Slower startup in 24.04 - don't understand 'systemd-analyze critical-chain'

[Chris Green]

On Mon, Oct 07, 2024 at 02:25:36PM +0200, Ralf Mardorf via ubuntu-users wrote:
> On Mon, 2024-10-07 at 13:05 +0100, Chris Green wrote:
> > I don't quite understand this, why does run-user-1000.mount start at
> > 12.981 seconds and then subsequent things are back to 5.xxx seconds?
> 
> Hi,
> 
> because systemd remains a PITA!
> 
Yes, well, apart from that! :-)


> Does
> 
> # systemd-analyze blame
> 
Not really:-

    chris at t470$ systemd-analyze blame | grep systemd
    1.643s systemd-journal-flush.service
    1.194s systemd-backlight at backlight:intel_backlight.service
     281ms systemd-udev-trigger.service
     280ms systemd-logind.service
     156ms systemd-journald.service
     118ms systemd-udevd.service
      86ms systemd-modules-load.service
      67ms systemd-tmpfiles-clean.service
      66ms systemd-tmpfiles-setup.service
      55ms systemd-fsck at dev-disk-by\x2duuid-F897\x2d5D74.service
      54ms systemd-remount-fs.service
      51ms systemd-backlight at leds:tpacpi::kbd_backlight.service
      41ms systemd-tmpfiles-setup-dev-early.service
      40ms systemd-sysctl.service
      39ms systemd-random-seed.service
      33ms systemd-binfmt.service
      24ms systemd-update-utmp.service
      18ms systemd-update-utmp-runlevel.service
      16ms systemd-tmpfiles-setup-dev.service
      13ms systemd-user-sessions.service
      13ms systemd-rfkill.service

There's nothing produced by 'systemd-analyze blame' that looks as if
it consumes enough time to account for the time the system takes to
boot to the GUI.  The top few items are:-

    chris at t470$ systemd-analyze blame
    5.737s NetworkManager-wait-online.service
    2.804s home-chris-tmp.mount
    1.643s systemd-journal-flush.service
    1.439s NetworkManager.service
    1.266s fwupd.service
    1.233s postfix at -.service
    1.194s systemd-backlight at backlight:intel_backlight.service
    1.137s syncthing at rcfg.service
     750ms dnsmasq.service
     719ms blueman-mechanism.service
     650ms apport.service
     592ms accounts-daemon.service
     ...
     ...


So what accounts for the 'run-user-1000.mount @12.981s' in
systemd-analyze critical-chain?

> show something suspect related to "systemd-fsck" or else? On my Arch
> Linux install the slowest "systemd-fsck" entry took 28ms.
> 
> FWIW on top of the blame list is "6min 8.290s fstrim.service", but this
> had no noticeable effects.
> 
That's the trouble with 'systemd-analyze blame', something like
fstrim.service doesn't prevent the GUI from starting so it doesn't
matter if it takes a while.

-- 
Chris Green