Keeping SSH sessions alive with systemd-inhibit | Hacker News Reader

Keeping SSH sessions alive with systemd-inhibit

kd8bny.com

65 pointskd8bny19 comments

Yes, this is a bodge. And yes, it should be unnecessary to jam in an inhibitor lock taken around the lifetime of an interactive shell permanently in the background.

The author wants xyr system to not suspend either when there's someone active on the GUI or when xe is logged in over the network. Theoretically, systemd already has a whole mechanism for making this happen.

The pam_systemd.so hook into PAM allows systemd-logind to track SSH sessions in its replacement for the Unix login database — all of that stuff under /run/systemd/users/, /run/systemd/seats/, and /run/systemd/sessions/. And in theory systemd-logind could respond negatively to a suspend request if there is an active SSH login session even if the GUI is idle.

In practice, it does not quite have the logic for achieving this. It's close, but it does not pass around enough information for this to be done as, say, a PolicyKit rule.

* https://github.com/systemd/systemd/blob/main/src/login/login...

This has the same purpose as caffeinate command on Mac and will prevent the system from suspend while a command is running. This is useful with tmux or long compilations or tests or whatever.

On Linux I rather prefer to configure my laptop to never suspend when connected to a power supply. For me the compilations and tests heavily drain battery and should not be run typically on battery.

But it turned on Mac Apple really wants to suspend and configuring the same behavior is somewhat non-trivial. So in past when I used Mac i often had a terminal window running

caffeinate -i sleep infinity

to prevent any suspend until I hit Ctrl-C there.

The underlying API is called Power Management Assertions.

I think this could be made into systemd (user) service fairly easily. Then you wouldn't need to worry about `disown` or tracking the PID, you'd just `systemctl --user start|stop prevent-sleep.service`

If you want to be extra fancy, you could even write small program that calls the dbus api directly and then just waits to be killed. Avoids the turducken of waiting processes.

Something unclear about the explanation: systemd-inhibit is not running `sh & disown`, it's running `sh` and `& disown` applies to the systemd-inhibit command. If you actually daemonized the inner command, systemd-inhibit would release the lock immediately.

Cool trick though! I didn't know about systemd-inhibit.

Of course systemd killing your SSH session is a intentional feature, not a bug.

Consider disabling suspend from the logind.conf and suspend, if at all, only at explicit user request.

I think the author’s point was that this is usually desired behaviour because it’s a desktop and only occasionally do they want to stop that.

systemd doesn't kill your SSH session (if you'd WoL your PC there's a good chance you can continue typing unless your local terminal detects that the other end goes down).

It's more that when you configure systemd to suspend your computer when there is no physical activity, it will suspend your computer when there is no physical activity.

The author wants their computer to suspend automatically.

I think modifying your system configuration to never suspend is a much worse solution than using the tool designed to prevent suspending the computer while a specific program is running to prevent suspending the computer while that specific program is running.

It'd be easier if `sshd` would permit you to wrap the incoming command line/shells so `sshd` would spawn your session with systemd-inhibit, but I don't think that's possible?

Unwarranted suspends by logind have always been a menace, i have been fighting them for a decade.

> It'd be easier if `sshd` would permit you to wrap the incoming command line/shells so `sshd` would spawn your session with systemd-inhibit, but I don't think that's possible?

sshd already links against systemd (by distro patches, not per upstream) but since SSH is a "legacy protocol" and its users are "uneducated troglodytes", this is not going to happen.

I think this is because OpenSSH is an OpenBSD project, and both systemd and OpenBSD refuse to support each other.

> SSH is a "legacy protocol" and its users are "uneducated troglodytes",

Who are you quoting here?

I assume that's a parody on systemd developers' attitude towards The Unix Way (TM).

This github issue comes to mind:

https://github.com/tmux/tmux/issues/428

What is the problem being solved? That your device is going to sleep or that it is interrupting SSH sessions? If it's the latter turn off keep-alives and an idle SSH session will live through sleeps and network interruptions.

Why not fix the keep alive time? Yes, there's no wrong way to do it, but IMHO the nix philosophy states you should at least try to stick your problem and your solution together.

https://www.golinuxcloud.com/keep-alive-ssh-sessions-in-linu...

https://en.wikipedia.org/wiki/Keepalive

Sounds complicated. I just use autossh from the CLI and it reconnects if my laptop (or the remote machine) wakes up again.

i think i used a laptop for a few hours in my life before I decided to simply disable all autosleep features.

computer is on, it's on. lid close, screen is off. done.

That's fine. In fact you're right that laptop and desktop power management is generally best done manually by expert users. That how I have things set up too, more often than not.

The use case for wakelocks (a longtime Android feature from which this is conceptually derived) is phones, though. Send a quick snap, throw it in your pocket, and expect (1) you get the notification for the reply when it arrives and (2) the device lasts until it gets back to the charger at bedtime.

That's highly non-trivial and absolutely not amenable to manual power management. Is systemd the right answer? Maybe not, but that's clearly where the feature is aimed.

That’s a very specific use case and solved with a combination of hardware and software (solved badly with s2idle from Microsoft). With computers, you don’t expect notifications from sleep state. It’s either active (even idling) or on standby (you don’t expect it to wait). There’s no standard for active threads in deep sleep mode (restricted API, short TTL) that phone OS boast)

Right, but the software side of the stack needs a way for subsystems to say "I'm important, don't suspend", and that's what this systemd feature is for. The fact that it's based on top of an S3 suspend to RAM means that the laptop can't do it well, not that it isn't a useful feature.

(Also I have to quibble with the "very specific use case" idea: it's the standard use case. Laptops are the weird and rare edge cases these days.)

But on a laptop, I do want this case to be manual. Even on a phone, I would like to have more control over stuff like Notification and the likes (I think ios have background process or something).

Reply on news.ycombinator.com