TL;DR: systemd breaks badly (at the moment) when having to deal with non-standard setups (ZFS, cryptsetup)...
As for the eternal systemd debate... if someone as knowledgeable as the author (his publications: http://www.complete.org/Publications) can't debug it... we'll have fun times ahead.
> TL;DR: systemd breaks badly (at the moment) when having to deal with non-standard setups (ZFS, cryptsetup)...
Come on, it's a little more specific than that. I have machines with zfs, machines with cryptsetup, all of them run arch and systemd, never had any problem. But of course, I don't have sysvinit scripts.
> TL;DR: systemd breaks badly (at the moment) when having to deal with non-standard setups (ZFS, cryptsetup)...
So you're saying that upgrading a custom tweaked configuration grown over the years to be just perfect for that user is going to break when converted to non-customized stock systemd packages?
The configuration in question has grown over the months and likely used multiple iterations to get right. You can't expect a stock systemd package to be a perfect drop in replacement.
What we really need in order to form an opinion of how badly or how well systemd is going to work, we need real-world examples of systems being upgraded into which only minimal tweaking of the init system has gone.
Like all the Arch and Fedora users who are not really complaining about systemd making their systems unstable or causing issues with day to day usage.
I should be able to expect exactly that, because systemd's supposed ability to work as a drop-in replacement for sysvinit[0] was touted. But suddenly it's unreasonable to expect it to live up to its promises. Convenient.
No. I think the linked article is pretty clear where the problems are. It's basically assumptions made by systemd that are not always true and from the looks of it a possible bug regarding the ordering of service files and reaction to events. Besides that broken or unusable console output without plymouth (I can't really tell from the article if systemd is responsible for that).
That's all fine and possible expected.. my only concern is just that this appears to be very difficult to debug. The author is as knowledgeable as you can hope for in a Linux user and is unable to get meaningful debug output.
All of these things are IMHO legitimate concerns that should be addressed by the systemd community or the Debian systemd packagers.
> we need real-world examples of systems being upgraded into which only minimal tweaking of the init system has gone
I am in that group (having a bunch of debian servers, ranging from "~5 years old upgraded one release at a time" to "~6 months old and installed from scratch"); converted them all to systemd; have found it to work fine and I am enjoying the extra features :)
TL;DR: systemd breaks badly (at the moment) when having to deal with non-standard setups (ZFS, cryptsetup)...
As for the eternal systemd debate... if someone as knowledgeable as the author (his publications: http://www.complete.org/Publications) can't debug it... we'll have fun times ahead.