:: Re: [DNG] Reaping orphan processes.
Pàgina inicial
Delete this message
Reply to this message
Autor: Didier Kryn
Data:  
A: dng@lists.dyne.org
Assumpte: Re: [DNG] Reaping orphan processes.
Le 04/03/2023 à 22:22, Hendrik Boom a écrit :
> On Sat, Mar 04, 2023 at 09:38:04PM +0100, Didier Kryn wrote:
>> Le 04/03/2023 à 18:44, Hendrik Boom a écrit :
>>> One thing I've always wondered about.
>>> Why is it necessary to reap orphan processes?
>>> Why can't they just terminate by themselves?
>>>
>>     There are two reasons at least why zombies must be reaped:
>>
>>     1) every process, even orphanned, occupies some space in RAM
>>
>>     2) the whole range of process IDs will be used after some time, making
>> it impossible to give an ID to a new process, therefore making it impossible
>> to create new processes. Only once a zombie process has been reaped, its pid
>> is made available again.
>>
>>     Processes don't vanish when they terminate because the possibility must
>> be given to examine the way they terminate, their exit code. The exit code
>> tells wether the process has detected an eror or has been killed by a
>> signal. This is the way the interactive shell, for example, reports command
>> failures: the shell reaps the terminated process, examines its exit status
>> and reports errors to the terminal. Every process is responsible of reaping
>> its children processes. If the parent process dies before the child, this
>> one is called an orphan, as you would expect, and pid1 can and should reap
>> it. When a process terminates, its parent, or pid1, receives the SIGCHLD
>> signal. There is also a concept of subreaper which can slightly complicate
>> this picture.
> I see. That's how it works. Now I get to wonder why the kernel cannot detect when a process has no one to report to ...
>
> Well I guess this case has to be handled somehow and this is the way Linux chose to handle it. Maybe Unix too.
>
> Maybe it's desifned this way to avoid baking the treatment of orphans into the kernel. Is there anything else that init might want to do with a terminated process?


    What init does with zombies is declared in /etc/inittab, which is a
peculiar sort of /program/ that init interprets. The programs declared
there are eventually restarted when their process terminate. For example
when a user logs out from a virtual terminal, getty terminates and init
restarts it. inittab is written in a unique and limited declarative
language.