[Pkg-ltsp-devel] 0.82debian2 - what's coming?

[Gustavo Franco]

On 3/8/06, Petter Reinholdtsen <pere at hungry.com> wrote:
> [Gustavo Franco]
> > By the way, don't you have any input on the keyboard problem
> > described by me some messages ago ?
>
> I have no idea.  It reminds me of a problem I saw a long time ago,
> where the keyboard would lock up when the ps2 mouse was accessed.
> Never really figured out that one.

We will see, i'll merge the patch in my tree and inform you here,
Otavio hit the same bug too. I think you can reproduce it if you're
not using ldm (he uses wdm and me gdm).

> > What about a possible new one that i see using the latest package
> > (both server and client with a fresh chroot) with a 2.4 kernel in
> > the client that is: The system stops during the boot. It run well
> > each daemon and isn't running any ttys (i tried set -x in bootlogd
> > and it exit with 0 after that the system stops). Changing the kernel
> > from 2.4 for 2.6 in the client hides the bug. Are you using 2.6 only
> > ?
>
> No idea.  Are you testing with bootchart to visualize the boot
> process?

No, but i think the image will show the boot process until the last
init entry and stop. It seems not to be a bug in the middle of the scripts
but it seems that getty isn't running. I tried to hack bootlogd adding the
getty call, it seems to be running (it takes a while to run the next
step) and the
script exit sucessfully.

> > Closing, i would like to comment that i'm testing initng (for my CDD
> > only) to see if i can speed up the client boot and deuglify it a
> > bit.  It needs a 2.6 kernel and a ltsp-start-client init "wrapper",
> > that i'm writing and will submit to the initng Debian maintainer,
> > everything else seems to be ok.
>
> Personally I believe initng is the wrong approach, as there really is
> no need to stop using the existing init.d scripts, when one can
> document dependencies using LSB headers and run scripts in parallell
> using the new rc script in etch and sid.
>
> But I am very interested in measurements on improvements, compared to
> the current implementation in sysvinit, and a few suggested
> improvements of the rc script (C implementation was published on
> initscripts-ng list).

It's clear that simple replace the entire init system isn't trivial
and shouldn't be
done due to the possibility of document the dependencies as you said. I've
decided to try init-ng in my cdd that uses ltsp, because it works well
in my sid
desktop and already contains the needed dependencies, it just lacks one file so
it seems to be the shortest path right now, not the better in the mid,
long term.

It was just a "look at this cool new toy" and see if we can speed up
the things without
too much work, keeping in mind that we need to release until the end of the next
month. Deuglify the boot would be cool too. I mean, we could use a alternative
terminal to show the same output as we've now and let only what a
regular Debian user
already read in a "normal" setup. I'm now talking about the Debian
LTSP, what's your
opinion on this? I'm not sure if it should be high priority now thought.

Thanks,
-- stratus