Re: [gentoo-dev] Init replacement

[Wouter van Kleunen <kleunen@cs.utwente.nl>]

On Sat, 3 May 2003, leon j. breedt wrote:

> hi,
>
> i as well looked into doing this, and came to the conclusion that the
> following are desirable attributes for a replacement init:
>
>     1. secure daemon startup (i.e. as different user, in chroot, etc).
> basically,
>        incorporating the functionality of start-stop-daemon into init.
this is definitly a good option, i'll definitly add this

>     2. daemon monitoring/controlling at an init level. i.e. you have a
> guaranteed
>        working way of checking service status, sending signals, etc.
> very useful
>        would be the ability to execute configurable commands/scripts on
> status
>        change. basically, incorporating similar functionality to djb's
> daemontools.
Yes, that is already in there

>     3. standardized, controllable output to TTY. i.e. be able to
> suppress anything
>        that a daemon/service tries to output, but log it somewhere. it
> should be
>        possible to flag something that could be interactive though, like
> fsck.
>        abstracting this would make it trivial to support different
> startup displays,
>        like a progress bar, or graphical/icon based startup.
Yup, that is also supported.

>     4. deep dependency checking when determining what to start up/stop.
> and when restarting,
>        only start up dependant things which were running when stopped.
>
Yes. Well, for the moment, you can only stop services that do not have
dependent services running. I plan to make that stronger

> it seems like you've thought of most of these cases. don't throw away
> the wisdom
> incorporated in sysvinit though. i'd recommend trying to make your init
> have the same
> semantics from an external point of view where possible so that
> userspace utilities
> such as wall, shutdown, etc, keep on working.
That is definitly possible. sysvinit uses a fifo to communicate. I could
write an IO part in pinit that could handle these commands. But i could
also write a pinit alternative.

 >
> i'm not entirely sure i'd want to use XML for everything though. i think
> a nice
> approach would be to put the *metadata* about a service in XML, but the
> actual
> mechanics of starting/stopping the service in a seperate script. of this
> script,
> you expect things like that if it starts a daemon, its pid will be the
> pid you get
> when you exec it. have a well-defined interface you expect from the
> scripts,
> either return codes or text they output on success/failure, but scripts
> are
> very useful for executing a collection of commands. have the script
> start with
> #!/sbin/your-interpreter so that executing the script on its own still
> executes
> it in its managed environment.
>
Yes. I have thought of adding a script-service. But i removed it, because
i do not like scripts. I agree that they are convenient for executing a
collection of commands, but bash is a very weak programming language.

I will think about adding scripts. Maybe just to lower the difference
between my init and sysvinit. But rather not bash, bash is ugly :-(

it would be nice if people wrote more scripts using functional languages.
(haskell, miranda, etc...)

> i'm extremely interested in development of this, don't stop development
> :)
>
> regards,
> leon.
>


--
gentoo-dev@gentoo.org mailing list