pxc has joined #nixos-aarch64
<samueldr> gchristensen: how did the idea of using the disk as swap space to allow bigger tmpfs look in the end?
<samueldr> good, bad or terrible?
<gchristensen> seems like a fine idea, but I haven't done it
<gchristensen> it'd be "easier" than making it "install" to disk
<samueldr> yeah, no need to clean the disk or anything
<gchristensen> right
<samueldr> the box might be getting a bit full again, tried collecting, there's a bunch of builds artifacts keeping roots alive though :/
<gchristensen> that is another thing we could do, find roots and delete them
<samueldr> I'm 100% confident we just had "disk full" builds fail on ofborg
<gchristensen> ouch, time for a reboot / fix for sure. I'll take a look at swapping -- I forgot to finish the deploy for clever's account, too :/
<gchristensen> clever: sorry for the delay on that
<samueldr> gchristensen: wait a sec before rebooting :)
<gchristensen> ok
<samueldr> :/ might have overstimated the "a sec"
<gchristensen> no worries
<gchristensen> there is a lightning fast fix available to us
<gchristensen> ok added 5g of swap, working on 50g
<samueldr> I wonder how the tmpfs has to be triggered to see it
<gchristensen> hmm I suspect I accidentally made my second swapfile `bs` instruction 500gb instead of 50gb
<samueldr> bs=10240 count=52428800
<samueldr> 10K * 52428800
<samueldr> so yeah
<gchristensen> oh well, I'll cancel it somewhere along
<samueldr> it will handle it by itself AFAIK
<samueldr> dd can't really do much once it filled the target device
<gchristensen> of course :) but I don't want to take up all the space
<gchristensen> [root@aarch64:/home/grahamc]# free -g
<gchristensen> total used free shared buff/cache available
<gchristensen> Mem: 125 12 0 67 112 44
<gchristensen> Swap: 197 4 192
<samueldr> root tmpfs didn't increase (understandably)
<gchristensen> :)
<samueldr> apparently mount -o remount,size=
<gchristensen> at least memory pressure is reduced
<gchristensen> want to do it, samueldr?
<samueldr> yeah
<gchristensen> go for it
<samueldr> it works
<samueldr> sudo mount -o remount,size=100G tmpfs /
<gchristensen> nice
<gchristensen> samueldr: let me know when you're done: I'll disable swap and add a partition to the disk
<samueldr> gchristensen: do it
<gchristensen> you're done?
<gchristensen> I can wait
<samueldr> done with the copy of the thing yeah
<samueldr> otherwise you can wait for "????" for the sd-image build which may take a good while :)
<samueldr> where "????" is a unit of time
<samueldr> but I would suggest instead doing it
<gchristensen> hmmm not sure I can swapoff :)
<samueldr> oh! I thought you were restarting the machine, deploying
<gchristensen> there we go
<gchristensen> I need to add the partition before I can reboot
<samueldr> the whole block device wouldn't have been enough? mkswap it at boot, to be sure?
<samueldr> maybe I'm misunderstanding something, but that means that there's a stateful bit in the partition existing?
<gchristensen> ah
<gchristensen> yes, you are missing something :)
<gchristensen> /dev/sda has one partition which is mounted to /persist, which contains the SSH host key, and ofborg private configs
<gchristensen> but that partition is taking the entire disk, so I can't easily alter it
<samueldr> I see
<samueldr> gchristensen: I thought the machine would be rebooted, will it? (I'm waiting to restart the thing)
<gchristensen> well ...
<gchristensen> hmmm
<samueldr> hmm?
<gchristensen> ok, yeah, I'll just kick it over for now and do the partition stuff next reboot. swapon reall made it hard to do :)
<samueldr> yeah, you'll have a clean slate if you reboot now :)
<samueldr> (and you did AFAICT)
<gchristensen> yeah, just did
<gchristensen> though it is trying to unmount the swap before continuing the reboot :
<gchristensen> ok issued a harder reboot, since linux wasn't going to give up easily
<gchristensen> ok samueldr should be up
<samueldr> :)
orivej has quit [Ping timeout: 250 seconds]
iqubic has joined #nixos-aarch64
iqubic has left #nixos-aarch64 [#nixos-aarch64]
pxc has quit [Ping timeout: 252 seconds]
pxc has joined #nixos-aarch64
orivej has joined #nixos-aarch64
<v0|d> anybody had luck w/ serial uart1 on pi3?
pxc has quit [Ping timeout: 252 seconds]
orivej has quit [Ping timeout: 246 seconds]
orivej has joined #nixos-aarch64
orivej has quit [Ping timeout: 246 seconds]
orivej has joined #nixos-aarch64
<gchristensen> this is probably the coolest thing you'll see today: http://visual6502.org/sim/varm/armgl.html
<samueldr> does it run nixos? /s
<gchristensen> its simulating a mass rebuild, now!
pxc has joined #nixos-aarch64
<sphalerite> whoooooa
<samrose> if I am seeing little boot partitions appear on my emmc device like mmcblk2boot0 179:64 0 4M 1 disk I am thinking those are generations of nixos rebuilds?
<samueldr> nope
<samueldr> those are generally unused, but a type of partitions for emmc devices, which could be used by the platform to hold e.g. the bootloader
<samrose> ah
<samrose> they are disks not partitions so that makes sense
<samrose> (my bad I called them partitions)
<samueldr> they're partitions at a higher level, but unrelated to the common disk partitions :/
<samrose> ah ok thank you
<samrose> samueldr: on these aarch64 boards does the nixos-install approach still work?
<samueldr> "on these" is a trick question :D
<samueldr> yes and no
<samueldr> using the sd-image*.img it's easier to just nixos-rebuild after configuring via /etc/nixos/configuration.nix
<samueldr> but yes, you can also do it using nixos-install to another device
<samueldr> as long as you're careful enough to allow it to be bootable according to your platform
<samueldr> (e.g. the unused disk space for the embedded bootloaders)
pxc has quit [Ping timeout: 260 seconds]
<samrose> so, easiest to just copy image from sd card, to emmc, then to edit /etc/nixos/configuration.nix on emmc target, then nixos-rebuild there
<samueldr> you could even boot and install using the UEFI boot iso, which I did on my rpi3
<samueldr> as long as the board supports booting UEFI
<samueldr> (rpi does NOT support booting UEFI stock, but there are UEFI implementations usable on it)
<samueldr> samrose: for your last sentence, yeah, it's the classical arm sbc way, well tested
<samueldr> though there's one major caveat
<samueldr> /boot is annoying to use as it can fill up quickly
<samrose> so you probably disable that boot before you nixos-rebuild eh?
<samueldr> last times I think I forgot and did it the first moment I was forced to :)
<samrose> if you disable boot, do you typically need to copy over components to the remaining parition to achieve booting?
<samueldr> uh, no, at rebuild stuff will work (IIRC)
<samrose> ok, that makes sense
<samrose> also, if I actually build may own sd image, I can just leave out /boot
<samrose> (as I have done in my x86 ISO)
<samueldr> probably, as long as u-boot can read from the partition flagged as bootable
<samrose> cool I'll try that down the road
<samueldr> eventually, I'd look at defaulting to still having a FAT32 partition, but used exclusively for the raspberry pi files
pxc has joined #nixos-aarch64
Thra11 has joined #nixos-aarch64
Thra11 has quit [Client Quit]
pxc has quit [Quit: WeeChat 2.3]
<samrose> is there any reasonable way to build OS image and channels with hydra for aarch64? Should the hydra instance itself run on ARM64
<gchristensen> do you have hydra?
<samrose> gchristensen: I have a hydra server running on x86 hardware (on aws)
<gchristensen> ah
<samrose> it works great for building channels and ISO for that variant of the OS
<gchristensen> so the hydra server doesn't need to be ARM but it does need an ARM hardware
<gchristensen> as a build machine
<samrose> ah, so I could do an agent
<samrose> hydra agent
<samrose> that is arm
<gchristensen> right
<gchristensen> Nix and Hydra call them build machines :)
<samrose> heh, yes build machines
<samrose> maybe I could get that going on packet.net?
<gchristensen> you could, yeah, there is some c1.large.arm capacity in NRT1 and SJC1
<samrose> or I could get a local arm dev machine with more resources and wire it up to be a build machine for hydra
<gchristensen> not sure you'll be able to (easily) get a local arm dev machine with more resources than c1.large.arm, but yeah
<samrose> because I am not sure if I can even use the full capacity of that c1.large.arm
<samueldr> it's uh, got all the beef
<samrose> I suppose I could use their api, spin up, do my thing, and power off
<gchristensen> sure
<gchristensen> you'd have to fully deprovision it, not just power off
<gchristensen> you can also use their spot pricing to get it for cheap
<samrose> might just do it. Seems less ridiculous than trying to maintain my own local hardware
<samrose> (for me)
<gchristensen> I also don't like to admin fiddly, expensive, and weird hw
<gchristensen> :P
<samrose> if it wasn't running all the time, the costs would be negligible
<samrose> (if packet.net wasn't running all the time)
<samrose> "Imagine that you could take advantage of 96 cores for just $.50/hr? What would you do?"
<samrose> lol
<sphalerite> samrose: unless a raspi is enough for your purposes, in which case maintaining your own local hardware is fine ;)
<gchristensen> the ~$70 of raspi equip buys a substantial amount of build time at $0.50/hr :)
<samrose> sphalerite: yessir, that's what I've been doing to date (although with bananapi as the target instead raspi because my company already went all in on this hardware sadly. But at least I can actually now run the damn OS on it thanks to you all)
<gchristensen> tbh I'm a fanboi so
<gchristensen> take everything I say with sufficient salt
<samrose> yeah, it might be reasonable to just set up raspi machines locally too. it seems an option worth considering anyway.
<samueldr> but your fancy packet server can't do hdmi and audio things locally :3
<samrose> true there is no way to test HDMI (we don't need audio, but do at least need HDMI)
<samrose> we actually will require to have the exact bpim64 hardware set up as a test hardware phase prior to rollout
<gchristensen> what're you making?
<samueldr> sounds fun :) must be nice having an actual goal other than "make the damn thing work"
<samrose> gchristensen: we have machines that will act as nodes for a distributed hosting system, where hosting is limited to specific kinds of apps https://holo.host/
<samrose> there are intel x86 boxes, and these bananapi m64 boxes
<samrose> we are running an overlay over Nixos on both
<samrose> the machines will auto-update, auto-garbage collect
<samrose> autoupgrade based on channels from our hydra server, and we are merging nixos channels upstream
<gchristensen> neat
<sphalerite> samrose: oh neat, didn't you work with mayflower also?
<samrose> sphalerite: I came into it not really knowing much of anything about NixOS months ago, but I started crash coursing myself on it because I was convinced it would be a much better pathway than the alternatives for an autoupdating linux distribution. I made a lot of progress, and mayflower helped me udnerstand how to do the overlay, and did some of the work etc
<samrose> they were a big help
<sphalerite> samrose: I recently joined mayflower :D
<samrose> ah cool
<samrose> seems like lots of people in EU are into NixOS, and people over here in the states haven't really caught on yet
<gchristensen> some day
<gchristensen> I hope to have a nixcon: graham's backyard
<sphalerite> not so sure about the not really caught on yet bit, but yeah it does seem to be quite Europe-centric
<samrose> well yes, anyone who understands it, and becomes aware of it seems to become really into it
<sphalerite> I mean, one of the major US retailers is using it :D
<samrose> wow, I did not know that
<sphalerite> What does Target actually use it for? Is that publicly available information?
<gchristensen> uhhh
<gchristensen> building ... software...? :)
<gchristensen> data science and optimisation team uses it
<gchristensen> I hope Intel is scared and pulling out some stops
<gchristensen> ARM CPUS is getting fast, and quickly: 32 cores at 3.3GHz(turbo) in a single socket? 1TB RAM? crazy pants
<gchristensen> only 125W too
<sphalerite> gchristensen: well the frequency isn't really very telling. A current 3.0GHz CPU from Intel and a 3.0GHz Pentium with the same amount of cores will still perform very differently. And that's with the same basic instruction set!
<gchristensen> true enough
<gchristensen> however
<gchristensen> I think the direction here should cause Intel serious concern
<gchristensen> we're watching innovator's dilemma unfold in real time
<sphalerite> innovator's dilemma?
<gchristensen> basically a massive, successful industry leader is killed by a cheaper, smaller company going after the low end
<gchristensen> and that eventually the quality of the product of the low end becomes high enough to take the leader's business, meanwhile the leader already lost the low end
<sphalerite> hm interesting
<gchristensen> I think the book was worth reading
<sphalerite> Christensen… Coincidence?? I think not!
<gchristensen> you caught me!
orivej has quit [Ping timeout: 244 seconds]
Dezgeg has quit [Ping timeout: 264 seconds]
orivej has joined #nixos-aarch64