Ubuntu is developing a new init program

Gullible Jones · 2006-08-27 13:38:17

It's called upstart, and I think it might be more compatible with Arch's BSD initscripts than initng. Not sure though.

deficite · 2006-08-27 17:37:06

Looks interesting. I like the unification they're trying for. I can't wait to mess around with it.

cactus · 2006-08-27 18:33:12

Yes. It sounds quite interesting indeed.I can't wait to give it a spin myself too.

I really like the idea of an init daemon that fires off tasks, than a series of non-connected scripts. In theory, it would make it possible for tasks to be started only with the permissions they need from the get go, instead of starting as root and then releasing permissions. Since they are only recieving a notification/message/task, instead of the script itself performing the startup and deciding what it needs to do.

dtw · 2006-08-28 08:13:18

/me spies more work for self on horizon...

Dusty · 2006-08-28 14:37:16

dtw wrote:

/me spies more work for self on horizon...

Not in the long run. These initscripts are going to make every sysadmin and distro maintainer's life easy, assuming they do it as right as they say they are.

Dusty

test1000 · 2006-08-28 15:11:06

Have anyone looked into how this compares to http://smarden.org/runit/ ?

archuser · 2006-09-16 12:19:20

Would this eliminate the complaint I get from mount when booting that "/dev/sda1" doesn't exist?

It's no major problem, it just bugs me if I'm there whilst my system is starting up.

Snowman · 2006-09-16 19:42:56

What is /dev/sda1? If it's a USB device that isn't always plugged in at boot up, use the noauto optin in fstab to not have it mounted automatically on boot up:

/dev/sda1       /mnt/hd_ext    ext3 noauto,user,rw,suid,dev,exec,async   1  0

archuser · 2006-09-16 21:17:58

Yeah it's my phone's usb stick. The fstab has no auto entry so I assumed it wasn't being auto mounted but that it was complaining because the particular device didn't exist in /dev.

I'll add the noauto entry. Cheers.

Snowman · 2006-09-16 22:02:20

I guess you were using the "defaults" option which implies auto.

Cerebral · 2006-09-17 00:46:54

Wow, that actually does look really interesting - a central point for job control, whether it be from boot, from timed jobs (replacing cron), and so on.

This could be pretty cool.

phrakture · 2006-09-18 15:09:23

upstart, initng, rinit, blahblah-init.... none of them really change anything. It's an artificial problem - System V init is not "slow", it is the processes that are called that are slow.

Sure, moving from sequential to event-based is generally a good thing, but you could do that now with an existing sys V setup, using tools to work with inotify or even directly by udev.

In fact, doing this all with udev may be possible right now. The only thing you know at boot is that your root filesystem is up and running. From there, it should actually be almost trivial to use udev rules to react upon certain events. The only one missing would be network up/down.

Dusty · 2006-09-18 15:26:29

phrakture wrote:

In fact, doing this all with udev may be possible right now. The only thing you know at boot is that your root filesystem is up and running. From there, it should actually be almost trivial to use udev rules to react upon certain events. The only one missing would be network up/down.

udev rules never seem almost trivial to me, but I'm a syntax geek. Now that you've mentioned this idea, I want to know your opinion -- is it worth it to implement it? Would this be something that we might want to do in Arch (perhaps because its "simpler" than this upstart?) Or did you mention this more as an indication that the whole idea is unnecessary and ubuntu would be wise to switch to BSD style init scripts, and we're already doing everything right?

Dusty

phrakture · 2006-09-18 16:08:17

Well, I think what I'm trying to say is more the fact that this is an artifical problem. "Oh no, init is slow and out-dated!" is not true. Remove everthing from inittab except some sort of "/path/to/time /path/to/print done" as initdefault, and tell me how long it takes to run init. Now, shut off all DAEMONS, set all non-critical FSes to noauto, and set MOD_AUTOLOAD to no, then throw /path/to/time in front of the rc.sysinit call.

What upstart is trying to do defies pareto's principle - your big hogs are a) mounting large filesystems, b) running through uevents, and c) starting all the on-boot stuff you have.

Trying to optimize the order and timing of these isn't gaining much.

Now, all things considered, I think that redesign is generally good, but redesigning something that's not problematic is meerly a waste of time.

cactus · 2006-09-18 16:26:14

they are also aiming to replace cron, and the one that runs missed crons..can't remember the name right now.. (is it 'at'?).

I think it is interesting, and doing something interesting to even a few people, is not often a waste of time.. sometimes very cool and useful things can fall out of the process.

archuser · 2006-09-18 19:33:29

I've read the page about upstart and don't quite understand it.

Is it some kind of 'on demand' system? Like for example: I haven't used my network connection yet; neither has anything started up at boot that requires a network connection. So the 'network' daemon hasn't been started yet. But as soon as something starts up that requires a network connection it says "Hey, is the 'network' daemon running?". The system says, "No, hang on, I'll just start it now" and it starts. Then after a period of time at idle (the network daemon), perhaps settable by the user, the system stops the network daemon and frees the resources.

hypermegachi · 2006-09-19 05:02:31

http://blog.thelinuxfr.org/index.php?20 … t-sur-edgy

frankly...i'm unimpressed. the system in the blog is a a64 3200+, and switching to the new system gained 3 seconds...not to mention that none of the services were heavy duty like a database server. my arch system running on a p4 2ghz with more services boots 10 seconds faster than upstart on this blog system.

granted, it's still in its infant stages, it's already hit edgy, which means in a month it will become production.

FUBAR · 2006-09-19 10:36:14

cactus wrote:

they are also aiming to replace cron, and the one that runs missed crons..can't remember the name right now.. (is it 'at'?).

anachron?

deficite · 2006-09-19 12:31:06

I, for one, am not interested in speed or uber-optimization. If I was I'd go Gentoo and waste away a quarter of my free time :-P. What I AM interested in is that this system makes SENSE. Yeah, you can do a lot of things with the tools we already have, but I can think of a few examples of extremely fugly, inelegant solutions simply because nobody wanted to start from scratch on something. This may not be the case with init, but I commend them for having the guts and time to do this (I think it's a nice thing myself)

mmccaskill · 2006-09-27 13:10:55

From my understanding of the article the goal of upstart isn't speed but flexibility. I don't think the writer is complaining about the speef (or lack thereof) of sysvinit. It sounds like a good idea but I am not well versed in this area.

hypermegachi · 2006-09-27 15:08:18

loving messing around with alpha and beta stuff i installed knot 3 on my laptop.

surprisingly, it beat out the one in the blog. i got 36 seconds, and my laptop is only a 2200+....so i have no idea how that a64 was so slow.

arch booted on the laptop at the exact same speed.

we need more people with more bootcharts.

Arch Linux

#1 2006-08-27 13:38:17

Ubuntu is developing a new init program

#2 2006-08-27 17:37:06

Re: Ubuntu is developing a new init program

#3 2006-08-27 18:33:12

Re: Ubuntu is developing a new init program

#4 2006-08-28 08:13:18

Re: Ubuntu is developing a new init program

#5 2006-08-28 14:37:16

Re: Ubuntu is developing a new init program

#6 2006-08-28 15:11:06

Re: Ubuntu is developing a new init program

#7 2006-09-16 12:19:20

Re: Ubuntu is developing a new init program

#8 2006-09-16 19:42:56

Re: Ubuntu is developing a new init program

#9 2006-09-16 21:17:58

Re: Ubuntu is developing a new init program

#10 2006-09-16 22:02:20

Re: Ubuntu is developing a new init program

#11 2006-09-17 00:46:54

Re: Ubuntu is developing a new init program

#12 2006-09-18 15:09:23

Re: Ubuntu is developing a new init program

#13 2006-09-18 15:26:29

Re: Ubuntu is developing a new init program

#14 2006-09-18 16:08:17

Re: Ubuntu is developing a new init program

#15 2006-09-18 16:26:14

Re: Ubuntu is developing a new init program

#16 2006-09-18 19:33:29

Re: Ubuntu is developing a new init program

#17 2006-09-19 05:02:31

Re: Ubuntu is developing a new init program

#18 2006-09-19 10:36:14

Re: Ubuntu is developing a new init program

#19 2006-09-19 12:31:06

Re: Ubuntu is developing a new init program

#20 2006-09-27 13:10:55

Re: Ubuntu is developing a new init program

#21 2006-09-27 15:08:18

Re: Ubuntu is developing a new init program

Board footer