Thanks for the link Brian. Very valuable. Have you discussed moving cntvct-log<file:///CentOS/automotive/src/boot-time-analysis-tools/-/tree/main/cntvct-log> to systemd-analyze? IMHO, systemd can have better control when the target is reached to record the wall time.
Umut From: systemd-devel <systemd-devel-boun...@lists.freedesktop.org> on behalf of Brian Masney <bmas...@redhat.com> Date: Thursday, 26 September 2024 at 03:07 To: dharm...@microchip.com <dharm...@microchip.com> Cc: sereni...@inventati.org <sereni...@inventati.org>, systemd-devel@lists.freedesktop.org <systemd-devel@lists.freedesktop.org> Subject: Re: [systemd-devel] Exploring Minimal Systemd in Initramfs for Faster Boot On Wed, Sep 25, 2024 at 03:53:36PM +0000, dharm...@microchip.com wrote: > yes something similar to this, I will experiment this and get back to you. > > and I think since the egt service and its libraries depend on the full > rootfs, integrating initramfs might not provide significant benefits in > terms of faster launch time. The time saved by using an initramfs to > launch basic services earlier would likely be offset by the delay in > mounting the root filesystem, which is necessary for accessing the egt > libraries. There are some drawbacks to putting your early service in the initramfs and having it persist across the switch root to when systemd is started again from the root filesystem. - The initrd is a cpio archive, and increasing the size of the initrd is going to increase the kernel boot time since it will need to uncompress and extract the larger cpio archive. - Any services started from the initrd will be started before the SELinux policy is loaded. Services started from the initrd will run with the kernel_t label. - Services started from the initrd can't depend on almost anything like mounts, devices, services, dbus, etc so it's difficult to develop software of any complexity. - Adding all of these dependencies to the initrd is only going to move the timing bottlenecks booting from the root filesystem to the initrd. - Directives like Restart=always will likely not work as expected. Personally I think that you should have the initramfs be small enough to only get you to the root filesystem and start your early service from the root filesystem. Shameless self plug: Here's a presentation that I gave last month at DevConf.us in Boston with Eric Curtin and Ed Chong about some of the boot speed optimizations that a group of us worked on at Red Hat: https://pretalx.com/devconf-us-2024/talk/HXVPHC/ . The talk also lists some upstream patch sets that you will want to consider including in your kernel to help with boot time. Brian
