<bqv>
ah yes, of course this becomes problematic as soon as i consider that i'd have to revamp my secrets management, in darcs
<bqv>
(hohoho, close one, i nearly thought i'd uploaded all my secrets unencrypted to the darcs clone, but of course it imported via git so doesn't read them plaintext)
<aleph->
Hmm. What's the idiomatic way to start a service that depends on another service? In this case, redis
<aleph->
Wonder if there's an example I can look at
* aleph-
goes to look through github
<ajs124>
just use systemd unit ordering? with like, requires, after, before, that kind of stuff?
<aleph->
Ugh, doy. Right
<bqv>
one thing i never managed to properly do with native systemd is the "restart service B after service A starts/restarts"
<ashkitten>
bqv: if you find out i also don't know this
<bqv>
at this point i think it's just not doable, and the easiest way is to have a systemctl restart as an ExecStartPost
<bqv>
ashkitten: oh so i'm not the only one, wow
<ajs124>
just reboot the machine
<bqv>
lol
<ashkitten>
✨ ajs124
<{^_^}>
ajs124's karma got increased to 4
<ajs124>
I think require might be able to do this. AFAIR systemd has passable documentation for all of this. Not terribly accessible, but their manpages are actually decent.
<ashkitten>
my feet hurt so bad, i've been doing vr for a while today
<bqv>
ajs124: i trawled the manpages for ages trying to work out how
<ajs124>
I've been running nginx units which don't properly reload or restart the service under the right conditions for at least months
<gchristensen>
bqv: I think you want a bind
<ajs124>
BindsTo sounds like it, yes. documented in systemd.unit(5)
<gchristensen>
and PartOf
<bqv>
does that suffer from that if service A stops, service B stops?
<bqv>
because that's unacceptable for my specification
<bqv>
and that seemed to be the big issue systemd is unable to solve
<bqv>
__monty__: wait, this won't work, flakes don't support all these new esoteric dvcses
<bqv>
that's why i was considering mercurial, nix at least knows how to read git and hg, iirc
Arahael has quit [Ping timeout: 258 seconds]
Arahael has joined #nixos-chat
neeasade has quit [Remote host closed the connection]
MichaelRaskin has quit [Read error: Connection reset by peer]
slack1256 has quit [Remote host closed the connection]
waleee-cl has quit [Quit: Connection closed for inactivity]
drakonis has quit [Quit: WeeChat 2.8]
<bqv>
Hm. I'm running git-bug, to import all github issue data from nixpkgs to my git clone. I started on the hour, its now :20, and only 1000 issues are done
<bqv>
Theres nearly 90k total now, thats 1/90 done in 20 minutes
<bqv>
So the whole thing might take around 2000 minutes, or just under 12 hours
<bqv>
Yikes…
cole-h has quit [Quit: Goodbye]
<bqv>
import error: API rate limit exceeded imported 3405 issues and 1359 identities with default bridge
<bqv>
Marvellous
<bqv>
The ratelimit resets an hour from now, so I guess its "x per hour", so if its an hour on, hour off, it'll take around a full 24h to import at best
<ashkitten>
i can't even get xrgears to build, so i can't test if the xwayland patch is working
<ashkitten>
i know steamvr doesn't work right
<ashkitten>
openhmd in nixpkgs is missing a pkgconfig file
<ashkitten>
it sure seems as though steamvr isn't even trying to get a drm lease
<JJJollyjim>
Omg there's an xrgears?
<JJJollyjim>
I wanna experience the gears in full VR glory
<gchristensen>
infinisil: I need to keep a bunchof stuff in there :)
<samueldr>
/tmp considered harmful
<infinisil>
Ah, and I guess tmpfiles is more granular than clearing everything
<energizer>
doesn't seem necessary to delete stuff during rebuild/switch
<energizer>
in /tmp
<gchristensen>
I agree
<manveru>
gchristensen: that ipxe stuff is really driving me insane... apparently it only works on a handful of instance types...
<gchristensen>
ouch
<infinisil>
Maybe it should be a systemd service that runs on a timer
<manveru>
and i have no idea what they have in common
<manveru>
i just tried every instance type in eu-central-1 (above 4GB ram and below 1USD/h)
<manveru>
these work: `m4.large t2.large m3.large c4.xlarge d2.xlarge r3.large c3.large`
clever has joined #nixos-chat
clever has joined #nixos-chat
clever has quit [Changing host]
<manveru>
i tried for the longest time with t3a or t3 and it never would do anything... not being able to connect to the network
<manveru>
so i suspect that ipxe maybe doesn't have the right network drivers for some instances?
<manveru>
no clue how to debug that though :P
<gchristensen>
it is probably because ipxe is a barely capable implementation of a system and aws skips all the junky stuff and plops you in to a good kernel right away
<gchristensen>
and so ipxe doesn't know what to do with Nitro
<manveru>
that does indeed match the ones that don't work...
<gchristensen>
this is the same reason ipxe can take 1 hour to download a 1GB file trough a 10g NIC
<gchristensen>
it barely knows how to use the hardware
<manveru>
i was probably lucky that i tried t2 for my first attempts... or i would've given up ages ago :P
<gchristensen>
a more reliable thing could be convince clever to build not-os to fetch an image from your image store and use clever's kexec stuff to hop in to t
<clever>
not-os would be faster to dl, compared to the thinest nixos netboot image
<clever>
and then you have proper linux
<clever>
there is also haskell-init
<clever>
not-os is about 40mb of squashfs, haskell-init is about 2mb of elf and thats it
<manveru>
at this point i'm really not sure anymore what to choose...
<gchristensen>
could even teach nixops to do this :)
<clever>
gchristensen: thats in my nixops ticket, having nixops generate a tiny image (netboot or ami) and then copy-closure into /mnt via that
<gchristensen>
yeah
<gchristensen>
though I'd prefer it use user-data to tell theimage what to boot, and have nixops not be involved in getting the proper system there
<manveru>
jup
<manveru>
i think we just need to fix the crappy user-data handling we have atm :)
<gchristensen>
oh not even userdata
<gchristensen>
erm, that
<gchristensen>
I'd like nixops to upload some artifact, and then specify that artifact in metadata, and a shim boot up and fetch the closure based on that metadata and enter into that exact closure
<gchristensen>
not some sort of NixOS configuration expression embedded in the userdata
<clever>
gchristensen: my problem with that, is that i want nixops to pre-build things (because a lot will be shared within a cluster), and i dont want the 1st generation to be mis-aligned (different nixpkgs)
<gchristensen>
yes, we're talking about the same thing
<gchristensen>
except yours requires nixops in the loop, mine would allow for autoscaling groups
<clever>
you could do it via a binary cache
<gchristensen>
yeah
<gchristensen>
exactly
<manveru>
as long as you don't need private repos involved...
<clever>
you could just upload the entire closure (secrets and all) to a cache, then just give the user-data a way to access it
<clever>
manveru: it could be an s3://bucket based cache, that uses IAM instance data for privs
<manveru>
well, as long as it can run a vault agent, i can get what i need
<clever>
gchristensen: the thin image could have 2 modes, this mode would format the disk, download a closure from cache->/mnt/ and fixup the bootloader
<clever>
gchristensen: but my original plan, would ssh in, and copy a closure over ssh
<gchristensen>
AH
<gchristensen>
ah*
<manveru>
just need a way to run arbitrary code from user-data and anything is possible...
<gchristensen>
I don't want hacks :(
<manveru>
i call it flexibility :)
<manveru>
hack is what we have now
<manveru>
i basically need to build a specific revision of nixFlakes and vault, retrieve a deploy key and maybe some certs, and nixos-rebuild a flake...
<manveru>
really not sure how you'd generalize that
<manveru>
so my previous solution was to make a configuration.nix with a systemd oneshot service that does just that from the user-data...
<ashkitten>
aughh
<ashkitten>
why does my usb host controller keep dying
<cole-h>
I want to get an Ergodox soon-ish... Is there anything I need to be aware of?
<cole-h>
energizer: Why, exactly? Assuming money is no object.
<energizer>
cole-h: because money is an object :)
c74d has joined #nixos-chat
<Ashy>
i think the main thing to be aware of is you might not like it
<Ashy>
i hated mine and sold it
<Ashy>
back to a good old das keyboard
<cole-h>
What exactly did you dislike?
<energizer>
what features do you want in your keyboard cole-h?
<cole-h>
Preferably QMK. Preferably ergonomic, because I'm planning on using a keyboard for the rest of my life. Preferably mechanical, because I'll probably be gaming on it as well.
<Ashy>
couldn't get used to the thumb clusters (even though i love the idea of them)
<infinisil>
Ashy: How long did you try?
<Ashy>
also i swap between laptops and the desktop setup too often
<infinisil>
I have an ergodox ez, I do use the thumb clusters, but only the main 2 keys, the others not at all
<Ashy>
a few weeks
<Ashy>
it did teach me to use my thumbs for alt and cmd though
<Ashy>
and capslock is ctrl+esc as is correct
<infinisil>
I think one can get used to pretty much anything with enough time :)
<Ashy>
true, heh
* energizer
wonders if keyboards will be around in 40 years
<infinisil>
As an example, I'm now used to programmer dvorak with the odd number layout
<infinisil>
From left to right: 7531902468
<cole-h>
what
<cole-h>
What is that cursed number row
<cole-h>
Explain yourself
<Ashy>
nice, i'm a normal dvorak user but i tell anyone who asks to not bother so long as they learn to touch type qwerty
<Ashy>
looks like it's optimized for symbols rather than number ordering
<infinisil>
Hehe, the creators thought it would be a good idea to put the most often used numbers near the index finger column
<infinisil>
So 0, 1, 9, 2 are there
<infinisil>
And odd even because who knows :P
<gchristensen>
Ashy: I'm of the same opinion
<gchristensen>
I switched when my hands hurt too too bad and I tried a bunch of stuff all at once. I don't really think qwerty helps
<gchristensen>
dvorak*
<infinisil>
I think it's really hard to compare. I was using querty (actually quertz) before too, but I couldn't tell if switching to dvorak helped
<infinisil>
I feel like my fingers need to move less when writing english at least
<Ashy>
gchristensen: yeah, i swapped to dvorak when i was in college and had the time to spend a month getting back up to speed
<Ashy>
i wouldn't do it again even though it feels slightly faster and comfier
<bqv>
I used programmer dvorak once
<bqv>
It ruined vim for me
<bqv>
Thats why I had to go back
<bqv>
I also started to forget qwerty
<gchristensen>
oh cool the minimal plasma ISO is 95.9% reproducible
<bqv>
energizer: I hope so, unless something more accurate turns up. I wouldn't ditch a keyboard for more mice, but I would for something with more buttons
parsley936 has quit [Remote host closed the connection]
<energizer>
gchristensen: are those stats reported somewhere?
<bqv>
infinisil, Ashy: yeah, I'd kinda take issue with that. I made it work, for a while, but I was unhappy about it
<infinisil>
I have no problem with hjkl either, they might even be better with dvorak
<infinisil>
Well not better, but also very reasonably placed by coincidence
<Ashy>
yeah fair enough, i'm so used to it now that i really struggle with vim in qwerty so maybe part of the muscle memory is positional rather than mnemonic