Re: recovering zombie process

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On Jan 17,  2:14am, Brian K. White wrote:
}
} It's impractical to use on linux though because your kernel is always a 
} moving target not supported by the module.
} It's wonderful on commercial OS's where you have the same kernel as
} everyone 

There are plenty of commercial distributions of linux that deliberately
stay on relatively stable kernel versions, and free distributions that
mirror those commercial ones closely for the same reason.

} > If the parent has crashed, there's no hangup.
} 
} Then shouldn't the next higher parent do it?

No.  Why would it?  It doesn't know what those jobs are; it doesn't
even get CHLD signals about them.

HUP signals are sent by the TTY driver layer when a connection is lost,
to the process group associated with that TTY.  In the case of the kind
of terminal multiplexer he was using, there's only one job in the
process group of the user-facing TTY, and that's the multiplexer itself.
All the other shells are in process groups attached to their respective
pseudo-TTY devices, and they haven't lost those connections.

On a normal lost connection the multiplexer would get the HUP and be
responsible for propagating it.  If the multiplexer has crashed, the
chain of responsibility is broken.