[nylug-talk] ubuntu looking to replace init

[Joseph M. Gaffney]

On Tuesday 12 September 2006 16:17, Ruben Safir wrote:
> Not it is **not** what it is supposed to do.  It **is** broken under these
> conditions and it **should** behave that way and not silently just work
> around the broken hardware.

If you dont have a usb hard drive attached, or your network card doesn't have 
a connection, or a piece of hardware otherwise is not detected or working 
(non-essential for a computer being turned on, obviously), your computer 
should simply stop working? 

Sounds silly to me.

> And that is where these discussions start to get tacky.  I don't want some
> programmer from HP defining what behavior to have when things aren't
> working as I set them up.

So much like your init scripts, you'd change your upstart configuration.  I 
don't know that some programmer from HP has access to upstart or a sysv style 
init, but sure.  Ok.

> I will fix my computer.  I don't need the system to do anything other than
> what I tell it to and if I tell it to connect to the internet with a broken
> network card, then that is darn well what I expect it to do.  And if it
> fails, ***I*** will figure out what to do by **observing** the lag in the
> machines initiation and reading the system logs from the kernel.

Apparently you think you lose all ability to configure everything that could 
possibly be configured by using upstart.  Don't worry, I'm sure some sort of 
text file will be available, and I'm sure you can use binary ascii to fill it 
out.

But seriously, what you're suggesting is that I, as a user, *must* observe lag 
in the machines initiation, and read system logs, when a simple event driven 
system can prompt me when there is an issue, or a new device becomes 
available, etc?

Sorry, I don't believe in more work than necessary.

> > > > No, you don't want that.  You want your system to be
> > > > smart enough to know that there is no cat5 cable, so its not going to
> > > > get an ip address - the two are related.
> > >
> > > Among other issues, who in their right mind turns off the computer?
> >
> > Laptop users.  People who conserve power.  College students.
>
> Not on my laptops.  I have a 10 hour batery and a plug.

How nice for you.

Most people don't.

> > > > Now, a cable is the nice one - you can
> > > > simply plug it in.  However, what if something is wrong with your
> > > > network card?  Is it necessary to wait, or wouldn't you like to get
> > > > right into the CLI or  your DE of choice to resolve the issue?
> > >
> > > No software can 'resolve' the issue when harware is broken. It can
> > > cover it up and prevent the useful syslog information which would tell
> > > me the kernel reporting that the hardware is misbehaving.
> >
> > "Error: IP Configuration not started. Network connection not present." or
> > some such.  Theres a problem with that?
>
> It does that anyway now upon initiation.  I need more information than
> that. And once the system is up, I need a lot more information.

How would upstart stop you from any of this?  The current init daemon simply 
runs a series of scripts.  The proposed upstart daemon runs a series of 
scripts based on user-driven events as specified.  

> > How is it "covering" things up?
>
> I walk away from the machine while it boots and come back and it is
> **working**
>
> Or it happens while running and the system is **working** when it should be
> broken.

Again, your computer is simply broken when a single device isn't working 
properly, or a single daemon doesn't load? Sounds like a tough system to use 
if you ask me.

> > > > Lets take it further.  Say you use a variety of servers, www, ftp,
> > > > whatever, specifically set up for use over a network.  Well, if it
> > > > were tied into Upstart, you wouldn't have to wait for each of them to
> > > > load and find there is no connection.
> > >
> > > Yes this was tried and largely, althouh not absolutely, disliked.  This
> > > was what inet (not init), the super dameon did and it was largely a
> > > security nightmare.
> >
> > And again, you wouldn't *have* to tie them in.  Its also called an
> > example. Want more? Go to the upstart site for better ones, and better
> > explanations. From your comments I get the impression you haven't.
> >
> > Firewire drives can be added randomly, PCMCIA cards the same, power
> > consumption is an issue, USB, now wireless usb, flash media, so on - the
> > traditional method, without any helper application, does not exactly
> > serve these purposes well.
>
> It does serve them well.  I don't want events turning on kernel modules
> tied into the initiation process.  These are used based events and I don't
> want them in init.

No, they aren't.  When say, a usb drive is connected, a user based event would 
be working with that drive, like deleting a file.  However, theres also 
system actions that must be taken.... initializing the drive, detecting 
partitions, firmware may need to be loaded.

Am I comfortable at the CLI? Sure.  My girlfriend? Not so much.  We're well 
beyond the days of simply replying rtfm, and have moved into Linux becoming 
the kind of OS where most everything Just Works, but you can still make it 
work how you want.

> > Should I change the run level every time I plug a
> > device in?
>
> I'm not sure what your getting to because I run my USB digital camera every
> day as an end user without root privledges.  I also put PCMCIA cards into
> my laptop.  There is a seperate programming running that does this, just
> were it belongs, ie:  not in init.

You mean it belongs to a DE, a la GNOME Volume Manager? D-BUS, which is 
turning into an all-you-can-be type app?

Does it really belong there, or is a camera's drive no different than the 
drives permanently installed in your PC?  Does it still have a filesystem 
which can be checked?

I have to disagree wholeheartedly.  The way its being done now duplicates 
work, since there are systems already in place to do precisely the same 
thing: mount a volume.

>
> You used the word intelligent and computers together.
>

I used intelligent to compare init and upstart.  Upstart has more logic than 
init, in theory.

> I only know what I read here.  And one of the things 30 years of computing
> expereince and 45 years of living experinece gives one is a good ear for
> such things.

It would be quite easy to check out http://wiki.ubuntu.com/ReplacementInit and 
see things explained by those more knowledgable than myself.

>
> > You also don't seem to be aware that people have
> > dynamic hardware configurations, and needs vary within a moments notice.
>
> Which means what exactly?  Do you think that a new firewire device will
> work without a kernel drive, at a moments notice?  Because it aint.

Not everything needs to be loaded *all* of the time, and thats the point.

> And wont this system be convinent.  Now any comercial vendor will be about
> to make a non-free piece of software work by simply following the 'simple'
> do everything API for this new init thingie which is more than init and
> cron combined.

Ok once again you're veering OT.... and you've completely lost me.  Where does 
non-free software fit in?

> It says security nightmare, freedom nightware, debugging nightmare, and
> usability nightmare all over this....as it has been described here....thus
> far.

I fail to see a security nightmare, as security is being taken rather 
seriously by the developers.  Who and what can trigger events is determined 
by their existing priveleges.

Freedom nightmare? I'm sorry... wtf?

Debugging nightmare? How?

Usability nightmare? Are you possibly suggesting init is user-friendly?  Many 
persons using Linux for years have no idea what to do beyond "init 3" 
and "init 5".  I'd bet even less know what "init 6" will do.

> > I have more than a bit of experience in that regard.
>
> I only have a small bit, but I'm still not hearing anything useful that I
> would need but see many many pitfalls of the regular user of GNU systems.
>
>
> Ruben
>

-Joseph M. Gaffney aka CuCullin