Announcement

Collapse
No announcement yet.

Fedora 38 Change Approved To Mandate Quicker Reboots/Shutdowns

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

  • Fedora 38 Change Approved To Mandate Quicker Reboots/Shutdowns

    Phoronix: Fedora 38 Change Approved To Mandate Quicker Reboots/Shutdowns

    Last month a change proposal was filed for aiming to yield faster reboots and shutdowns of Fedora Linux by shortening the time window that services can block the shutdown process. A modified version of that change proposal has now been cleared by the Fedora Engineering and Steering Committee...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Michael
    For whatever reason I can't post on that Fedora 38 thread below. Just wanted to say there:

    That sounds like a nice change because one of the worst annoyances that I have on Linux is when I go to reboot and some random Wine process from a Steam game holds the process up.

    Now I wonder why I haven't made that change myself....

    Comment


    • #3
      I still think 15 seconds would be more than enough. Anything specific that needs longer (perhaps httpd or database waiting for a connection to finish) can generally have a custom set timeout anyway. The common SysV init defaults felt more sane here.

      Comment


      • #4
        To be honest, on modern systems IO is capable of 100mb/s and above. So within 10 seconds, a process should be able to do > 1GB of IO and other things. Yes, this depends on many other things and this number is pretty arbitrary, but IMHO no regular application/service quitting should not take more than 10–15 seconds.

        One example I know is I2P which tries to gracefully migrate clients using it's tunnels to other tunnels by not accepting new one, but letting the old one live for a while. Cutting of tunnels harms your rating on the network. Tho this is only minimal and a reboot on such a system is rare - so even there I would not justify more than 15s.

        I'd be even happy to have way less then 45s as the default, like 30 or 20 seconds.

        Comment


        • #5
          Originally posted by kpedersen View Post
          The common SysV init defaults felt more sane here.
          And they were…?

          Comment


          • #6
            Originally posted by zdzichu View Post

            And they were…?
            By default, 3 seconds. SysV-init was much faster than Systemd

            Much shorter than FESCo's (somewhat arbitrarily) 15, 45 and 120 seconds. But honestly if a process isn't finished after 3 seconds (and doesn't have a custom longer clean shutdown timeout because its a server) means it is probably broken. So why wait longer?
            Last edited by kpedersen; 18 January 2023, 05:02 PM.

            Comment


            • #7
              Fedora 38 Change Approved To Mandate Quicker Reboots/Shutdowns
              It's good to see that they've finally decided to drop systemd.

              Comment


              • #8
                Originally posted by Draget View Post
                To be honest, on modern systems IO is capable of 100mb/s and above. So within 10 seconds, a process should be able to do > 1GB of IO and other things. Yes, this depends on many other things and this number is pretty arbitrary, but IMHO no regular application/service quitting should not take more than 10–15 seconds.

                One example I know is I2P which tries to gracefully migrate clients using it's tunnels to other tunnels by not accepting new one, but letting the old one live for a while. Cutting of tunnels harms your rating on the network. Tho this is only minimal and a reboot on such a system is rare - so even there I would not justify more than 15s.

                I'd be even happy to have way less then 45s as the default, like 30 or 20 seconds.
                I think you meant 100 MB/s (megabytes) not 100 mb/s (mebibits). But that's only on raw uncompressed throughput. Once you add various compression algorithms you quickly become CPU bound even on extremely high throughput NVME storage. I've seen LZMA2 with 12 threads drop down to under 20 MB/s on certain kinds of files.

                Not saying that systemd timeouts shouldn't be trimmed down. They should be! 90+ s timeouts on shutting down certain notoriously bad programs is ridiculous for any measure! Just that certain assumptions also need to be checked at the door, like possible versus actual throughput.
                Last edited by stormcrow; 18 January 2023, 05:16 PM.

                Comment


                • #9
                  Originally posted by kpedersen View Post

                  By default, 3 seconds
                  3 is just the default run level noted there. There isn't a default timeout for sysvinit, certainly not one shared by distributions.

                  Comment


                  • #10
                    Originally posted by kpedersen View Post
                    I still think 15 seconds would be more than enough. Anything specific that needs longer (perhaps httpd or database waiting for a connection to finish) can generally have a custom set timeout anyway. The common SysV init defaults felt more sane here.
                    Which is why I think that even 5-10 seconds is plenty.

                    -- I have the honour to be,
                    Your most obedient servant,
                    London, January 20, Jasper W. Rogers.

                    Comment

                    Working...
                    X