keventd not working?

[aanderse]

the other night i tried out the keventd functionality for the first time - but it doesn't appear to be working properly for me

when i boot up i don't see any pwr/ac condition, or anything like that
i was poking at the logic in keventd a bit and it is creating the proper condition via symlink... but then the symlink is immediately deleted

i ran inotifywait and then plugged my laptop in, this is what i saw:

aaron@framework ~> inotifywait -m /var/run/finit/cond/sys/pwr
Setting up watches.
Watches established.
/var/run/finit/cond/sys/pwr/ CREATE ac
/var/run/finit/cond/sys/pwr/ DELETE ac

if you have any ideas what might be happening here i would appreciate it!

thank you 🙇‍♂️

[troglobit]

Sorry for the late reply.

I have honestly not looked at keventd for quite a while, so this malfunction might be caused bit rot. It was initially created to tie into the condition subsystem to work with a UPS I have at home, but that system was never fully deployed.

I've had some plans and ideas though, like expand on it a bit to connect it for more udev-like tasks. If you have a minute over to have a look at this that would be very appreciated. (I'd probably ask Claude code myself for a first review anyway ...)

[aanderse]

I have honestly not looked at keventd for quite a while, so this malfunction might be caused bit rot.

understood, looking at some git history this makes sense 👍

expand on it a bit to connect it for more udev-like tasks.

that sounds really cool!

---

i was mostly just curious about this but you really caught my interest with your udev comment so i will poke around with this a bit more when i find some spare time and report back

thanks!

[aanderse]

i dug through the code and found out that keventd is working, just not how i expected

my expectation was that even if no services had listed sys/pwr/ac as a condition that the condition would still show up in initctl cond dump - this is not the case... and in hindsight it seems somewhat obvious

once i actually added sys/pwr/ac as a condition to a service then reloaded everything worked fine 😅

i suppose this line was the source of my confusion

sorry for all the noise!

---

I've had some plans and ideas though, like expand on it a bit to connect it for more udev-like tasks.

any interest in sharing those ideas? i have been battling with mdevd recently so mostly i'm just curious 😄

[troglobit]

i dug through the code and found out that keventd is working, just not how i expected

Glad to hear! It's still very rudimentary, so I wouldn't have been surprised if there had been something else.

my expectation was that even if no services had listed sys/pwr/ac as a condition that the condition would still show up in initctl cond dump - this is not the case... and in hindsight it seems somewhat obvious
once i actually added sys/pwr/ac as a condition to a service then reloaded everything worked fine 😅

Ah good point, that's a detail we probably should mention in the docs.

i suppose this line was the source of my confusion

Heh, yeah that code is a bit sneaky. It was actually @wkz who wrote that, almost ten years ago now. To this day I still have to remind him to add comments :laugh:

sorry for all the noise!

Not at all!

---

I've had some plans and ideas though, like expand on it a bit to connect it for more udev-like tasks.
any interest in sharing those ideas? i have been battling with mdevd recently so mostly i'm just curious 😄

Absolutely! To set the scene, I think [system/10-hotplug.conf(.in)](https://github.com/troglobit/finit/blob/master/system/10-hotplug.conf.in] paints a pretty good picture of the sad state of things.

I'd like to see a tighter integration of the capabilities of what (e)udev/mdev/mdevd into Finit, without "doing a systemd" and gobbling one of them up, but I've seriously considered adopting mdevd[^1] and integrating its capabilities into keventd. Keeping the logic but with a syntax closer to udev rules, to ease adoption. This would allow us to provide 10-hotplug.conf as an example of how-to-do-it without keventd, and make startup, auto-loading modules, etc. a lot easier.

[aanderse]

very interesting to hear

i have read through the source code of mdevd and i appreciate how straight forward it is so i suppose integrating that into finit shouldn't be a terrible task?

eudev is not very responsive these days so maybe there is some reasonable justification to bring this in house?

thanks for sharing your thoughts!

[troglobit]

i have read through the source code of mdevd and i appreciate how straight forward it is so i suppose integrating that into finit shouldn't be a terrible task?

Exactly, that's my conclusion also.

eudev is not very responsive these days so maybe there is some reasonable justification to bring this in house?

Can't be the easiest job chasing the tail of systemd-udevd all the time, and the likeness in the name brings some pressure as well.

I'll have a chat with my coworkers, but I'm leaning in favor of this, it could be a worthy problem to tackle and would likely add to the value offering by Finit.

thanks for sharing your thoughts!

any time

[aanderse]

i recently found out that mdevd-coldplug was "accidentally" working for me - i was using its preferred (and very fast) execline functionality to symlink /dev/disk/by-uuid paths into place and then switched to simple shell scripts which is not as fast... and my /boot partition wasn't mounted - seems like i was just getting "lucky" in the past (because execline is fast) and my /boot partition was being mounted

i have switched my /etc/fstab to mount boot by its /dev/nvme... direct path for now... but if finit integrated a device manager i can imagine there would be better opportunities here. sure there could be other simple solutions involving modifications to bootmisc plugin or whatever... but just thought i would mention here as an example

[troglobit]

I've started playing around with expanding the scope of keventd on the devman branch, if you're interested. It's still very early (untested) and quite rough around the edges. Next step is to add a rules engine and I'm leaning towards mdev/mdevd style rules.

[aanderse]

I've started playing around with expanding the scope of keventd on the devman branch, if you're interested. It's still very early (untested) and quite rough around the edges.

oh wow! this is amazing! the most immediate question i have - could you please support netlink rebroadcast? this will allow people to build software with libudev-zero and run graphical applications with keventd!

Next step is to add a rules engine and I'm leaning towards mdev/mdevd style rules.

looking forward to it! please keep me in the loop 🤩

[troglobit]

could you please support netlink rebroadcast? this will allow people to build software with libudev-zero and run graphical applications with keventd!

Sure thing! I've just pushed another set (568b22d) of completely untested code 😅 which should do the trick.

Next step is to add a rules engine and I'm leaning towards mdev/mdevd style rules.
looking forward to it! please keep me in the loop 🤩

Will do!

[aanderse]

i gave the code a read through now and this is great! it is so understandable!

question about startup - is the idea to use some readiness notification after the coldplug code runs (or doesn't run, if not specified)? this would match the suggestion finit documentation has on other device managers, right?

[troglobit]

i gave the code a read through now and this is great! it is so understandable!

Glad to hear! 😃

question about startup - is the idea to use some readiness notification after the coldplug code runs (or doesn't run, if not specified)? this would match the suggestion finit documentation has on other device managers, right?

Something like that, yeah. Not really gotten to the integration point yet 😊