gchristensen changed the topic of #nixos-chat to: NixOS but much less topical || https://logs.nix.samueldr.com/nixos-chat
mog has quit [Ping timeout: 246 seconds]
mog has joined #nixos-chat
lassulus_ has joined #nixos-chat
lassulus has quit [Ping timeout: 240 seconds]
lassulus_ is now known as lassulus
lopsided98 has quit [Ping timeout: 240 seconds]
lopsided98 has joined #nixos-chat
lopsided98 has quit [Read error: Connection reset by peer]
lopsided98 has joined #nixos-chat
<elvishjerricco> Obviously setting `sync=always` in ZFS is bound to be a lot slower. But shouldn't it at least be the same speed for synchronous writes?
<elvishjerricco> I'm trying to figure out why using a zvol for swap isn't working. It just locks up when you need swap.
<elvishjerricco> It seems like it's really just being unreasonably slow
<elvishjerricco> Testing with `dd conv=sync,fsync`, it seems any dataset or zvol with `sync=always` is like 10x slower than without.
<elvishjerricco> I'm also running `sync` immediately after just to be sure, and that's returning instantly
<elvishjerricco> Odd. It actually seems that the speed increases with the `bs`. With `bs=8K`, it's like 10MB/s max. With `bs=500K`, it's about 400MB/s
<ekleog> elvishjerricco: intuitively, sync=always makes a sync after each call to write(1)
<ekleog> which means that for dd, it makes a sync after each block of size bs
<ekleog> given a sync is awfully slow… well
<elvishjerricco> ekleog: Yea. But shouldn't synchronous IO do that too?
<elvishjerricco> hence the `conv=sync,fsync`
<ekleog> I'm not really sure what “synchronized IO” means in the context of dd
<ekleog> that seems unrelated to sync
{^_^} has quit [Read error: Connection reset by peer]
<ekleog> well, to sync(1)
<elvishjerricco> From `man dd` https://www.irccloud.com/pastebin/vUdJzVKH/
{^_^} has joined #nixos-chat
<ekleog> fsync adds a sync at the end of the complete file write
<elvishjerricco> Hm good point
<elvishjerricco> Still, I should be getting much better than 10MB/s
<ekleog> well, a sync is *slow*
<elvishjerricco> Only as slow as the disk's physical IO, right?
<ekleog> and 500K/8K*10MBps = 600MBps not far from 400MBps, so that's consistent with the idea that sync's take the most time in your test
<ekleog> it also runs the kernel block write scheduler or something like this, I think
<elvishjerricco> So how can I test this better?
<ekleog> hmm… I'm not familiar with it, but look into kernel perf tracing? ISTR there's something like kprof or similar designed to have an idea of where the kernel spends time
<ekleog> but just so I understand, what kind of disk is it?
<ekleog> if it's a spin disk, even if it's fast doing lots of small operations will be much slower than a big one
<ekleog> if it's a ssd, you should get optimal performance by setting the block size equal to the ssd block size
<elvishjerricco> ekleog: This is an nvme ssd
<elvishjerricco> so 10MB/s is very surprising to me
<elvishjerricco> ekleog: Also, why would doing many smaller syncs be slower than doing one big sync?
<ekleog> elvishjerricco: a ssd writes data by blocks
<ekleog> meaning that if you write less than a block, it will still rewrite a full block
<ekleog> so this will also wear your disk faster
<ekleog> as the kernel batches the writes into a single block if you do one big sync
<elvishjerricco> ekleog: Ah. So swap on SSD sounds kinda bad in general?
<elvishjerricco> Since blocks on SSDs seem pretty large
<elvishjerricco> The kernel would have to know to dump full blocks to swap at once.
<ekleog> https://superuser.com/questions/728858/how-to-determine-ssds-nand-erase-block-size seems to imply that the block size is likely around 4MB
<ekleog> well, I don't think swap on ssd is bad in itself (at least not more bad than with any other disk)
<elvishjerricco> So you think the kernel is smart enough to dump to swap in full blocks?
<ekleog> spin disks are worse, because they need to physically spin up to the place where it's being written every time
<ekleog> I think it's doing as good as it can :)
<ekleog> (that's the reason for the name “block device”, btw)
<elvishjerricco> It sounds like if I set the volblocksize of the zvol to the SSD's block size, I might see the problem go away
<elvishjerricco> The ZFS FAQ suggesting using `getconf PAGESIZE`.
<elvishjerricco> So maybe that's just very wrong for SSDs
<ekleog> well, I don't know at all about zfs, so can't say it, but so long as you force sync on each write, it'll be slow, because no matter what the fs does it'll have to write at least a block on each write
<ekleog> just curious, why do you want to set sync-on-each-write?
<ekleog> (also, please note that my analysis here is a bit simplified, depending on the type of flash memory there are actually other things that are done to improve the performance even when blocks smaller than full blocks are written, but that depends on the actual hardware)
<elvishjerricco> ekleog: Well if the zvol is going to be for swap, you really don't want ZFS to keep any writes in memory; you just want it to go straight to disk
<elvishjerricco> Because if you're writing to swap, you probably don't have any memory to cache the writes to
<ekleog> I'd be assuming that zfs would just buffer a full page before writing it to disk
<ekleog> but then, again I don't know anything about zfs :)
<ekleog> (or about the precise implementation of lvm, swapfs, etc., for that matter)
<ekleog> but typically linux doesn't start swapping when it's already too late, it always keeps a bit of margin for the stuff required to actually swap
<ekleog> vm.swappiness
<ekleog> oh, and vm.min_free_kbytes, too
kisik21 has joined #nixos-chat
<elvishjerricco> Hm. I'll play around with not having sync=always
<etu> Heck yeah! I've booked hotels and flights to the correct city: https://events.ccc.de/2018/09/11/35c3-call-for-participation-and-submission-guidelines/
<etu> :D
<etu> Living on the edge
<etu> I hope to see some nix-people there :)
Ericson2314 has quit [Ping timeout: 250 seconds]
kisik21 has quit [Ping timeout: 244 seconds]
__Sander__ has joined #nixos-chat
<sphalerite> Does anybody know of a server provider which offers ARMv7-capable servers that's not scaleway? I've been having a number of minor issues with scaleway, but unreliable storage is something I really c don't want to have to work around…
<etu> Oh, they seem to have armv8
<etu> Maybe that's close enough?
<sphalerite> etu: nope, unfortunately the packet machines don't run 32-bit ARM code
<sphalerite> at least the thunderx ones that they actually have available publicly
<ekleog> sphalerite: ISTR online.net has some arm servers
<sphalerite> ekleog: online.net is the parent of scaleway so I'm not sure that'll be any better, but thanks!
<sphalerite> ekleog: yeah it's just the scaleway offering they're showing :(
<ekleog> oh yeah good point
<ekleog> didn't remember the offer name :°
<ekleog> hmm… maybe on some cloud stuff you can find armv7 VMs?
<sphalerite> doesn't look like it
<sphalerite> I could run my own ARMv7 VMs, but I don't want machine emulation-level slowness >.>
<sphalerite> and the one ARMv7-capable server I've been able to use other than scaleway C1s is currently unavailable
<ekleog> hmm soyoustart can apparently have some armv7… but it's for storage servers :/ https://www.soyoustart.com/en/server-storage/
<ekleog> so you'll basically get a rpi
<ekleog> cortex a9, iirc that's exactly a rpi
<sphalerite> "stocks are currently being replenished"
__Sander__ has quit [Ping timeout: 240 seconds]
__Sander__ has joined #nixos-chat
<sphalerite> Welp. I've ordered another chromebook.
<rawtaz> oops
<rawtaz> what are you going to use it for?
dmc has quit [Quit: WeeChat 2.2]
<sphalerite> rawtaz: it's an aarch64 one so I won't have to build everything myself
<sphalerite> well I don't have to build absolutely everything myself with the current one (armv7) thanks to Dezgeg's cache, but a lot of stuff. Typically takes ~36h, when everything goes smoothly
<sphalerite> which it rarely does
<sphalerite> and I don't have a full-featured browser for it
<rawtaz> so its for release and test building NixOS?
<sphalerite> no, it's for actually using
dmc has joined #nixos-chat
<rawtaz> ok :)
<sphalerite> I like the chromebook I have because it's small, light, it wasn't very expensive, has a long battery life, and can run without any proprietary software
<rawtaz> yeah thats great
<rawtaz> so you throw NixOS on it
<rawtaz> what model did you get?
<rawtaz> i see there's multiple vendors as well
<sphalerite> the one I have is the ASUS C201
<sphalerite> the one I've ordered is the C101PA
<sphalerite> also from ASUS
<rawtaz> ah, the flip one?
<rawtaz> does touchscreen work with nixos?
<rawtaz> its nice that you can wrap it like that
<sphalerite> the touchscreen in my big laptop works just fine, although I rarely use it
<sphalerite> every once in a blue moon I draw something on it in the GIMP
<rawtaz> :>
<sphalerite> and I've set up a script to flash "DON'T TOUCH :(" whenever the screen is touched. Because friends don't touch friends' screens when pointing at them.
<rawtaz> yeah.. i dont think i have any friends, based on that :/
<sphalerite> lol
<rawtaz> quite a nice feature that :D
<rawtaz> perhaps you can make it show some horrible scary picture all over the screen when they touch it
<sphalerite> I considered adding a screaming sound-effect or something but that might be too obnoxious
<rawtaz> like that old labyrinth flash "game" that in the end show a horror picture
<rawtaz> yeah that too!
<sphalerite> yeah I just run feh with full-screen, just that the image I display is just httpts://sphalerite.org/dump/no-touchy.png
<sphalerite> it just displays it on the whole screen for one second, then disappears again
<rawtaz> haha its great
<sphalerite> ironically I wrote it using the touchscreen
<rawtaz> of course :)
<rawtaz> so do they ever reply "but it's a touchscreen!"?
<sphalerite> I only wrote it recently and it hasn't claimed any "victims" yet
<rawtaz> hehe ok
<rawtaz> anyway, really awesome idea :) looking forward to it being used
<rawtaz> this alone could be a reason to get a touch screen
<sphalerite> I know right?
<sphalerite> best use of my touchscreen: keeping my screen smudge-free
<rawtaz> :D
lopsided98 has quit [Ping timeout: 252 seconds]
lopsided98 has joined #nixos-chat
jtojnar has quit [Ping timeout: 252 seconds]
jtojnar has joined #nixos-chat
lopsided98 has quit [Ping timeout: 264 seconds]
lopsided98 has joined #nixos-chat
<ekleog> need to buy a laptop with touchscreen
<ekleog> definitely do
<sphalerite> ekleog: for the same purpose?
<ekleog> yup
<samueldr> when "lapping" the laptop, I like scrolling pages with touch
<ekleog> (and for drawing japanese when I don't know the reading of characters, that was my initial motivation, but it was not enough to make me actually buy a laptop with touchscreen)
<sphalerite> samueldr: I do that every now and then, but two-finger touchpad scrolling is almost always more convenient
nbp has joined #nixos-chat
* nbp never heard of this channel before
<samueldr> :)
<samueldr> qwerty-lafayette is interesting for the typographically exigents, but everything is covered with canadian french + compose key; dead keys are especially useful there
<nbp> samueldr: So I highly recommend the Querty-lafayette despite the fact that you have to execute xkbpcomp at the beginning of each session
<nbp> samueldr: having symbols on the left hand while pressing the Ralt key is really a live changing experience when coding.
<samueldr> I do believe though that the advantages compared to french canadian are much less than when compared to azerty :/
<samueldr> (but never actually used an azerty layout, so can't say for sure)
<sphalerite> I've used azerty. I hate it with a passion :p
<joepie91> azerty
<samueldr> french canadian looks like the baby bear of the keyboards, where QWERTY isn't good for languages with accents, AZERTY seems clunky for programming, and french canadian is just fine :)
* joepie91 shudders
<nbp> I used azerty for a while, and stopped when I learned to code.
<samueldr> (I'm obviously biased, having learned that layout first)
<nbp> and now that 7 years since I switched to qwerty-lafayette, and I would recommend the coding layer to everybody.
<rawtaz> nbp: it's because it doesnt actually exist
<nbp> and the compose key is nice for writting French without too much pain.
<nbp> rawtaz: because you hands stay at the same place.
<nbp> rawtaz: with the right thumb on the Alt key.
<rawtaz> :)
<nbp> You have HTML on the first line, C-like languages on the second line, and parentheses on the third.
<nbp> and I have not yet get used to the keypad yet, but I should…
<nbp> ^ on the third line … that's part of my modifications of it.
<nbp> to have “();” as a sequence of keys.
<nbp> ^ Both of them have the ({[]}) symbols at weird locations.
<nbp> even Qwerty
clever has joined #nixos-chat
<clever> sphalerite: www.oneclickroot.com this site just feels fishy, lol
<clever> ok, recovery mode, system status=official, frplock=on, warrantyvoid=0, secureboot=enabled!
<samueldr> clever: device?
<clever> samsung tablet, SM-T560NU
<samueldr> do you want to switch to lineageos or keep a samsung rom?
<rawtaz> where in the Nix manual is the "Nix daemon" explained?
<clever> samueldr: keep the current rom and all data
<rawtaz> sorry, wrong channel
<samueldr> oh, keeping data is always harder :/
<clever> yeah
<samueldr> most of the time unlocking the bootloader also wipes (which is good imho)
<clever> yeah
<clever> with the kindle fire hd7, there was an exploit in the backup restore process, that allowed you to write to files you shouldnt
<samueldr> though I don't know about samsung / odin based devices, all my recent experience comes from the different fastboot-based devices
<clever> and that allows you to trick the OS into thinking its on a VM
__Sander__ has quit [Quit: Konversation terminated!]
<clever> which then means `adb shell` just gives root
<clever> i have used odin to root a galaxy S3 without data loss
<samueldr> I'm peeved at how OEMs can sell you devices that you can't control :(
<clever> i can undestand wiping data when disabling secureboot
<samueldr> yes!
<clever> so you can control it, and you cant bypass the lock screen
<clever> also, it is now stuck in odin mode, i cant shut it off
<clever> and it tells you, do not turn off the target, lol
<clever> Bus 001 Device 008: ID 04e8:685d Samsung Electronics Co., Ltd GT-I9100 Phone [Galaxy S II] (Download mode)
<clever> from lsusb
<clever> windows registered it as com3, lol
<sphalerite> clever: maybe fastboot can tlak to it?
<clever> the guide says to use odin, and the screen does say its in odin mode
<sphalerite> but odin won't talk to it?
<clever> odin is downloading...
<sphalerite> oh right
<ldlework> allah is doing
<joepie91> lol
<joepie91> that's gotta be one of the most famous spammers by now
<dmc> Allah?
<samueldr> AFAIUI, samsung devices don'T do fastboot at all
<sphalerite> clever: huh, f-droid says "no versions with compatible signature" for usb mountr
<clever> :S
<dmc> @ joepie91
<samueldr> installs here
<joepie91> dmc: yeah
<sphalerite> clever: do you have any other ideas how I might be able to clone the BIOS config across all these machines from linux?
<sphalerite> clever: the BIOS has an option for "replicated config" which allows me to store the config to a USB stick but as mentioned I don't have one >_>
<clever> sphalerite: silly bios vendors, lol
<sphalerite> we need coreboot everywhere!
<maurer> If you're a Euro, and your MEP voted in favor of the new copyright directive, please send them a message telling them how hard they've fucked up
<clever> samueldr: had to go into developer options and toggle "oem unlocking"
<samueldr> for most "google-like" devices, this only toggles the bit, but doesn't unlock for unsigned images
<samueldr> the unlocking for unsigned image (with fastboot oem unlock generally) will erase the device
<clever> yeah, the "FRP lock" is now off, but secureboot is still on
<clever> the oem unlock also asked for the lock-screen code before allowing it to toggle
<clever> but odin still cant see it, i suspect the drivers are missing
<samueldr> ages ago, with a galaxy S 1, I used heimdall on linux for this
<samueldr> (but it was so different, no oem unlock in the menus, all bootloader controlled)
<sphalerite> Hmmmmmm, coreboot supported "HP Compaq 8200 Elite SFF PC" at some point. The machines I'd like to run it on are "HP Compaq 8000 Elite SFF PC"s…
<clever> sphalerite: step 1, when running flashboot as root, can you detect the flash chip, is it in writable mode?
<sphalerite> clever: you mean flashrom?
<clever> yeah
<clever> ive yet to find a machine that flashrom can write on, but i can read the bios on 2 of my machines, the router and nas
<sphalerite> … nope flashrom can't read it :/
<sphalerite> I tried flashrom -p internal -r rom
<clever> then you would either need to physicaly remove and reflash the chip, or fix flashrom first
<clever> sphalerite: i gave the fishy one click root a try, and after installing adb drivers, it claims it will work
<clever> but then demand you login or register on their site, lol
<sphalerite> oh dear
<clever> sphalerite: and if you do fork over a name&email, they demand $39 to root it, lol
<sphalerite> jeez
<clever> sphalerite: odin on windows fails to detect the device in odin mode
<clever> heimdall in nixos detects the device, but fails to flash any parition
<sphalerite> nice
<andi-> clever: are you running windows in a VM? I had to use odin on my OPO3T and figured (after a few days) that the stupid driver only works in usb2.0 mode and VirtualBox (that I used for clicky usb-passthrough) only supported 1.0/1.1 without the unfree parts..
<sphalerite> FWIW windows works in qemu-kvm too
<samueldr> OH, virtualmachine
<andi-> yes, but I didn't want to fiddle with the usb passthrough (never did that before) and wanted my phone back after a few days :)
<samueldr> I would stray away from using a VM as it *can* cause issues
<andi-> well I do not run windows anywhere.. Didn't want to install it either
<samueldr> from memory, the reason I had to use heimdall is that odin failed, and a quick search basically said that it could stop the process in an unexpected manner
<sphalerite> but yeah VMs don't work well for fastboot either
<andi-> in the end it worked flawless..
<samueldr> though, for fastboot, using virt-manager and usb passthrough it worked quite well for postmarketo
<samueldr> postmarketos*
<andi-> after I figured that USB stufff..
<samueldr> (I did have to add a couple different usb IDs so it'd be flawless)
<andi-> after a co-worker spent a day customizing debian VM images I just spent an hour with the nix tooling to create debian VMs... I'd say my approach is much nicer then using packer but I am biased :D <3 nix
<sphalerite> lol
ma27 has quit [Quit: WeeChat 2.2]
ma27 has joined #nixos-chat
lopsided98 has quit [Ping timeout: 250 seconds]
lopsided98 has joined #nixos-chat
<clever> andi-: odin in virtualbox lacked drivers for the odin mode
<clever> andi-: odin on real hardware was using the teensy microcontroller drivers
<clever> andi-: i did suspect it was vbox and 1.0 only support to blame, and switched to real hardware
Ericson2314 has joined #nixos-chat
<joepie91> today's DIY project: https://i.imgur.com/4YpHN7n.jpg
Ericson2314 has quit [Ping timeout: 240 seconds]
* andi- watches those hanging builds on release-18.03...
<andi-> It somehow bothers me more then usually..
<ldlework> sphalerite: do you think my idea about theme unification is sound?
<sphalerite> I have no idea what you're talking about
<ldlework> oh ok
<LnL> andi-: stuck builds?
<andi-> LnL: >13h "sending inputs"
<LnL> what machine
<andi-> ike.ewi.tudelft.nl.
<LnL> hmm, doesn't show up in hydra-bad-machines
<LnL> I could try to restart them if you give me a link, but I'm not sure if that works for this
<andi-> https://hydra.nixos.org/build/81212009 is the other one I saw
<andi-> (same machine)
<LnL> that second one seems to have worked
<LnL> but the other one doesn't stop
<LnL> heh, it's in the list now :p
Guanin has joined #nixos-chat
<andi-> LnL: did you abort them again? That firefox test was start fetching things, looked good and now it is back to queued :/
<LnL> it's still building dependencies, no?
<andi-> ahh, thats why the output went empty
* andi- still gets confused by hydra
MichaelRaskin has joined #nixos-chat
<LnL> I think 19 and 20 are the scripts and 21 is the actual test
<LnL> ah no, that's another script... :p
<andi-> looking at ike in general it seems like he/she/it started stalling jobs 23h ago :/
lopsided98 has quit [Ping timeout: 240 seconds]
lopsided98 has joined #nixos-chat
<clever> [clever@amd-nixos:~]$ adb shell id
<clever> uid=2000(shell) gid=2000(shell) groups=2000(shell),1004(input),1007(log),1011(adb),1015(sdcard_rw),1028(sdcard_r),3001(net_bt_admin),3002(net_bt),3003(inet),3006(net_bw_stats),3009(readproc) context=u:r:shell:s0
<clever> ok, so adb does work
<clever> and yep, fastboot doesnt detect it
<LnL> andi-: "fixed"
<andi-> LnL: thanks, what did you do? Shoot it? :-)
ma27 has quit [Quit: WeeChat 2.2]
<LnL> yeah so apparently you have to find the right job to abort
ma27 has joined #nixos-chat
<LnL> if 2 jobs depend on the same build that gets stuck you have to find the one that started it, not the one that's waiting
<LnL> hydra is weird :p
ma27 has quit [Client Quit]
ma27 has joined #nixos-chat
jtojnar has quit [Remote host closed the connection]
jtojnar has joined #nixos-chat
ma27 has quit [Quit: WeeChat 2.2]
ma27 has joined #nixos-chat
ma27 has quit [Client Quit]
ma27 has joined #nixos-chat
jtojnar has quit [Quit: jtojnar]
jtojnar has joined #nixos-chat
tertl3 has joined #nixos-chat
<gchristensen> "The Java Virtual Machine used to launch <oXygen/> XML Editor is not supported.JAVA_HOME=/nix/store/llz3210kc2fmidfizx68k0hl8qzb5nzp-jetbrainsjdk-152b1248.6/jreIt is recommended to run <oXygen/> XML Editor on Linux with a Java Virtual Machine1.6.0_05 or later from Oracle." hehe
<samueldr> :)
<joepie91> are you in XML hell? :P
<gchristensen> no
<LnL> aren't you still on holiday?
<LnL> what in the world are you doing with java an xml there
<gchristensen> yes :) had to come back from my no-electricity, no-cell-phone, no-wifi paradise island to let in a repair man
<gchristensen> heading back to vacation tomorrow morning
<LnL> just posted a fun store about upgrading nix
<LnL> story*
Guanin has quit [Ping timeout: 252 seconds]
lopsided98 has quit [Quit: Disconnected]
lopsided98 has joined #nixos-chat
Myrl-saki has quit [Quit: WeeChat 1.6]
<gchristensen> so I've been playing with oXygen and it is really nice. I wish a license wasn't $200.
lopsided98 has quit [Ping timeout: 240 seconds]
<gchristensen> it is so weird to me that orivej, instead of adding 2fa to their regular account, created a whole new github account to use 2fa on.