Comment by owyn

I feel the same way. The big part for me is that it tells us that owyn doesn't use tab completion if they're forgetting about the ".service" part. Sure, I don't remember either, I don't have to.

I'll add the abstraction for anyone confused

  program [command [subcommand]] [flags] [object]
  e.g.s
  systemctl status sshd.service
  systemctl enable --now sshd.service
  touch -c test.sh
  echo 'Hello World'
  echo "Hello ${USER}"
  

Anything in brackets is optional and might not appear or be available. By command I mean a category of commands. Such as 'pip install' vs 'pip uninstall', which are sub-programs inside the main program. But this can have layers such as 'uv pip install'. Often flags can be used in any order because you'll just loop over all the arguments but this is still the standard order.

There's also the two actor pattern

  program [command [subcommand]] [flags] source destination
  e.g.s
  cp /foo/bar/baz.txt "${HOME%/}/"
  scp -i "${HOME}/.ssh/foo" ${HOME}/to_upload.sh user@remote:~/
  dd if=/dev/urandom of=/dev/diskToDestroy
  rsync /mnt/hdd1/ /mnt/hdd2/

When it comes to starting and stopping services I want the verb to go last. Way easier to press up, backspace backspace backspace o p to change service ssh start to service ssh top. This is a frequent pattern I follow as I start/stop/restart/reload. Having to go back at least one word adds keystrokes that aren’t necessary.

I think the previous post mentions the `service` command, which has `service foo start` pattern.

I'm used to `service` command, so I have muscle memory which `systemd` breaks. The way `systemd` commands are laid out is better, it just messes with `service` command muscle memory.

I on the other hand DO get this complaint. It’s less about the “word order” and more remembering the words to type. As a software engineer I have to google stuff all the time. I haven’t touched C#’s PLINQ in five years, I doubt I could do it from memory. Likewise, I can count on one hand the number of times I will set up a systemd service in a given year.

Many (most?) of us are all over the place, expected to “wear many hats” and aren’t in a single IDE or language all day every day. Certainly if I were a “Linux admin” I would have systemd pretty committed to memory. Anyone else probably wouldn’t.

For me:

* /etc/init.d/my-service stop

and Ubuntu’s:

* service my-service stop

both lurk in my brain.

Plus, you could always write a quick shell script to swap the arguments if it detects this specific failure. Inability to remember or learn which commands expect which arguments is, at some point, not the responsibility of the software to fix. Unfortunately, this field does necessitate some amount of memorization and ability to problem-solve.

If I had to guess it's because of the 'service' command, which goes 'service foo start'.

It took me ages to unlearn that pattern for using systemctl, even though as you say: it's far more consistent

It’s because service was the other way around, I’m sure.

sudo service my-service start

sudo service my-service stop

it's the same mentality that brought us the git design - the easiest, least typing options are rarely, if ever the thing you want to do.

Instead, these invocations give cryptic messages, throw errors, or sometimes, even break things.

The most common and helpful things are hidden deep behind multiple flags and command line arguments in manuals that read like dictionaries more than guides.

I'm always at a complete loss as to how such decisions are made. For instance, "git branch -vv" is the useful output you would like to see, every time, that should be "git branch". Why not make the current output, "git branch -qq"? Is a humane interface too much to ask for? Apparently...

I know people defend this stuff, but as a senior engineer in programming pits for 30 years, they're wrong. Needless mistakes and confusions are the norm. We can do better.

We need to stop conflating elitism with fucked up design.

In case you don't know, you can use the `-n` argument to `systemctl status` to tweak the log output, e.g. `-n0` to disable the log output and `-n40` to get more than the default 10 lines.

In my experience, most of the times a service fails, it fails on startup (misconfiguration, missing resources). So the status is OK. Also, you can always request more lines, if you think that's helpful.

This wouldn't be my problem with systemd. Not by a long shot.

> My biggest annoyance is "systemctl status" gives you just enough of the service's log to make the output take up most of the terminal each time you run it but never enough of the service's log to get a useful picture of what's actually happened with the service lately.

How about

systemctl status foo | tail

It's also worth noting that this is a fairly standard pattern in /etc

  /
  | etc
  | | fail2ban
  | | | action.d
  | | | fail2band.d
  | | | filter.d
  | | | jail.d
  | | ssh
  | | | ssh_config
  | | | sshd_config
  | | systemd
  | | | network   # network wide context
  | | | nspawn    # containers
  | | | system    # system wide context
  | | | user      # user wide context
  

Personally I like it more than

  /
  | etc
  | | cron.d
  | | cron.daily
  | | cron.hourly
  | | cron.monthly
  | | cron.weekly
  | | ...
  | | firewall
  | | firewalld

Keeps things less cluttered. Hierarchical categorization is >> than lateral

It could have been named 'global' instead of 'system'.

I tried, but found it very confusing around the user session. Apparently an environment variable need to point to a started dbus. I can't get consistent results over local session vs ssh vs su vs sudo

I used to do this but frankly it's easier to run a system-level unit as whatever user you want and keep all the files in /etc instead of scattered around /home.

The user-level units are most useful when running an actual multi-user system. If you trust your users to not abuse them, anyway.

When I'm really working with a service and get tired of typing sudo systemctl restart or status all the time, I'll just do a quick alias or two right then and there.

alias s=sudo systemctl status alias r=sudo systemctl restart

And if I'm only working with one service, I'll throw the service name in there too.

> cron just does what it says on the tin with no problems.

I can name a rather large problem with cron that systemd timers solve handily: long-running job duplication. When jobs take longer to run than the space between their triggering times, duplicates start piling up. I've had to rescue numerous systems from such states, which are difficult to detect until things have gotten quite bad. Sure, you can write a bunch of boilerplate to handle this yourself with cron, but with systemd timers it's all handled for you along with other niceties like capturing all output in journald and ensuring that the next run starts at soon as possible.

Launchd is not used in systemd - why would anything be wrapped in it? The config is simple - User for the user and Group for the group. You can print it the result of "id" if you're not sure what the result is.

> ...cron just does what it says on the tin with no problems.

Yeah, tell me about it.

We had a production-down support ticket filed for one of the things that I work on at $DAYJOB. The customer's VM's disk was full. Why? Because the 'timer unit' that was supposed to run logrotate every day had never been run and was never scheduled to run. (The VM had been up for a month at this point.) No other customers had ever reported this issue, and we'd not changed anything about that timer unit or what it is supposed to run in ages.

No amount of gyration or agitation with 'systemctl' and friends OR rebooting of the VM kicked the cron replacement into proper functioning. 'logrotate' was simply never being scheduled to run. This fucker was WEDGED, and the tools weren't helping us understand why.

We read through all the docs on all the various kinds of units and don't see what we're doing wrong. We do a BUNCH of digging, and find an -IIRC- open Github issue from years back where someone was running into this problem. More or less the last word on the issue was Poettering saying something like "Well... actually, now that we've said that that particular cascade of options isn't actually supposed to work, now I'm not so sure that they're NOT supposed to work.". And that was -apparently- that.

IMO, when your cron replacement can be easily configured in such a way as to never even try to run the thing that you scheduled it to run, you really need to go back and rethink how you've built your cron replacement.

Yeah, and you can omit the ".service" suffix, and you could use ".timer" suffix, too, if you have a timer for that service.

As much as I do not like systemd, I do not think its commands are an issue, and I use "systemctl status" and "journalctl" as well with some flags at times.