Peter Duffy Explains SystemD
Ein Volk, Ein Reich, Ein Führer! (it is not optional)
"Apropos the most recent [systemd] postlette from today," one reader told us, pointing to a cache of new criticisms. "Some of the longer e-mails there have material for a proper post. One of the key questions raised is why systemd is designed to be mutually exclusive to all other approaches."
See Version 256 of systemd boasts '42% less Unix philosophy'.
Lots more at dyne.org in recent days, but here's one that stands out:
Just to be clear . . .
The only thing which I said (at least I hope I said) was that for me, the elephant-in-the-room question is: why is systemd not optional? That's the question which in my view should be asked of anyone who evangelises about systemd or thinks it's a good idea. I have asked it a few times, and never got an intelligible or intelligent reply.
I dislike systemd for all the reasons usually given, and a few more besides. But I dislike it the most for not being optional. I see that as genuinely tragic. If ONLY it had been optional! Those who liked it could use it. Those who didn't could use something else. It would have been a win-win, because LP and co. would have got constructive criticism and maybe even praise, instead of endless hatemail, and as a result, systemd might have actually got better (yes, I know - it's difficult to contemplate that possibility). And the linux community wouldn't have been divided against itself.
I've never found the "faster boot time" argument as in any way convincing, and obviously I've not used systemd-based distros enough to know whether it's borne out in practice. I've always assumed that it was probably based on the obsession with parallelising the startup of processes - which to me always seemed a bad idea anyway. Say that for some reason, a disk or other device responds a few milliseconds more slowly than usual: that could cause a knock-on effect which at worst might stop a system from booting. If processes start in an orderly and predictable way as configured, something like that might have have less impact. (I did once have to deal with a box which would only boot intermittently: it was running ubuntu, and I eventually found that someone had decided to "upgrade" it to a systemd-based release. From what I could make of the logging information (not a lot), it appeared that something like the above was happening. Once the box was rolled back to a pre-systemd release, the problem went away.)
The thing which permanently set me against systemd was that when I first tried it (testing RH7), the box wouldn't shut down. It just gave a message about "asserting shutdown state" or some such pompous baloney, and then hung. I had to hold down the power button. When the system came back up, it complained about ungraceful shutdown and ran a full set of disk checks. Maybe not so bad if the box is on my desk. Rather more of a problem if it's a mission-critical server in an unmanned server farm, I'm trying to sort out a major problem on it, and it's 03:00 in the morning. (Strangely enough, I noticed on one of the FB linux groups only a few days ago that someone else had just had the shutdown hang problem.)
systemd is clearly not an init system. It's more like a comprehensive replacement for everything, developed in Microsoft's proprietary platform and led by a Microsoft employee. █
