
non visible to cygwin (via e.g. process status (ps)) but visible to Windows
(Task Manager)

The function doesn't fail and it doesn't report any error, but sometimes
(hard to reproduce deterministically) it creates one, or more, extra
processes of the same binary



Have anyone noticed and/or have some information/ideas about this ?



Kristian

Re: posix_spawnp creates ghost processes

[Andrey Repin]

Greetings, Kristian Ivarsson via Cygwin!

> We're having a rather complex application and have noticed a rather weird
> behaviour that I cannot find any information about

> We're using posix_spawnp and sometimes it creates extra "ghost-process(es)"
> non visible to cygwin (via e.g. process status (ps)) but visible to Windows
> (Task Manager)

Cygwin may spawn processes for housekeeping. F.e. if you "program &" in shell,
an additional shell will be spawned to launch the "program" and notify the
parent shell when it exits.

> The function doesn't fail and it doesn't report any error, but sometimes
> (hard to reproduce deterministically) it creates one, or more, extra
> processes of the same binary

> Have anyone noticed and/or have some information/ideas about this ?

Are these processes correctly exit when all childs exit?


-- 
With best regards,
Andrey Repin
Friday, April 24, 2020 11:08:35

Sorry for my terrible english...

Sv: posix_spawnp creates ghost processes

[sten.kristian.ivarsson]

> Greetings, Kristian Ivarsson via Cygwin!
> 
> > We're having a rather complex application and have noticed a rather
> > weird behaviour that I cannot find any information about
> 
> > We're using posix_spawnp and sometimes it creates extra "ghost-
> process(es)"
> > non visible to cygwin (via e.g. process status (ps)) but visible to
> > Windows (Task Manager)
> 
> Cygwin may spawn processes for housekeeping. F.e. if you "program &" in
> shell, an additional shell will be spawned to launch the "program" and
> notify the parent shell when it exits.
> 
> > The function doesn't fail and it doesn't report any error, but
> > sometimes (hard to reproduce deterministically) it creates one, or
> > more, extra processes of the same binary
> 
> > Have anyone noticed and/or have some information/ideas about this ?
> 
> Are these processes correctly exit when all childs exit?

Well, no

The issue is that due to some other issue, the main program hangs in some
kind of blocking mode (don't know why yet)

I exit that with SIGINT (via console and Ctrl+C) and there are a few
processes running that were created by posix_spawnp according to the CygWin
TTY, but according to Windows, some of the processes are duplicated
processes

If I kill -9 them in CygWin, the ones not visible are still running
according to Windows Task Manager

> --
> With best regards,
> Andrey Repin
> Friday, April 24, 2020 11:08:35
> 
> Sorry for my terrible english...

Re: posix_spawnp creates ghost processes

[Ken Brown]

On 4/24/2020 3:30 AM, Kristian Ivarsson via Cygwin wrote:
> Hi all
> 
> We're having a rather complex application and have noticed a rather weird
> behaviour that I cannot find any information about
> 
> We're using posix_spawnp and sometimes it creates extra "ghost-process(es)"
> non visible to cygwin (via e.g. process status (ps)) but visible to Windows
> (Task Manager)
> 
> The function doesn't fail and it doesn't report any error, but sometimes
> (hard to reproduce deterministically) it creates one, or more, extra
> processes of the same binary
> 
> 
> 
> Have anyone noticed and/or have some information/ideas about this ?

Cygwin's posix_spawn uses execve under the hood.  The latter creates a new 
Windows process, does some housekeeping, and then transfers control from the old 
process to the new one.  There will be a short period of time when both 
processes exist from the Windows point of view.

Ken