Let's boot with Upstart (GDM ready in 15s)

[jdong]

Golly.

A fair bit of that sounded like sorcery/black magic to me!

Sorry I didn't mean to direct all that at you I was just trying to make everyone who is interested aware of some of the things we'd have to do.

I've registered UpHack in Launchpad: https://edge.launchpad.net/uphack and pushed up two bzr branches, including my event.d.

Is you're setup pretty standard? Could I just import all your scripts and give it a blast. Is that advisable.

That's exactly how I plan on spinning the first LiveCD. The most obvious problem I can see is that Readahead needs to be disabled when booting off the LiveCD; I think I'll have the readahead script grep for the bootup arguments indicative of a LiveCD and abort if found.

Before I spin a livecd, I'd like to reimplement all my event.d files in reference to a very standard (i.e. freshly installed Hardy) system.

I may need some help implementing some of this stuff. As I've no idea at all about it. I'm keen to learn though.

Sure. I don't expect anyone to understand what I'm saying right now until I set up some rudimentary documentation on a wiki page. Not only is Upstart very sparsely documented, a lot of these scripts are the creation of myself with zero documentation whatsoever.

I'm a bit busy these few days, but expect sometime over the weekend to get another big push of data.

[bobbocanfly]

Would definately give this a test on a spare partition, might even be able to help out in some way. That boot speed is brilliant.

[smbm]

Sorry I didn't mean to direct all that at you I was just trying to make everyone who is interested aware of some of the things we'd have to do.

I've registered UpHack in Launchpad: https://edge.launchpad.net/uphack and pushed up two bzr branches, including my event.d.



That's exactly how I plan on spinning the first LiveCD. The most obvious problem I can see is that Readahead needs to be disabled when booting off the LiveCD; I think I'll have the readahead script grep for the bootup arguments indicative of a LiveCD and abort if found.

Before I spin a livecd, I'd like to reimplement all my event.d files in reference to a very standard (i.e. freshly installed Hardy) system.




Sure. I don't expect anyone to understand what I'm saying right now until I set up some rudimentary documentation on a wiki page. Not only is Upstart very sparsely documented, a lot of these scripts are the creation of myself with zero documentation whatsoever.

I'm a bit busy these few days, but expect sometime over the weekend to get another big push of data.

Cool.

I'll keep an eye on the Launchpad page.

[mech7]

hmm with me the booting is fast... its the logging in and actually loading the desktop thats soooo slow

[jdong]

hmm with me the booting is fast... its the logging in and actually loading the desktop thats soooo slow

I'm working on a way to "fix" that too but it's unrelated

[bash]

Wow really neat work. Things like that really awake the geek in me. I'm really curious about this and always found boot up to be something you could definitly tweak out a couple of seconds. Now I don't know anything about hacking my way around the whole startup process. But I quite like to learn (by doing ) it. Except the next 3 weeks I'm packed with big uni papers. Guess no fun for me.

What comes to mind though is that you basically have to modify every process that is started during booting. Now you got that setup for your system and the idea of a LiveCD is great as well. But wouldn't you run intro troubles if the user would start the install others programms that get started at boottime. For example I have got a mysql server running on my desktop box because I also use it as a MythTV setup. Wouldn't these programs "just come" with the "old way" of booting. Or can upstart be setup more generally so that it doesn't matter in the end what services/demons your start during boottime? (This is me asking, not knowing the whole bootup system particularly good)

[jdong]

. But wouldn't you run intro troubles if the user would start the install others programms that get started at boottime. For example I have got a mysql server running on my desktop box because I also use it as a MythTV setup. Wouldn't these programs "just come" with the "old way" of booting. Or upstart be setup more generally so that it doesn't matter in the end what services/demons your start during boottime? (This is me asking, not knowing the whole bootup system particularly good)

That was what I meant about #4 on the 2nd level TODO. Currently, you're right, that job would be ignored. We need to either:

(1) Look in rc2.d for scripts we don't know about, and notify the user with information on what he can do to get Upstart to recognize them.

(2) #1, but automatically convert it to an Upstart job

(3) At the end of Upstart boot, execute all SysV jobs that aren't Upstart Jobs sequentially.



I need to think this through more, as each of those options is appealing in its own way.

[macogw]

That was what I meant about #4 on the 2nd level TODO. Currently, you're right, that job would be ignored. We need to either:

(1) Look in rc2.d for scripts we don't know about, and notify the user with information on what he can do to get Upstart to recognize them.

(2) #1, but automatically convert it to an Upstart job

(3) At the end of Upstart boot, execute all SysV jobs that aren't Upstart Jobs sequentially.



I need to think this through more, as each of those options is appealing in its own way.

I like #3. The other two...automagic generation of config files...sound a bit scary.

I'm totally up for playing with this though. I should take a dpkg dump first though, so I can do a set-selections and reinstall my packages if reinstalling Ubuntu becomes necessary

[igknighted]

Why not a combination of all three? Give the user a setting to be notified of how to change it (#1) or to have it automagically converted (#2). Then in the meantime #3 could be a default behavior (which I suppose should also be able to be turned off) that would boot the scripts that haven't been added yet.

[Meson]

Is it possible to just implement this as rc3,4,or 5? It should make it easier to test on the same exact system rather than installs on seperate partitons (which have different radial distances along the disk - dunno if that is negligable or not).