Slashdot Mirror
Ask Slashdot: Can You Say Something Nice About Systemd?
ewhac writes: "I'm probably going to deeply deeply regret this, but every time a story appears here mentioning systemd, a 700-comment thread of back-and-forth bickering breaks out which is about as informative as an old Bud Light commercial, and I don't really learn anything new about the subject. My gut reaction to systemd is (currently) a negative one, and it's very easy to find screeds decrying systemd on the net. However, said screeds haven't been enough to prevent its adoption by several distros, which leads me to suspect that maybe there's something worthwhile there that I haven't discovered yet. So I thought it might be instructive to turn the question around and ask the membership about what makes systemd good. However, before you stab at the "Post" button, there are some rules...
Bias Disclosure: I currently dislike systemd because — without diving very deeply into the documentation, mind — it looks and feels like a poorly-described, gigantic mess I know nothing about that seeks to replace other poorly-described, smaller messes which I know a little bit about. So you will be arguing in that environment."
Nice Things About systemd Rules:
Bias Disclosure: I currently dislike systemd because — without diving very deeply into the documentation, mind — it looks and feels like a poorly-described, gigantic mess I know nothing about that seeks to replace other poorly-described, smaller messes which I know a little bit about. So you will be arguing in that environment."
Nice Things About systemd Rules:
- Post each new Nice Thing as a new post, not as a reply to another post. This will let visitors skim the base level of comments for things that interest them, rather than have to dive through a fractally expanding tree of comments looking for things to support/oppose. It will also make it easier to follow the next rule:
- Avoid duplication; read the entire base-level of comments before adding a new Nice Thing. Someone may already have mentioned your Nice Thing. Add your support/opposition to that Nice Thing there, rather than as a new post.
- Only one concrete Nice Thing about systemd per base-level post. Keep the post focused on a single Nice Thing systemd does. If you know of multiple distinct things, write multiple distinct posts.
- Describe the Nice Thing in some detail. Don't assume, for example, that merely saying "Supports Linux cgroups" will be immediately persuasive.
- Describe how the Nice Thing is better than existing, less controversial solutions. systemd is allegedly better at some things than sysvinit or upstart or inetd. Why? Why is the Nice Thing possible in systemd, and impossible (or extremely difficult) with anything else? (In some cases, the Nice Thing will be a completely new thing that's never existed before; describe why it's good thing.)
We will assume out of the gate that systemd boots your system faster than ${SOMETHING_ELSE}, so no points for bringing that up. Bonus points are awarded for:
- Personal Experience. "I actually did this," counts for way more than, "The docs claim you can do this."
- Working Examples. Corollary to the above — if you did a Nice Thing with systemd, consider also posting the code/script/service file you wrote to accomplish it.
- Links to Supporting Documentation. If you leveraged a Nice Thing, furnish a link to the docs you used that describe the Nice Thing and its usage.
By
eldavojohn
Systemd has a nice ring to it. The way the syllables roll off my tongue pleases me greatly. It could be the title of a great anime series. It could even be the lost name of an ancient forgotten god-king. It might even be the name I give my first born. It sounds much more authoritative and genuine than sysvinit or upstart or inetd. For instance from my non-technical fourth grade perspective this is what I interpret the others to mean:
Contrary to your base assumptions, systemd does not actually boot faster on my Pentium II (Intel inside) system. I just like the way it sounds.
My work here is dung.
I've done migrational upgrades to System-d with Arch Linux with zero problems in addition to using it with new installations. It works fine, and I'm still really confused about the jihad-level hatred it seems to engender in some people.
AntiFA: An abbreviation for Anti First Amendment.
Betteridge's law of headlines is an adage that states: "Any headline which ends in a question mark can be answered by the word no.
Is this the right headline to apply it on?
I like that in the future when the integration is more complete, I'll be able to install a database or a webserver and then once in a full moon when a cosmic ray hits the process and kills it, systemd will just restart it.
Yes, you can do that with other tools too once you've learnt your lesson (many years ago I had 1.5 year uptime with Apache and then it suddenly crashed) and I am using one of those at the moment. What I like is that this will just work out of the box for newbies and veterans alike with no clunky configuration/interfacing.
If you have a service called 'foo', then 'systemctl status foo' not only shows you whether the service is running, it also shows you the last 10 log entries created by that service. This is great when the service failed; usually the error message will be right there.
How does it do this? Well, because all processes created by the service are in the same cgroup, all of the log messages (and even anything they print to stdout, which would have been lost otherwise) can easily be tagged with the service name (and a bunch of other metadata). You can use journalctl to query the journal for the logs from a specific service, and systemctl status does this for you.
Using the information gleaned from this year old bug I was able to create a Tower of Hanoi solver using the dependency resolver's attempts to determine whether a NFS filesystem should be mounted after the network comes up or not. Every attempt to start Apache (which depended on the filesystem) represents moving a disc to the middle tower.
My problem with things like systemd is not that they have nice features - lots of things have nice features. Windows' Shadow Copies is a lovely feature that's a lot easier to configure that some alternative equivalents, etc.
It's that they put those nice features into some new paradigm of configuration when you could have just added them to the existing system.
'systemd' needs to stay optional, and I mean that explicitly. optional. Not "default", and not "optional, in the sense that you then have to do the maintainer's job for them because they are too lazy to consider people not using systemd, because systemd is the default, and the maintainer does not want to consider the people that dont want systemd, regardless of the reasons or circumstances." kind of way.
Systemd is potentially useful for only a subset of the linux ecosystem, and forcing this kind of change is a bad thing.
Please allow me to explain why:
Systemd seeks to be a "Be all, end all solution to system initialization", which ultimately means that it will itself try to cover every possible thing that its maintainers believe needs to or should happen during system init. That in and of itself means that it will be large and cumbersome; exactly the things that embedded linux should avoid, where ultra-minimalism is king. (We are talking systems that have just a few dozen megabytes of memory, and just a few hundred megahertz of processor power at the most. Having all that gobbled up by the init system as soon as power is applied is not going to win you any trophies, and boldly asserting that embedded devices need to obey a desktop paradigm is throwing the baby out with the bathwater.)
This is especially true with "reference distributions", like debian. Debian "console only" deployments with tools like debootstrap are reasonably common with embedded devices, as are deployments that make use of chroots for specific sandboxed services. A chroot does not need a full blown init like systemd. It is best served with a simple init script. Building a distro with the intention of killing simple init, and replacing it with a monolithic solution like systemd will make service daemons much more difficult to control in this way, and will actually rob core functionality away from the distro that goes that route--- exclusively in favor of desktop flavored deployments.
Linux is more than desktops.
Linux is routers.
Linux is home automation systems.
Linux is servers performing specialized functions.
Linux is so much more.
Please be more considerate about trying to force systemd into debian. Optional is OK. Optional like "gnome vs kde vs xfce vs $ManagerHere". Not "optional" like "unity on ubuntu". Debian is a reference distro, upon which many other distros are based. It has already found its niche in the linux ecosystem. Please dont try to reinvent it.
The idea of booting services in parallel is nice, but the problem is that apparently it doesn't have a way to specify that you need to wait for a dependant service to hold off until the initialization of the dependancy is complete. Systemd considers it "booted" as soon as it launches, which causes people problems with unreliable network initializations and such that have been resolved for Sys-V style init scripts for years (if not decades.)
I do not fail; I succeed at finding out what does not work.
Reliable servers don't just crash.
Obviously you haven't learned your lesson yet. Hardware failures happen.
http://uselessd.darknedgy.net/ProSystemdAntiSystemd/
This comes from an "anti-systemd" source, but tries to cut through a lot of the controversy and hostility shown on both sides. Bear in mind, you only see the anti-systemd view on Slashdot, but you get just as much idiocy on the other side as well. For example, the Poettering "death threats" were actually a joke made by a bunch of people in an IRC channel. Here, read the log (ctrl-F "hitman"):
http://logs.nslu2-linux.org/livelogs/maemo/maemo.20130215.txt
What the fuck, you think you can set rules for discussion on Slashdot?
Sadly he can't force the rules on anyone, no matter how useful those rules would be for such a debate. So, instead of actually doing what the poster wants, we instead get comments like yours which add absolutely nothing to whether or not systemd is any good.
And yes, it's a question I was considering posting on too. It would be nice to see some useful information.
So thanks for clogging the top level posts up with useless, pointless crap. You have done your fellow slashdotters a great service showing you can be angry on the internet.
SJW n. One who posts facts.
I've been starting to migrate many of my services at home to containers to make them a bit easier to maintain (a bit of a tangent - having 5 containers instead of one host with 5 services means that you have to do 5x as many updates, but each update can at most break one thing at a time). This was trivial to do with systemd-nspawn.
With a command line that barely fills a terminal line I can launch a container, have it boot systemd inside the container, have a few bind mounts, and have it get its own IP like a lightweight VM. Within the container systemd just does whatever it is told to do, like launch ssh so that I can get in, configure the network, and launch whatever services the container was intended to provide. The container journal logs are symlinked back to the host log directory, so they're really easy to look at from the host.
Sure, you can do similar things with docker, but doing it with systemd involves less tooling in general.
Also, for simpler situations systemd-nspawn makes a VERY good substitute for chroot. In addition to doing everything chroot does, it starts a separate process namespace so you don't see outside processes from inside the container. It also automatically sets up /dev for you, sets up resolv.conf, etc - it can do all this while just spawning one program inside just like chroot does (so no need to run systemd inside). It can also set up bind mounts if you ask it to. When you exit it cleans up - no lingering bind mounts, or tmpfs, or /proc and such inside. Also, any mounts inside the container aren't visible outside, so you can run a backup on your chroot and not have it follow bind mounts, or try to save /proc/kcore or whatever. In fact, you could spawn 5 containers inside the same directory and they can have private /tmp and /dev and /proc while seeing changes each other make in the files in the actual chroot.
With more people jumping to FreeBSD, we might get better hardware support over here.
Prediction for end of Universe #42: Fencepost error in Quantum_bogosort.cpp
One thing I really like about systemd is that when you stop a service, it actually stops.
I used to run monit with openrc and when you wanted to restart a service you had to play games to ensure that it was really killed, and that the service state was cleaned up, and so on. Just telling openrc to stop the service just wasn't reliable at all - it worked well when nothing was wrong, but if nothing was wrong chances are monit wouldn't be doing anything.
Systemd is very effective at containing processes and their children and when you stop them, they are all gone for good. If you want to restart a service, systemctl restart service will get the job done 100% of the time, assuming the configuration/etc lets it restart. It does support graceful shutdown of individual services, followed by process genocide.
This also applies to things like cron jobs you launch through it. When the parent process ends, anything left gets cleaned up.
Systemd was forced down my throat by Arch Linux. I didn't know anything about the controversy back then, so I just thought: "There's probably a good reason for this, let's get to work".
I read some docs and I liked the security features a lot! You can tighten services easily with a declarative syntax.
Here's a snippet from my ntpdate.service file. You don't need much systemd knowledge to guess at what each line does:
PrivateTmp=true
ReadOnlyDirectories=/
InaccessibleDirectories=/boot
InaccessibleDirectories=/root
InaccessibleDirectories=/etc/ssh
LimitNPROC=1
DeviceAllow=/dev/null rw
DeviceAllow=/dev/urandom r
User=nobody
Group=nobody
CapabilityBoundingSet=CAP_SYS_TIME
NoNewPrivileges=true
I ended up enjoying that work and tightened things so much that I hit a bug, which was resolved in just a few days: https://bugs.freedesktop.org/s...
But I still don't know how to configure the network properly T_T
Reliable servers don't just crash.
Thanks for reminding me .. I always forget to add this to my source code:
#define RELIABLE
I am Slashdot. Are you Slashdot as well?
Just installing systemd over sysv speed booting up from a minute to 10 seconds, and shutting down from half a minute to under one second. I always hated Linux couldn't handle shutting the fuck down efficiently, now it does.
Or even better : The nice thing is systemD is not portable. It will infect only linux.
aaaaaaa
A small thing I've come to appreciate with systemd is that it actually cares about exit codes. This applies to any unit, including timer units (the equivalent of cron jobs). I ported most of my cron scripts over to systemd and suddenly started noticing scripts which had been having non-zero exits for ages, but fcron just didn't care about exit codes.
You can tell systemd to ignore exit codes for a process, or specific exit codes. However, I've found that in general using systemd I have a lot more awareness of abnormalities in my services.
Sure, you can often get away with ignoring exit codes, just as you can often get away with ignoring compiler warnings. However, in getting rid of them I fixed a few problems ranging from trivial to important, and my system is more robust for it.
Writing service files for my own daemons or modifying existing ones is pretty close to trivial. The files are short, easy to understand, and there isn't any risk of runaway child processes like there is with a sysvinit init script making them close to trivial to write and maintain. If anything I would say that's why so many distros are jumping on board.
I had to write service files as an early adopter, but it would also be useful for anyone rolling out their own daemons or that needed to tweak the behaviour of an existing service for their own needs. I imagine it would also lead to fewer packaging bugs.
In which case a nanny process restart is useless. Thanks for making my point, idiot.
Many hardware failures are transient, and a process restart is a very effective fix, at least in the short term. In the longer term, you'd better have monitoring in place so you know that the restarts are happening and can decide when to fix the hardware.
In addition, many process crashes are caused not by hardware failures, but by obscure software defects, and a process restart is not only effective at getting the production system back online, but arguably is a complete solution to the problem if the defect is sufficiently obscure that it's very rarely triggered, and hence not worth the large amount of effort required to identify and fix it.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
Systemd is modular:
- It's broken up into multiple independent processes, each of which handles one major thing well.
- It's broken up into libraries so that commonly used code (such as parsing config files) is implemented once and shared among the services, saving memory (because you know how shared object libraries work, right?) and ensuring that there's only one implementation of any one thing that needs to be tested and debugged.
- Interdependence among services is minimized, although as with any real, complex system, there are chains of dependencies.
- Dependencies on and among services are handled on-demand so that only the services you need are running (often started well after boot). As a side effect, boot time is shortened.
- Process 1 (init) is very small, with minimal functionality, in order to minimize the chances that it will crash.
The above are all true, or at least they are consistent with claims made by the developers. Sure, I have negative things to say, which are also true, but I don't want to add to the noise of all the false negative claims floating around.
I like it because it hasn't been proven to cause Tourette syndrome. The swearing it seems to cause is just a coincidence.
Systemd during a normal boot is doing some detailed logging of the boot process that let's you do a simper version of bootchart. This allows you to find out how long each service took to boot and also do a graph of them.
It's concerned complimentary to the actual bootchart...
[1] http://0pointer.de/blog/projec...
Back in 1993 I worked on a network process controller for Inmarsat-A. It had it's own custom init process group leader which erected unix domain IPCs between certain children, some shared memory segments, and performed a waitpid to relaunch any child process that died. I was new to the project and new to unix so when I suggested we use inittab to take over this task, the team lead just laughed. Init was not up to the task. The arrival of systemd finally addresses this use case: a process group leader that can care for of a hierarchy of tightly coupled processes. Systemd may be complicated, but so is the intended use case.
The reality is that before systemd showed up, there wasn't really any project with an active upstream that tried to solve the plumbing problem (I'm not talking about init in isolation here). Each distro had to invent their own hacks, some of them good, some of them not.
The fact is that the community that is beginning to form around systemd is much more healthy than the scattered bits and not-quite-fitting pieces we had before. Maybe that's sad, I don't know. I think that in the end, the unification around systemd will allow competitors to form (just implement the interesting subset of the systemd interface and you can integrate with all distros!). So long term we'll end up with a much more vibrant plumbing for Linux.
In which case a nanny process restart is useless. Thanks for making my point, idiot.
Your logic is flawed. Random bit flips sometimes happen. They don't necessarily take down the whole machine.
Poettering get to pretend that he's Linus. That, and it manages to start all of the services correctly on my system without hanging about 9 times out of 10. Aside from that? as previous posters have said, if a service stops/crashes for some reason? I want it to stay stopped until i can figure out what is wrong. MySQLD? out of disk space? I don't want that to keep on yo-yoing up and down until i have it fixed. I agree, SystemD should be optional, you want it on your phone/IOT device? fine. My server that I reboot about once every 2-3 months? I want init.
Background: I've professionally administered Unix and Linux machines for >25 years, including various BSDs, Linuxes, Irix, HP-UX, Solaris/SunOS, AIX, etc. I've been certified by several vendors or distributions, including, since 1999, Red Hat (which gives me quite a bit of background on their specific implementations over the years). I don't work for a company doing development of any OS or platform. heck, other than random 401K type aggregate ownership, I don't own stock in any company that cares about this issue at a deep level, to the best of my knowledge :)
Personal Bias about this thread: You pretty much lose all the credibility possible with me when you start of with "I ... dislike systemd because ... it looks ... like a poorly-described, gigantic mess I know nothing about ... ." (It's the "disliking something you know nothing about" that bugs me). Otherwise, I don't care much about this debate on a personal level. I currently admin boxes using systemd as well as everything else, and nothing about systemd has caused me anywhere near the heartache that it seems people who haven't used it much seem to feel about it.
Seriously, there're thousands of pages of documentation about what it is, how it works, and what most if not all of the design basis decisions are/were. I'll link you to a few of them because hey, you can get to slashdot and post, but you can't seem to use Google ;) (tongue in cheeck, of course). There're plenty of folks who DO have great detailed reasons on why they don't like bits and pieces of it, and you should be able to compare them to the various info I'm linking below.
Systemd has tons of upside and tons of downside. Most are pretty well detailed, although many of the gut reactions people seem to have to it are based on a lack of understanding about how it works and what it's compatible with, to wit "I can't use shell scripts for anything at startup anymore!" , "All of my old chkconfig or SysV scripts can't be included at all!", "It kills off syslog!", "The only reason it exists is to make laptops boot faster and in the server world we don't care", etc. Those are easily researched and the actual basis (or lack thereof) pretty easily found.
So, for why the systemd setup looks like it does, you can go back 4.5 years to where the announcement and rationale is described. Speed is part of it, as is device changability, as is double-forking, resource limits, and service state checking and recovery. Yes, it's a load of stuff. Definitely a system-wide approach VS a semi-random collection of various ways to do things all tacked together (which is, frankly, what most Unix and Unixlike systems are, through survival of the fittest).
http://0pointer.de/blog/projec...
http://0pointer.de/blog/projec...
http://0pointer.de/blog/projec...
http://0pointer.de/blog/projec...
http://0pointer.de/blog/projec...
Since RedHat's obviously the largest major proponent and arguably the source of the most production users, here's their documentation:
https://access.redhat.com/docu...
Here's the project page with loads of links about the software and uses cases:
http://www.freedesktop.org/wik...
And of course so many questions have been raised the developers have posted their rebuttals to myths or misunderstandings.
http:/
OK I'll try this. Traditionally Init services were about starting processes. They didn't have capacities for keeping processes running. Which meant that for processes that need to be kept running and for which there was a real possibility of failure (most of them) the init had to start a process management system which then started the functional process. This has been the status for years. With systemd there is a process maintenance component standard in the init system. Which means that processes not only start but are capable of being kept in a stable state easily and automatically. Process management stops being something system admins work hard at and instead becomes something that happens out of the box.
Sometimes, availability really is critical. In that case I want to take the risk of an automatic restart before the cause is investigated. However, it is important to appreciate that the approach is risky . The restart can cause cascading errors that change a reasonably simple issue into a multi day recovery operation.
Your history is a bit off here. Linux's earliest intent was as a workstation OS. An operating system that Unix users could administrate and on their x86 box. A $2k alternative to buying a $7k Solaris workstation. RedHat itself came from this desktop mold, before the LAMP stack made the real action cheap server alternatives. Mandrake, Caldera Desktop, Corel Desktop were all doing easy desktop before Ubuntu. What was unique about Ubuntu was: it was free and gave away the CDs. Everyone else was trying to make money selling the operating system.
What you are right about is that Ubuntu created a generation of Linux users with no experience with other Unixes or often other Linuxes. In so far as Linux was ever successful on the desktop Ubuntu was the form of that success. Your paranoia about "disruption" and "the establishment"... is simply that. Exactly what do you think in the 1980-90s you were doing to the mini computer culture of the generation before you when you made client server cheap and ubiquitous?
For example, RedHat in 1994-6 was selling a commercial X-Server as an option because XFree86 was too hard and too crappy. That wouldn't have been the case if no one cared about ease of use or just doing bug reports.
____
As for the "no plan" that was part of Linux culture from the beginning. It was one of the things that distinguished the GNU community from the 386BSD guys (and the children that spun off to be the today's BSDs).
Caveat: I am a server admin.
With systemd, one can't even remotely log a journal natively, which is par for the course in many server environments. You can't make this stuff up, please see bug #1098132 (https://bugzilla.redhat.com/show_bug.cgi?id=1098132) At the time I'm writing this, that functionality still just *doesn't exist* in systemd/journalctl. It was almost pushed into the last revision, but the bugs were show-stoppers so was pulled. Even if it does go into the next systemd revision, how long until it's really kosher for solid/production environments? 1 year? Two? Three? Yet it's being pushed as the default *now*.
To explain why this is important: if someone cracks in, the log files are going to be one of the first things they muck with. You have locked down syslog/et al, so you can tell if they muck with the binaries, but the log files themselves are considered compromised. Logging remotely at the same time before the data even hits the disk creates another layer of safety, which you simply can't do with systemd/journalctl. One day you will, assuming you want the binary journal and journalctl, but even if you wanted those now you couldn't. Yes, you could rsync over a copy of the files, but there's a reason why people aren't just doing that for regular logs and this is going long.
Over the last decades, people have worked out tried-and-true systems for documenting and verifying logging. So from an audit perspective, with journalctl you can't lock down syslog-ng and that process, because you've introduced an untrustworthy one above it. They know the current strengths and limitations -- to the point that there are legal/liability issues involved for many if their logging goes wonky. Journalctl adds a layer between the systems they have that work (and even have protocols in place for in terms of auditing), and as of right now adds a *wonky* layer that's very buggy. And by buggy I mean prone to corruption and simply not doing as its told.
Here's the real kicker: most of this issue would go away if systemd was willing to allow the user to not use journalctl and let things like syslog-ng have that data in the raw. It is choosing not to allow that as a tactic, as opposed to what the users and tech itself wants, and so here we are.
Also, this entire proposition by samzenpus is inane. When one thinks backwards from what the motivations might be, none of them are good and make me lose that much more respect for the site.
I came here to post something similar. We've seen a load of switchers over in FreeBSD land as a result of systemd, including one company that's currently a moderately large RedHat customer and plans to migrate before their current support contract expires.
I am TheRaven on Soylent News
i'm a small time admin, running about 100+ wireless nodes for artisans, private and commercial users. the -majority- of users running debian on desktops (autopilot upgrades rolled out), the rest are smartphones etc. windows is -actively- banned from the network. all routers, servers, firewalls are running debian, handrolled with lenny, wheezy and jessie, but all configs by the book, and audited by an experienced 2nd set of eyes. i'm 35 years in computing, telecoms and electronics. systemd is the worst single time-waster event in my professional career. although not yet enabled by default in jessie, it's causing a -lot- of disruption with basics like remote rebooting becoming an economical hazard (petrol costs), remote desktops not cranking up, auto logins not working (previously reliable packages like gdm3 break on upgrade due to forced dependencies to systemd related packages), endless hangs on startup/shutdown, many users complain about slow shutdowns, or machines not shutting down at all, services not starting. and yes: i understand how the beast works, and i can see the beast is doomed. the logging is a ---fucking--- nightmare, works fine on paper, but is making it impossible to get basic stuff done when in the real world things break, and taking the unreadable logs with it into the ether. i'm yet to see a machine that actually booted faster, or indeed shut down faster. agressive parallelization seems to apply ony for a subset of computing equipment i'm unfamiliar with. i can already hear the smart arse comments from corporate armchair admins: 'do this, do that ...' but the reality is that the problems caused by the yet flaky systemd is creating such an extra workload that is buries any hope of getting to grips with it. driving to each customer and wipe the systems fucked by systemd and re-install them with wheezy or carefully pruned jessie is cheaper, and we can all have our lives back. i never thought i'll be wading through backported software again ...
for regular user systemd brings no obvious advantages, at least not for their ancient harddrives on single core desktops - but instead a few ended up with trashed installs, e.g. by old non-critical errors being dragged into an upgrade and then breaking on (remote) reboot.
if i hadn't left the core machinery on lenny and wheezy i'd be in deep shit now, and likely going out of business.
stefan zalewski, burren.org
To date, there is no evidence that systemd causes ebola.
Wow. The usual complaint is the myth that systemd is monolithic. It's not. But now you're complaining that it's broken up into too many services? And how is this any different from sysvinit, which also starts any number of different binaries?
All they've done in systemd is write C code to start up services that used to be started instead by shell scripts and added the ability to make dependency resolution automatic so that services are only started when they need to be. So basically, they made it smarter and faster. The complaint that it's got too many binaries is moronic and a complaint just as well against sysvinit.
Systemd is modularized into a number of different binaries, each of which handles a different service. Thus, different functions are isolated from each other. This enhances stability and improves startup performance when not all need to be running first thing at boot time.
Oh, and all of systemd's config files are written in ASCII, in the traditional UNIX way. One cool thing they've done is made a single parser implementation (in a shared object library) so that all of the config files have the same syntax. Also, when you debug a problem with parsing for one service, you automatically debug it for all others at the same time.
Not to mention crashes caused by rare, hard-to-reproduce race conditions.
Indeed. That is one sub-category of the obscure software defects I mentioned. It's probably the best example, actually.
It's interesting to describe the approach used by many systems at Google (where I work; I'm now in Android but used to work on Google servers): a common pattern at Google is to crash immediately upon detection of any error. This is actually just a logical extrapolation from Google's long-used approach of building reliable systems on top of cheap, unreliable, commodity hardware, applying the same notion to individual software components. System designs assume that anything can fail at any time, and are built to recover gracefully, possibly with some degradation. So there is extensive infrastructure to fail a request over to to another process instance, and to automatically restart any failed processes. And of course there is extensive and detailed monitoring, with lots of statistical analysis of failure modes plus charting of everything to enable patterns to be recognized and various forms of alerting, ranging from automated bug filings, to e-mails, to pages delivered to on-call engineers.
Given that approach to fault tolerance, it's often very reasonable, at least for non-Java processes, to simply abort/crash whenever anything goes wrong. Restarts are automated and fast (for non-Java processes; JVMs are a bit slower to start) and monitoring and alerting take care of letting people know what happened and how often. This includes both hardware and software-related failures. Monitoring also pays special attention to processes that fail repeatedly (called "flapping") upon restart and generates high-priority alerts. The restart infrastructure will also slow and even stop the restarts of flapping processes.
Anyone who's used the googletest unit test framework for C++ may have wondered about the extensive support for and documentation of so-called "death tests", which allow you to verify that your code crashes when it's supposed to, and in the right way. This is a consequence of this particular approach to fault tolerance; if crashing is part of your reliability plan, you need to test that your code crashes when and how it's supposed to.
None of this has anything to do with systemd, of course. The fact that a strategy is effective in Google's environment is utterly meaningless in single-server contexts. In this case, though, auto-restart plus monitoring and flapping control seems like something that could usefully work in many contexts, perhaps even as the default.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
systemd's logging system is a huge improvement over old legacy style text logs. It has more early boot logs since it can log while still in initramsfs, and with kdbus it will probably get even earlier and late logging; the goal is "metal-to-metal" logging.
Having structured and indexed logfile (called journal) is a huge improvement in many ways; it allows for rich meta-data like monotonic timestamps, high precision time stamps, UUID's, exe file names and paths, and incredible easy and powerful sorting and filtering with tools like "systemctl".
If you try to add meta-data like monotonic timestamps in flat text files, they become very long and cluttered, making them hard to read for humans.
Another problem with flat text files are the fact that watch scripts depends on the exact wording of the daemon output strings; if the developers changes the wording, third party scripts will fail. In a perverted sense, the exact wording have become a API that cannot easily be changed or extended.
Not so with systemd; it has a stable API and lots of language bindings. It is easy to make a watch script that targets the stable field's.
Some CLI examples: I have used full length options for readability and since systemd have excellent bash completion for everything, it doesn't involve much more typing.
journalctl --boot -1 --priority err
Show all log entries from previous boot only, that have log level "error".
journalctl --since -5m
Show log entries generated the last 5 minutes.
journalctl --follow --unit smartd.service
Follow (tail) the smartd daemons log output.
journalctl _KERNEL_SUBSYSTEM=usb --priority warning
Show only messages generated by the kernel USB subsystem that have log level "warning"
journalctl --field _PID
The "--field" option makes systemctl show all values of the following journal field. In this case it will show a list of all PID's that have ever written to the journal. Want to see every executable that have ever generated log output, substitute _PID with _EXE. Notice the underscore. Field values with underscores have kernel guarantee for being correct.
It is also easy to track two services that you suspect are causing each other problems:
journalctl --output short-monotonic --boot -u NetworkManager.service -u chronyd.service
This will show the log entries for NetworkManager and a sNTP client from the current boot, and show the output with monotonic timestamps. Nice.
It happens about 40 times a day on you average PC
No, it doesn't. I operate lots of machines with ECC RAM. I've seen ECC correction. Bits flip, but nowhere near that frequently. It's very rare.
For those who think SysVinit style init systems is what Linux should be using the next 30 years, there is Slackware. It is a nice general purpose distro that is very traditional. So nobody is forced to use systemd if they don't want to.
Until some key functionality used by people is no longer available in that distro due to decisions made upstream to no longer support the code base, or other dependencies.
If I use KDE - which I do - then packages for that become unavailable at some point in Slackware given the above. That means I will be forced to use systemd if I want to continue using KDE; which also means I will have to change distributions, assuming Slackware remains systemd free, as well.
Not trivial. Not easy. Not freedom of choice.
That is simply a lie.
Lodragan Draoidh
The more you explain it, the more I don't understand it. - Mark Twain