Ubuntu Systemd Plans: Will Be Default By Ubuntu 16.04 LTS

[steveriley]

Regarding systemd migration sequence and timing... anyone remember how did Arch do it? I didn't have an Arch system built during the switch, so I wasn't able to witness it. But I don't think it took all that long.

[Skrapion]

Yes, it's true for systemd too. Actually, systemd is much better at that sort of functionality. It's not limited to task-level dependencies like Upstart is; it can launch a service when, for instance, a file is created, or a socket is opened. In addition to ensuring that everything the service needs actually exists at launch time, it also allows you to do some neat things with late launching. For instance, you can set it up so that sshd isn't constantly running, but instead only runs when somebody connects to the external socket.

[rgap]

systemd achieved 2-second boot times two years ago. Is that fast enough for you?

Well it all depends what boot time UpStart has in comparison.

My personal experience with boot times is that systems that delay processes boot a lot faster than those that try to run everything at once. So I was curious if this is true for SystemD. Maybe it isn't. This is what I was wondering.

[Skrapion]

systemd achieved 2-second boot times two years ago. Is that fast enough for you?

[rgap]

Will SystemD on Ubuntu increase the boot time?

I know Lenovo has resorted to using an UpStart like system where process loading is delayed until booted. Anybody know if SystemD will result in longer boot times on Ubuntu 16.04?

[Skrapion]

With this and the Mir story, I guess 16.04 is the new Ubuntu pie in the sky? The date they put on anything they don't know when they can have finished?

All it really means is "It's not going to be in 14.04". 16.04 is the next LTS release, so that's the next "important" release. They're not overly concerned about what goes into the interim releases.

[Jumbo21]

They're waiting for Debian to do the job, so that they can copy/paste it

Not surprising, Canonical takes forever to do anything that someone else hasn't already done for them. And even then, they take ages to finish anything. So glad I don't use Ubuntu, or any distro based on it

[Luke]

Let's just hope they get the migration path right.

They're waiting for Debian to do the job, so that they can copy/paste it

The real issue will be figuring out the migration path for those who are not reinstalling. This probably means some way of ensuring that packages that are started by Upstart still work until they get updated. Fortunately most top level apps (seemingly ALL that I have) do not install files in /etc/init with Upstart, thus should be init system agnostic.

It's system packages that will be the issue here, all the DE's will need to cope with this at the same time. Fortunately it seems that the display managers handle them, and once lightdm/gdm/kdm/whatever is opened by init the DE's don't seem to rely on the init system any further, I've got Cinnamon, gnome-shell, IceWM, and KDE all installed, and don't see anything in /etc/init that appears to be from any of them.

If things get ugly, an upstart compatability job like the currents sysVinit compatability job might be a workaround, but the now global Linux transition to systemd should put pressure on system software developers to get off Upstart fast once it is deprecated.

If they do this right, it should be possible to pin the init system when the changes begin, unpin it when the system packages that are started by init are ready, and get an upgrade that works. Certainly I would take a fresh OS snapshot (a dd copy of the root partition) just before starting an init system changeover.

I will have to learn enough about sytemd myself to rewrite my boot message scripts that give notice in Plymouth of disk mounting events. Right now they are Upstart conf files in /etc/init. SysVinit can't do the same job as it does not respond to external events, I will need a way to write the files for systemd to detect mounting of a disk and run plymouth-message

[Spittie]

Wow, this is in two years. Is it really going to take this long to migrate? I'd say a few months should be enough to get it done. And they basically have most of the work already done in Fedora and other systemd distributions.

They're waiting for Debian to do the job, so that they can copy/paste it

[nerdopolis]

The link to the session notes points back to the Phoronix article.

And the link still isn't fixed. Can anyone find the oeiginal link?