Re: Process Management (was: Re: Reliability of RPC services)

[Marcus Brinkmann]

At Thu, 27 Apr 2006 16:12:55 -0400,
"Jonathan S. Shapiro" <address@hidden> wrote:
> 
> On Thu, 2006-04-27 at 21:14 +0200, Marcus Brinkmann wrote:
> 
> > Michal is right, the defining relationship between processes is the
> > parent-child relationship.
> 
> This is true in the example you are discussing, and in many other
> examples, but it does not generalize. In general, the defining
> relationship between processes is the relationship defined by the
> processes. Parent/child is only one possible relationship.

We have talked about this before.  In the Hurd, the defining
relationship is parent-child.  We have to define how process
management works, both to provide the actual implementation and the
user interfaces to control them, so we can't leave this question
unanswered.  Other models are indeed possible.

> > Instead, every process will know about its child processes.  These
> > child processes have local job numbers (not global process IDs)
> > similar to bash's %n.  Because the resources are distributed
> > hierarchically, forcibly killing a job will kill all its descendants
> > as well.  This is the right thing to do.
> 
> In Coyotos, the important relationship for purpose of killing things is
> the parent/child relationships of the space banks. The correct way to
> kill this JVM is to kill the space bank from which it was run. This will
> (recursively) kill everthing allocated from that bank, including
> sub-processes. The resources do not go away because the processes die.
> It is the other way around: the processes die because the resources are
> revoked.

This is indeed what I meant with "hierarchical resource distribution"
and "forcibly killing".  Thanks for expanding on that.

Marcus