systemd for 11.10 ?

Patrick Goetz pgoetz at mail.utexas.edu
Wed May 11 17:01:50 UTC 2011


On 05/11/2011 04:57 AM, ubuntu-devel-request at lists.ubuntu.com wrote:
> From: Reinhard Tartler <siretart at ubuntu.com>
> Date: Wed, 11 May 2011 10:55:09 +0200
>
> So you want to require screen users to break their current user
> experience by having them to write (system-wide) upstart scripts?

I'm completely befuddled by this sub-topic.  As soon as you detach from 
a screen, it automatically re-attaches to init and hence should not be 
affected by anything that happens to gdm (or a shell) subsequently.

>
> Use case: We have a lab of powerful ubuntu computer to which potentially
> thousands of students can log in via gdm or ssh and start long-running
> jobs in a screen session.

Has anyone actually tested this with systemd?  When you start screen, it 
starts a child process called SCREEN.  If the screen parent gets a kill 
signal (say if the parent shell is killed), SCREEN automatically 
detaches itself from screen and becomes a child of init.  Is the issue 
that cgroups are somehow being used to kill every related process with 
some kind of magic bullet, so the normal screen reattachment process is 
circumvented?  If so, then it sounds like screen just needs to be 
updated to work with cgroups.  (Note that I'm speculating here, and have 
no idea how cgroups are being used to send kill signals to related 
process groups.)

It's a long standing feature of UNIX that all the children are killed 
whenever the parent process gets the ax -- it sounds like people are 
arguing that it shouldn't work this way, which is kind of absurd.







More information about the ubuntu-devel mailing list