trinity-devel@lists.pearsoncomputing.net

Message: previous - next
Month: February 2014

Re: [trinity-devel] Notes to commit 77aed4a4

From: Slávek Banko <slavek.banko@...>
Date: Tue, 25 Feb 2014 03:59:22 +0100
On Monday 24 of February 2014 23:10:39 Darrell wrote:
> On Mon February 24 2014 3:42:56 pm Slávek Banko wrote:
> > Darrell,
> >
> > I have a note to commit 77aed4a4. I'm afraid that used grep is dangerous.
> > For example, my colleague has a login name 's' (yes, only one letter).
> > Such grep would return unwanted results. I propose a change roughly as
> > follows. This will make it possible to omit some grep and also condition.
> >
> > ps -u $USER -o pid= -o comm= | grep -w kdesktop_lock | awk '{print $1}' |
> > xargs -r kill -9
> >
> >
> > The second note is that this killing can be dangerous. If the machine is
> > configured for multi-user login (multi-seat, xdmcp), the user could have
> > multiple active sessions. Alternatively, user could have running some
> > applications remotely - dcopserver would be active for remote session.
>
> I am open to whatever works best. I developed those snippets because I do a
> lot of testing for the project. During testing I do things that many people
> are unlikely to do, but that is exactly what testing includes.
>
> For example, in my testing I routinely use Ctrl+Alt+Backspace. When I do
> that some Trinity processes are not terminated gracefully. The snippets I
> added are intended to ensure those abandoned processes do not impede the
> next startup of Trinity.
>
> Yet even when I am using my main office system, sometimes Trinity breaks
> when logging out. As you know, we have been trying to resolve those
> problems for a very long time. We have yet to find the runaway process,
> memory leak, or race condition that causes this breakage. So even on a
> production box sometimes I resort to using Ctrl+Alt+Backspace. If this
> happens to me then I presume other users are affected too.
>
> Similarly, at the end of starttde where I explicitly kill gamin, I have
> seen gamin hang a reboot or halt. Often when I logout of Trinity I then
> reboot or halt the system. Once in a while the system hangs until I open a
> console and kill gamin. I have that in my various system shutdown scripts
> but since Trinity uses gamin indirectly, there is no harm in killing a
> gamin process spawned by that user. Even with multiple logins by the same
> person, gamin just creates a new process. The kill command won't kill gamin
> processes owned by other users and I did not expect any problems.
>
> That is the background of the new snippets. Please modify the snippets to
> make them more robust and reduce errors or conflicts.
>
> By the way, my apologies for 6a571749. I should have caught that and would
> have done so this morning when I rebuilt tdebase with some test patches.
> Had I built tdebase yesterday I would have had my Homer Simpson moment
> sooner (Doh!).

I have no objections to killing gamin_server. However, killing kdesktop_lock 
for active session causes the user logs off - it's not very good. What is the 
effect of kills dcopserver for active applications I have not tested it.

Regarding commit 6a571749 / 44b8d19c, I rather surprised that this had an 
influence. I assumed that the updates is applied only on the earlier version 
of the configuration, and it will not affect the fresh configuration.

Slavek
--