<clever>
[725409.099332] systemd-journal(1162): READ block 399988800 on nvme0n1p1 (72 sectors)
<clever>
and if you enable block_dump, linux will log (to dmesg) every single IO action on the system
<clever>
just remember to turn it off when your done, and that the logs about that will cause more logs, infinite recursion!
<angerman>
Yea.
<clever>
the other big thing in a nix machine, would be /tmp for the actual builds
<angerman>
Time to reboot the node. I couldn’t nixops it anymore because /boot was too small at 128mb
<clever>
but you may run out of ram for some builds, if you try to tmpfs it all
<clever>
there is also an option for that...
<clever>
boot.loader.grub.configurationLimit
<clever>
angerman: most of the bootloaders have an option like this, which limits how many generation get put into the /boot/
<angerman>
clever: I don’t really build much on the pis they are deployment targets, not build machines.
<clever>
but how much disk it uses, depends on how many unique kernels you have, over the last $configurationLimit generations
<clever>
ah
<clever>
i'm (ab)using my rpi4 as a build machine currently
<clever>
cpu temp gets up to ~63c, hot enough that i can only touch it for a few seconds
<angerman>
Idk. Can rent on AWS a machine for 4-7ct/h
<clever>
i'm still stuck in the penny-pinching mindset, and avoid cloud resources when i can, lol
<angerman>
I used to use packet.net, but they’ve failed to bring up the aarch64 machines multiple times now. And I had 30-40min spin up times.
<angerman>
Apart from that they are either in Amsterdam or Tokyo, which limits me to ~2mbit up/down to those machines.
<clever>
when i was working on nixops packet support, i tried to spin up an arm machine to test that, but they where just out of arm's
<angerman>
AWS A1 instances are in SG, and I can get at least 10+Mbit. Not sure why I can’t get more. The line would give 1Gbit.
<angerman>
And the AWS instance is ready within seconds.
<angerman>
I *do* wonder however if using amazon Linux + install nix on it would be smarter as those are builder by the second.
<angerman>
clever: re pennies, I used to be that also a lot, but eventually I started using rented cloud resources more and more and found the experience for short bursts of need quite good.
<angerman>
Also now someone else has to care about all the hardware failures, ... :-)
<clever>
i had a not-os image setup to netboot on my pi's
<clever>
but i had found the netbooting unrealiable
<clever>
and one day, i opened a ticket and we dug thru packet traces, and i believe i found the cause, but the foundation has just stopped responding
<clever>
basically, the tftp client treats any network timeout as file-not-found
<angerman>
lol
<clever>
so if there is a random network outage, and it doesnt get a response for start.elf, it claims start.elf not found
<clever>
and if it doesnt get a response for kernel7.img (the proper one for that model), it tries kernel.img (for a different model), and then fails later on
<angerman>
Nasty
<clever>
depending on when the network drops out, it can cause vastly different side effects
<clever>
and the tftp server is my router, which has a bug in its NIC, causing random dropouts
<clever>
angerman: this configures nixos to support an iscsi disk for the rootfs
<clever>
angerman: i was using that to work around SD wear issues on my pi's, a few years back
<clever>
i had initially tried nfs, but that ran into even bigger nasty surprises
<clever>
first, xfs spreads the inode table evenly over the entire disk, and it picks an inode slot near the 1st data block, to reduce seek times
alp has joined #nixos-aarch64
<clever>
if the xfs volume is over 2tb in size, then your inode#'s get past 4 billion, and stop fitting in 32bit ints
<clever>
nfs properly passes those 64bit inode's on to the client
<clever>
if a 32bit userland, without large-file-support (same as that mmap problem) then tries to readdir(), it gets EOVERFLOW because the inode wont fit
<clever>
angerman: and then for the final surprise, a wide range of basic linux utils, treat EOVERFLOW as an empty directory!!
<clever>
so building the stdenv from scratch, over nfs, on a 32bit builder, with a 64bit nfs-server, results in basic tools claiming file not found, when it clearly exists
<clever>
after that huge headache, i gave up on nfs for the pi, and switched to iscsi
alp has quit [Ping timeout: 272 seconds]
* clever
heads off to bed
<angerman>
You do like those low level battles, don’t you?
ryantrinkle has quit [Ping timeout: 256 seconds]
alp has joined #nixos-aarch64
alp has quit [Ping timeout: 272 seconds]
<ar>
anyone had any luck with getting NICs to work on mainline kernel on lx2160/honeycomb?
<ar>
ok, got my usb ssd to show up when connecting it to the other usb port
orivej_ has joined #nixos-aarch64
orivej has quit [Ping timeout: 265 seconds]
greizgh has quit [Quit: greizgh]
greizgh has joined #nixos-aarch64
greizgh has quit [Client Quit]
greizgh has joined #nixos-aarch64
greizgh has quit [Quit: greizgh]
greizgh has joined #nixos-aarch64
greizgh has quit [Client Quit]
greizgh has joined #nixos-aarch64
orivej_ has quit [Ping timeout: 260 seconds]
orivej has joined #nixos-aarch64
t184256 has left #nixos-aarch64 [#nixos-aarch64]
t184256 has joined #nixos-aarch64
<samueldr>
dan_b`: I don't remember if you sait touch was working on the mobile nixos boot gui, but my gut feeling tells me it probably wasn't, if you want to test a fix https://github.com/NixOS/mobile-nixos/pull/154
<craige_>
A bit of a lazy pre-research question but is anyone aware of any aarch64 off-the-shelf or even kit-form wi-fi routers that I can get NixOS onto? clever?
<samueldr>
I think flokli had something along the lines
<angerman>
craige_: depending on what you are looking for, you could just use any scb, and plug a dumb AP onto it? (e.g. those ubiquiti dishes?)
<hexa->
flokli owns a turris omnia iirc
<samueldr>
sounds right
<samueldr>
armv7l though I think
<samueldr>
I wasn't sure about that
<flokli>
samueldr: turris mox too
<flokli>
which would be aarch64
<flokli>
but no NixOS (yet) on any of those
<craige_>
I'm trying to replace what I suspect is a dying TP-Link WDR3400 running OpenWrt angerman - so I'm haping to replace th ewi-fi and and routing feature in one nuit but i'm happy to break it out into multiple units - I've been looking at some multi-ethernet SBC's for example.
<hexa->
WDR4300*
<craige_>
That too :-)
<hexa->
tbh, the nixos hostapd module isn't very nice, yet
<hexa->
I aim to fix that some day
<hexa->
if you aim to get some minipcie or m.2 radios into an sbc, note that you'd need two radios to cover both bands
<craige_>
:+1:
<craige_>
Thanks hexa++ - still not sure what path I'm heading down yet.
<{^_^}>
hexa's karma got increased to 1
<craige_>
I'm still unsure where recent network issues have arisen from either. I can't actually pin it on my near decade old WDR4300, I just suspect it. angerman++ 's suggested path my be the most efficient path.