Announcement

Collapse
No announcement yet.

Debian To Switch To Systemd Or Upstart

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • The init system should be the last thing people should have to be wasting time deciding on about. Both systems work just fine. Upstart seems to have better backwards compatibility . SystemD seems to have better process logic .

    Comment


    • Originally posted by jrch2k8 View Post
      well as an gentoo user i admit openrc is sysV with nitro but after test systemd i just nuke openrc from my system entirely[just keep certain bashies needed for eselect that gentoo devs insist in keep them inside openrc instead of eselect ebuild]. the difference is just too huge to go back.

      I use RHEL 6/ubuntu too in some clients of mine and everytime i have use upstart i feel like im handling old days HPUX or Caldera.

      I mean the difference is big enough to inform my client's IT departments that for the next update they need a distro with systemd or it just won't work[thing that will help me remove around 7000 lines of code needed to deal with every sysV like nightmare out there + add amazing stuff like sandboxing/safe trackeable process spawn, engine switch when virtualized, focused logs, resource jailing, bullet proof child process clean, reliable subprocess spawn under demand(OMG), session integration(OMFG), timed tasks through dbus(no more f/d/g/h/u/j/k/l/cron lucky shots), self restart when crash + reliable logs through journald/dbus to send me back, 1 only one service file for any distro(with systemd ofc) OMG only 1 yes 1 i really mean 1 no tricks is realy really just 1]

      i do admit tho systemd more important goodies are for devs or transparent to the user
      OpenRC will be shifting to the new BSD init system which offers far more features than the standard SysVinit provides.
      This coupled with cgroup support, makes it a more enticing choice whilst not breaking compatibility with the other Debian kernel projects.

      Comment


      • From a purely user standpoint:
        1. custom Systemd scripts have been easier to write than Upstart scripts
        2. Upstart cannot load my screen brightness settings on boot for god knows what reason, but Systemd has no problem with it.

        The second point surprised the hell out of me. I won't ever go back to Upstart if only because now I don't have to turn-up my brightness from the lowest setting every. single. boot.

        Comment


        • Originally posted by Daktyl198 View Post
          From a purely user standpoint:
          1. custom Systemd scripts have been easier to write than Upstart scripts
          2. Upstart cannot load my screen brightness settings on boot for god knows what reason, but Systemd has no problem with it.

          The second point surprised the hell out of me. I won't ever go back to Upstart if only because now I don't have to turn-up my brightness from the lowest setting every. single. boot.
          I can confirm about the second point as I run an updated Fedora 20 Beta Test Compose 6 which also surprised me as well. That part is great especially for embedded device like smartphone or tablets.
          Systemd will really benefit GNUX Linux Debian at long term. Other variant like Debian kFreeBSD could use launchd which is similar to systemd.

          Comment


          • Originally posted by intellivision View Post
            OpenRC will be shifting to the new BSD init system which offers far more features than the standard SysVinit provides.
            This coupled with cgroup support, makes it a more enticing choice whilst not breaking compatibility with the other Debian kernel projects.
            this make no sense, can you elaborate

            Comment


            • Originally posted by Honton View Post
              OpenRC = Ron Paul. Do everybody a favour and stop your campaign. This is only about systemd and Upstart.
              only the article says this.

              Debian Developers do not think so: https://wiki.debian.org/Debate/initsystem/openrc

              Comment


              • Originally posted by Honton View Post
                OpenRC = Ron Paul. Do everybody a favour and stop your campaign. This is only about systemd and Upstart.
                Why did we have to parallel politics...
                As if blood wasn't boiling enough already on the topic of init systems, we start launching nukes.

                People (in the US at least) are so angry right now... it has become virtually impossible to think positively about the future. We've given up on issues they fought extremely hard against and are now just bracing for things to get much, much worse.

                Comment


                • My thoughts on systemd, etc..

                  The ideas behind systemd are pretty strong:

                  1. Easy to write, unified (close to a "standard") way of handling startup.

                  2. Fast start and stops.

                  3. Gets rid of a lot of replication of critical resource handling (e.g. socket handling, etc.)

                  4. Better association of daemon resources.

                  5. Targets vs. limited runlevels

                  Things I like about systemd are the fast startup and fast shutdown. But, everything is still very "beta", and I have a system right now where the kernel crashes badly on shutdown occasionally. Blame targets, etc... but ultimately, things are expected to just work right.

                  With regards to #1 and #5, way too many assumptions are being made. So, while it's nice to say that in general all systemd implementors will tend to use the config and targets provided, I think way too many assumptions are being made. To me it's more complicated (sorry). I mean, we can argue that having access to just a few "runlevels" in the old sysvinit days was a problem, but now we have the idea of targets... and in some ways, it's a dependency mess. So... while default targets may be provided as an example, I've found them to be difficult to work with so far with regards to building something very custom (and like it or not, Linux is the master king of custom). Runleves represented limit "states"... targets are actually a completely different paradigm, though you could argue that it's possible to create the idea of "states" using targets.

                  With regards to Debian's decision. Sounds like upstart wins because the votes are "stacked". With that said, because of the more intimate nature of systemd, I fear that anyone not using systemd will eventually find themselves having to fork too many things (if not programs, at least concepts). Not sure of the long term viability of staying away from systemd.

                  Systemd is still baking btw. Lots of bugs and even features being addressed.

                  And finally...

                  I personally believe the hardest hit will be enterprises where sysvinit is replaced with systemd. While systemd claims compliance with sysvinit (that is, they supposedly cover all of the scope of sysvinit even to the point of allowing you to use your "old school" scripts)... my experience is that because systemd isn't a mere "init" replacement, that enterprises will find difficulty just dropping it in and they will be forced to modify many things (and in the process will find the "hole" in the systemd design where they aren't quite able to handle everything... which makes sense, you can do quite a bit with old style sysvinit). In many ways, systemd is better, but it's better at giving you the features that you worked around (likely) a long time ago. So... I feel that the impact to enterprise customers when RHEL7 and SLES 12 are released will be pretty high. I'm not looking forwards to the fall out that is going to happen in those two releases.

                  Could things have been handled better? Not sure. I mean systemd brings many good ideas to the table. But the implementation is still a bit too radical (and complicated and buggy), no matter what the systemd folks say. There will be problems in deployment. Of course, there are other things that scare the pants off me as well, like KMS and such where support is being dropped for things that are actually needed in the enterprise. We'll see. Who knows, maybe the economy will rebound, everybody will buy "new" equipment and my fears will be all for nothing.... talking mostly about the KMS decisions, I think we're all going to struggle with systemd for at least a couple of years after the enterprise distros come out. Question: Is that ok?

                  Comment


                  • Originally posted by kazetsukai View Post
                    Why did we have to parallel politics...
                    There will be a vote, and politics is what happens before a vote.

                    Although the comparison to Ron Paul is flawed, as Debian typically uses Condorcet style voting (Cloneproof Schwartz Sequential Dropping) and not the anachronistic methods employed in public voting.

                    Comment


                    • debian debate

                      For a nice wiki-style debate with pros and cons for the different options, see:



                      far less polarized than the phoronix debate

                      Personally I think openrc would be a good match, but we will see what they decide

                      Comment

                      Working...
                      X