Fedora’s rc becomes dependency aware

fedora-logo-bubble
Harald Hoyer, a Red Hat developer, has released a first draft with details of the possible new Fedora init system. The /etc/rc.d/rc system will become dependency aware.

Almost all distributions are working in some way or the other at improving the init system: Ubuntu develops its Upstart, openSuse tries to improve the init scripts massively, and some distributions picked up projects like InitNG.

It was generally agreed that Fedora must also do something in this regard, however the plans to improve the situation for Fedora 7 have been delayed to Fedora 8. There were some discussions but really happened, and not even benchmark tests were done.

However, now Harald Hoyer updated several related wiki pages showing the possible future: /etc/rc.d/rc will become dependency aware making it possible to boot services in parallel. Also, D-Bus will be included to get more information about the system services. Also, the window manager will be fired up quite quickly, possibly even with flexible dependencies (sometimes you need network, sometimes not), to give the user the fastest possible feedback.

It is good to see some progress – and at least my tests with InitNG show that parallel booting can improve the boot time quite a lot compared to a standard boot sequence.

However, there are several things I don’t like about these new proposals: I have the feeling that all other attempts were not even really tested – although InitNG is already part of Fedora Extras, and although I for example also made upstart run on Fedora. Why were these attempts not tested in general?
I must admit that I cannot even see reasons against Upstart, because it does deliver the features needed: dependency handling, parallel booting, a connection to D-Bus. At least the first two are also handled by InitNG.

In general I miss a real discussion of the pros and cons of the existing solutions and existing attempts to solve the problem for Fedora. Scott Remnant, the main developer behind Upstart, spent quite some time comparing existing solutions with each other and figuring out the needs of today’s Linux distributions. I have seen nothing comparable from the Fedora developers yet. And with comparable I mean a paper like discussion of the topic, not the existing (worth reading) e-mail discussions.
After all I would like to know why there should be yet another solution developed while there are alternatives like Launchd, Upstart, InitNG and also openSUSE’s attempt available. I don’t like the idea of an init war.

But maybe the changes for Fedora 8 will be more temporary until it turns out which new init system really is worth porting?

19 thoughts on “Fedora’s rc becomes dependency aware”

  1. I don’t know if that’s the reason, but Mandriva has chosen another init because all existing init replacements break Sys V init booting strategy. And thus, legacy or third party software may not work at all.

    At least, it’s the case for initng. But no one paid attention to Mandriva implementation (probably too “simple”) and ubuntu, fedora and suse started to develop their own solution although Mandriva had this for several months… It’s the same with power management system… Well, that’s life.

    Number exceeds merit.

  2. red_alert: Of course I read it – and I also contributed to it.

    But “partly discussed” describes it quite well: there was no real discussion at all. For example Upstart was “discussed” by linking to the main paper (I was the one who first reported that it can work on Fedora), and InitNG was even linked to the wrong homepage before I added that it is already in Fedora Extras and that you can also provide bootcharts with it.

    Also, the discussion at the middle of the article are not real discussions – some thoughts are written down, that’s all. And the big projects (meaning these which are already in place in larger setups, like LaunchD, Upstart or for examples Sun’s attempt) are not even mentioned.

  3. Oh well🙂 I’m sorry, I did not really have the time to investigate it that deeply😉 I’m just happy that there’s anything going on. I think the past proved, that red hat / Fedora often take the better way, even if it’s a new apporach and there are others working already. IMHO.

    Btw, did you once talk to Harald directly? Maybe he can tell you more exactly why they’re going their own ways.

  4. I wonder what about einit … it seems to be much faster than upstart… altough it’s in an alpha stage right now (afaik)… but why don’t they even think about improving and then using it?

    (sabayon already has it in his repo and it’s working really fast… atm networking and sound doesn’t work)

  5. red_alert: No, I never spoke to him directly, but if I find time I might send him some questions. If this will happen I post it here.

    LordBernhard: it is no wonder that einit is much faster than Upstart because Upstart at the moment on most machines does nothing else than running the sysvinit scripts. Upstart *has* a lot of potential by replacing current init scripts with Upstart versions of them (which are then much faster), but the replacing is evolutionary and not revolutionary.
    Also, to quote the einit page:

    eINIT is designed for embedded devices and things like headless servers.
    Upstart is designed with backwards-compatibility to sysv in mind, eINIT is not

    Both are not aims Fedora follows.

  6. “It’s the same with power management system… Well, that’s life.”

    BS. Mandrive developed its own power management solutions and used its own wireless network manager instead of working properly and coordinating on Network Manager used by everyone else.

    “LaunchD, Upstart or for examples Sun’s attempt) are not even mentioned.”

    They are clearly mentioned in the wiki.

  7. jef: that’s wrong, they are not. In the quoted part I refer to the middle part of the article in the wiki where the “discussion” takes place. And there all these alternatives are *not* even mentioned.

    They are listed at the bottom of the article – but just listing something is not what I call discussing it.

  8. “They are listed at the bottom of the article – but just listing something is not what I call discussing it.”

    You said they aren’t even mentioned which is incorrect. If you had talked about a description I might have agreed. Wiki isn’t good for discussions really.

  9. jef: Read what I wrote:
    “Also, the discussion at the middle of the article are not real discussions […]. And the big projects (meaning these which are already in place in larger setups, like LaunchD, Upstart or for examples Sun’s attempt) are not even mentioned.”
    If you misunderstood this the understanding might be incorrect, but not my words.

    And about wiki and discussion: I’m missing a discussion in a paper like fashion. I also mentioned that clearly and explicitly: and that didn’t happen. That has nothing to do with a wiki at all.

  10. “And about wiki and discussion: I’m missing a discussion in a paper like fashion.”

    Well you are not subscribed to fedora-devel. Subscribe or read the archives before complaining.

  11. Jef: Yeah, nice – what about reading yourself first? As you might see in the post there is actually a link to a discussion on the devel list dealing exactly with that topic.

    And no, there was no paper like discussion posted there.

    So yes, I read before I start complaining. Unlike you.

    So if you have any further problems, be more detailed: if I missed something, just give me a link. If you just want to troll: I will not tolerate that here anymore.

  12. “And no, there was no paper like discussion posted there.”

    wtf is a paper like discussion?

    “If you just want to troll: I will not tolerate that here anymore.”

    Hmm. Just because someone disagrees with you they are a troll?

  13. “Just because someone disagrees with you they are a troll?”
    Of course not – but a person who keeps commenting on a subject without giving any kind of usable information or pointing out what the real problem is, that is a troll.

    For example: “read the archives before complaining”
    This is totally dumb in even two ways: first of all you show that you do not read my post (because there was a link given to the most important fedora-devel discussion), and second you do not show me why I should read the devel list. If you think there are valuable information hidden regarding this topic, give me link. Show me that it would be worth (again, regarding this topic) to subscribe to the list.

    About the “paper like”, again it would help reading my post above because there the term is used in the context of the analysis Scott did before he started Upstart. That analysis gives you a pretty good impression.
    But anyway, in short paper-like refers to the term “paper” used in scientific publishing, meaning basically an essay or an article dealing with the topic in a structured way: introduction, phrasing the needs, showing the current situation, pointing out current solutions, highlighting where they do fulfill the needs and where not, and of course the detailed explanation of the new solution or of the approaches to be taken in the future to extend a current solution to fulfill the needs.

    That’s what was done by Scott – and what has also been communicated – and what I’m missing by Fedora atm.

Comments are closed.