From: Herbert Poetzl (herbert_at_13thfloor.at)
Date: Sat 24 May 2003 - 20:43:16 BST
On Sat, May 24, 2003 at 05:11:36PM +0300, Alex Lyashkov wrote:
> On Monday 19 May 2003 10:50, Herbert Poetzl wrote:
> > Sorry Alex!
> >
> not problem :)
>
> > >
> > > trivial.
> > > It's need add new command in new_s_context syscall.
> > > "stop_vserver".
> > > My patch have list all process in context.
> > > code for this simular:
> > >
> > > write_lock(&task_lock); // exclusive blocking.
> > > for_each_task_in_context(ctx,tsk)
> > > {
> > > task_t *task = tsk->task;
> > > force_sig(SIGKILL,task); // or SIGTERM
> > > }
> > > write_unlock(&task_lock);
> > > and block creating new process in context.
> > >
> > > correctly ?
> >
> > sounds good ...
> >
> > I had some thought on this issue, and I guess it would
> > be the best solution, to freeze all processes (not
> > with SIG_STOP but) by NOT scheduling any of the vserver
> > tasks anymore? why?
> freezing add _NEW_ processes in context - old process scheduling as usually.
> You scripts have many childrens and more simply kills all process and restart
> vps. But if you block add new processes in context you script stoped himself.
good idea, but what about 'normal' processes 'correctly'
spawning new processes? this would fail ...
I guess it would be optimal, if the physical administrator
could actually 'freeze' a virtual server, have a deep
look at the processes, decide to send some signals to
them (when/before unfrozen), reactivate (unfreeze) the
virtual server (tasks) and be happy ...
the next step could, for example, be making the whole
vserver persistent and restart/reactivate a frozen/saved
server on another machine ...
(just an example, no need to do this jet ;)
> > - you could have a good look at the processes from
> > outside (couldn't you?)
> not. i don't see process outside vps. it`s kernel work.
hmm, shouldn't there be some way to take a look
at vserver processes, without being part of that
vserver?
> > - you probably could send any signal from ctx-1 or
> > from vserver <ctx> enter? (am I wrong?)
> not. from tools vserver_ctl.
> run example:
> vserver_ctl -x 4 stop sendsig sigkill
> but it's need restart vps.
why does it require the vps to be restarted?
or did I get something wrong?
> > - after unfreezing SIG_KILL signalled processes, it
> > should have the same effect as the kernel loop
> > above (shouldn't it?)
best,
Herbert
PS: is there a new patch on the way? just asking
before I start to dissect the snapshot-1052304359.diff
because I do not want to do it again next week ;)
>
> --
> With best regards,
> Alex