IRC chat logs for #ltsp on irc.freenode.net (webchat)


Channel log from 3 October 2015   (all times are UTC)

01:10work_alkisg is now known as alkisg
01:11
<alkisg>
vagrantc: sorry for the nbd-server-stop typo, thanks for the fix
01:13
<vagrantc>
alkisg: happy to be useful. :)
01:13
<alkisg>
So, phantomas says it's ok for an epoptes release
01:14
<vagrantc>
yeah, got the email
01:14
<alkisg>
I could try to integrate some of the patches that cyberorg and lbssousa sent, before releasing...
01:14* vagrantc is still struggling with what the best way forward for LTSP in Debian is
01:14
<alkisg>
NFS?
01:15
<vagrantc>
that's the main blocker...
01:15
encrypted swap needs some looking into
01:15
might have some time this weekend to dive in, but if i can't come to a resolution i'll still need to spend a bit of time switching the default to NBD
01:16
<alkisg>
Do you have to decide/change that now? Isn't NFS overlay support going to be fixed before stretch is released?
01:16
<vagrantc>
i struggle with uploading a package in which the default supported configuration is utterly broken...
01:16
<alkisg>
True, that... :-/
01:17
<vagrantc>
i guess it technically could depend on the jessie kernel... heh.
01:18
alkisg: the support is there for aufs in the kernel, but would have to maintain a dkms package for each kernel version
01:19
and haven't yet tested how well it works with linux 4.x
01:19
<alkisg>
vagrantc: or, persuade them to keep it compiled because of this problem?
01:19
https://lists.debian.org/debian-kernel/2014/12/msg00136.html
01:19
"I am assuming that this will cover the needs of Debian live systems, so I've dropped aufs from the Debian packaging. "
01:19
what if you tell them "no, it doesn't, let's do that when nfs is working"
01:19
...and have a link with the upstream issue for the broken nfs support
01:20
They can drop it in stretch+1 then...
01:21
<vagrantc>
i spent a bit of time hashing it over with ben (the debian kernel maintainer) and struggled for a couple days during debconf to test patches for NFS support ...
01:22
i probably need to start filing bugs on LTSP in debian to mark other bugs as blocking LTSP's bugs...
01:22
<alkisg>
For the "squashfs file over nfs" initramfs patches, while it's a good thing to have support for, I think I prefer nbd instead of that solution
01:23
<vagrantc>
i've still found NBD to be a little flaky when it comes to updating images...
01:23
<alkisg>
I didn't get that one, what do you mean?
01:24
<vagrantc>
the whole "restart nbd-server kicks off all the active clients" thing?
01:24
also, temporary networking failures seem to cause NBD to crash
01:24
<alkisg>
updating an image doesn't require server restart
01:24
<vagrantc>
but it doesn't free up the disk space from the deleted images until all the clients reboot
01:25
<alkisg>
teachers here report that "we have networking issues or unplug cables etc, but the clients then continue to work, why does epoptes thumbnails stop appearing"
01:25
(epoptes automatically disconnects after 15 seconds)
01:25
...maybe those issues have been resolved?
01:25
<vagrantc>
i seem to recall having those issues recently...
01:25
with this new install i've set up
01:26
<alkisg>
About the deleted image not being freed up, isn't that exactly what it's supposed to do?
01:26
If you could find a way to reproduce it, we could ping wouter or try if xnbd fixes it
01:26
<vagrantc>
yes, but i've run out of disk space on a few servers when frequently rebuilding images
01:27
and it's one of those mysterious "why is all the disk space taken up, when I can plainly see that there is plenty of disk space" mental frustrations
01:28
i've gotten in the habit of running "lsof | grep deleted" regularly
01:29
<alkisg>
If one is frequently rebuilding images, he should also reboot the clients frequently...
01:29
I don't find that a significant issue, I think that allowing the clients to continue to work while having already deleted the file is the correct thing to do,
01:30
except maybe for ls, df etc to be somehow more verbose and say where the empty space is
01:30
but that's another story
01:30
<vagrantc>
i agree ... but it's baffled myself and my *coworkers* on many occasions
01:30
<alkisg>
What I could like though is for mksquashfs, nbd etc to be able to produce and use diff images
01:31
Like, "compress this directory tree for me, but somehow make use of the previous image I did 5 hours ago"
01:31
Either to save compression time, or to make a second i386.img.diff file that either nbd-server could use directly, or the ltsp clients in /dev/nbd1 indirectly
01:31
<vagrantc>
i think i know how to do that...
01:32
there's mksquashfs -apend
01:32
<alkisg>
I think that doesn't compare things, it just appends them
01:32
<vagrantc>
or rather, -noappend
01:33
sure, so you generate the diff and append any changes...
01:33
<alkisg>
Like, first loop mount it read-only, then generate the .diff in /tmp/, and then unmount it and call mksquashfs -append?
01:34
*diff with rsync
01:34
<vagrantc>
something like that
01:34
kind of annoying, but would probably really speed up compression and save a lot of space
01:35
alkisg: could also do an overlay FS with the RO squashfs image, and then simply append what's in the COW
01:36
<alkisg>
For manual changes? Sure, but that doesn't reflect actual chroot or ltsp-pnp style changes, right?
01:36
I think I would prefer it if ltsp supported one small .diff image
01:37
<vagrantc>
might be possible...
01:37
oh, we could add the .diff images to the overlay layers
01:37
<alkisg>
like the concept of full backups every once in a while, and one differencial (not incremental)
01:37
Right
01:37
<vagrantc>
that'd be cool.
01:38
<alkisg>
That way the deleted images would be small, as only the differencial would be deleted
01:38
(frequently)
01:38
ltsp-update-image --quick or --full
01:39
<vagrantc>
like the premise, though do want to get a new upstream version uploaded before implementing... :)
01:39
<alkisg>
The default being quick if the .diff is under some size or with a date comparison
01:44
Btw I reported the nbd-server stop/start issue upstream in systemd, lennart says "won't fix": https://github.com/systemd/systemd/issues/1445
01:48
<vagrantc>
alkisg: i guess nbd-server should have a proper .service file, then...
01:49
<alkisg>
vagrantc: until all those services have proper .service files, a quick fix is to put the :pidfile = stanza in them
01:50
I'll put it in epoptes, it doesn't hurt...
01:53
vagrantc: so, what are your planning dates for releasing ltsp/epoptes? Do I have time to send more fixes in?
01:53
<vagrantc>
alkisg: reported a bug in Ubuntu/Debian about nbd-server start/stop ?
01:53
<alkisg>
No, I didn't
01:54
<vagrantc>
alkisg: might be worth doing, since upstream basically said "wontfix"
01:54
<alkisg>
I had spent 10 hours straight at that time, reporting other bugs with wily
01:54
so I thought I'd use my time somewhere where it would help more
01:54
<vagrantc>
alkisg: probably won't release either before late Sunday, more likely monday
01:55
<alkisg>
Cool
01:55* vagrantc has been saying that for each week for the last 3 weeks, notably
01:55
<alkisg>
For now, I'm troubleshooting a bug where thin clients don't login with lubuntu 14.04
01:55* vagrantc has just been kind of hoping the NFS issue would fix itself in newer kernels...
01:56
<alkisg>
If you delay it for 10 more days, you'll probably have some rpi2 related fixes from me as well :D
01:56
<vagrantc>
oh, nice.
01:56
<alkisg>
Someone sent me 3 of them, so I'll be playing a bit with them...
01:56
<vagrantc>
alkisg: i've got an rpi2 as well, so i could test the fixes, workarounds, etc.
01:56
<alkisg>
That's nice!
01:57
<vagrantc>
i've also got an older model rpi B (256MB ram)
01:57
<alkisg>
Basically I want the tftp folder to map what should go in the card's boot partition,
01:58
<vagrantc>
unfortunately, the boot partition needs non-free firmware bits :(
01:58
<alkisg>
and maybe an ltsp-create-bootable-image tool or something similar, to format/copy to the SD cards
01:58
<vagrantc>
notably, the GPU boots the CPU, not the other way around...
01:58
<alkisg>
Doesn't the raspbian chroot include those?
01:58
<vagrantc>
i think it's possible to include them, sure.
01:59
with rpi2, you should be able to actually run straight Debian armhf, and only pull in a handful of raspbian packages.
01:59
<alkisg>
Which one would you recommend?
02:00
If raspbian is focused to rpi, maybe they'd make a better default? For our examples/ltsp-build-client-rpi.conf sample file, that is...
02:02
<vagrantc>
honestly, probably both ... as raspbian is needed to support older rpi, and the performance gains from armv7 are probably not huge ... but the raspbian packages do lag a bit from Debian ... so i tend to prefer Debian for as much as you can get away with, and raspbian only for the necessary bits ("bootloader", kernel, etc.)
02:02
i guess you could use a hybrid approach with armel ...
02:03
<alkisg>
Well maybe we can just ship a couple of different example ltsp-build-client.conf files
02:03
<vagrantc>
which could support all of the above ... but armel takes more of a performance hit
02:03
<alkisg>
With different sources.list and default packages
02:03
<vagrantc>
yeah, i see valid use cases for all of the above
02:03
although really, raspberrypi foundation needs to get their kernel patches upstreamed !!!%#$@%@#
02:04
<alkisg>
Haha
02:04
Were you successfull in emulating a PI's boot process with qemu?
02:04* vagrantc wonders if the "bootloader" firmware is acceptible for debian non-free section
02:04
<alkisg>
I was only able to do it with some kernel that I downloaded from the internet
02:04
<vagrantc>
alkisg: can't imagine that would be feasible ... too much closed-source hardware/software
02:04
<alkisg>
...named kernel-qemu
02:05
<vagrantc>
it might just be pretending
02:05
<alkisg>
I just wanted to see a typical armhf booting, not specifically the pi
02:06
<vagrantc>
oh, i think the cubieboard is mainlined in qemu, but i haven't successfully booted with it
02:06
do have a physical cubieboard
02:06
<alkisg>
What loads the kernel and cmdline.txt file? Is something like uboot hardcoded somewhere in some ROM?
02:09
<vagrantc>
the GPU loads the binary firmware which loads the cmdline.txt and kernel and ... boots the CPU.
02:09
the rpi is a most unusual architecture
02:10
<alkisg>
ah, the binary firmware is also the "bootloader"...
02:10
can read vfat, load the kernel...
02:10
<vagrantc>
the rpi stuff is nothing like any of the other boards.
02:10
other boards look for code to execute in NAND or SD (such as u-boot), and then the code does whatever it does
02:11
<alkisg>
So cmdline.txt is completely pi-specific, not common to other boards
02:11
<vagrantc>
yes
02:12
some builds of u-boot support uEnv.txt, which sets some u-boot environment variables ... can also load and execute u-boot scripts (often .scr) which are basically just u-boot shell commands.
02:13
<alkisg>
Android also doesn't use uboot, right? So I don't think I have any device that uses it, yet...
02:14
So anyway, since integrating those patches about making LDM more verbose, I've had several regressions
02:14
E.g. now the verbose part says "user@server's password: "
02:14
...i.e. it waits for me to give it a password, while I already gave it to ldm
02:20* vagrantc was thinking of doing quick LDM and ltspfs (virtually no changes there) uploads
02:20
<vagrantc>
alkisg: but if the patches you're looking at are for LDM, i suppose i should wait
02:21
<alkisg>
vagrantc: my plan is to have LTSP in wily working, I don't know what that will involve...
02:21
(and also all other wily-related bugs at least file in launchpad, if not worked around in ltsp/sch-scripts)
02:21
<vagrantc>
alkisg: 11.04 ?
02:21
<alkisg>
*filed
02:21
15.10
02:21
<vagrantc>
alkisg: er, 16.04
02:21
oh.
02:21
<alkisg>
As a preparation for 16.04, yes
02:29
can't open display, nice lubuntu... :-/
02:29* alkisg hates xauthority issues
02:35
<alkisg>
Half of the times, the client boot process stucks at "A start jobs is running for LSB : raising network interfaces"
02:39
hey cool, systemd quickly shows that `networking start` spawned `udevadm settle`, which is still running
02:39
Once udevadm settle was done, 1 minute later, the boot process continued
02:40
(2 minutes later, the timeout period of udevadm settle)
02:43
OK lubuntu 15.10 works fine with the defaults, i.e. without LDM_DIRECTX=True (with ltsp-trunk of course)
02:43
Now to see why LDM_DIRECTX=True is now broken... :-/
02:45
<vagrantc>
but regular ubuntu works?
02:45
<alkisg>
...s/now/later/ :)
02:45
I'm not sure
02:46
I was trying with fat clients, not things
02:46
*thins
02:46
I'll check it out later on in the morning
02:46
<vagrantc>
oh wow
02:53alkisg is now known as work_alkisg
03:04vagrantc has left IRC (vagrantc!~vagrant@unaffiliated/vagrantc, Quit: leaving)
05:05
<cyberorg>
work_alkisg, i may have mentioned before, on suse we use unionfs-fuse for nfsroot since few years now
05:06
<work_alkisg>
cyberorg: ah, wait for vagrantc to show up, he's the one using nfs
05:06
Here we've developed and are used ltsp-pnp by default, so no nfs possible
05:07
*are using
05:08
<cyberorg>
also using prebuilt images solves building and compression time for everyone
05:09
<work_alkisg>
prebuilt images are not possible here, each school has its own needs, so they learn to use software center and then they go to sch-scripts and select update image from the menus
05:09
<cyberorg>
see https://github.com/openSUSE/kiwi/blob/master/modules/KIWILinuxRC.sh mountSystemUnionFS
05:09
<work_alkisg>
E.g. some need proprietary software that is not redistributable, impossible to include to prebuilt images
05:10
<cyberorg>
yup, then they have to unpack and repack the prebuilt image like https://sourceforge.net/p/kiwi-ltsp/wiki/Customizing%20Image/
05:11
<work_alkisg>
Selecting a package from software center is much easier, supported and documented
05:11
That's why we moved away from chroots
05:11
Even installing packages in chroots was broken at times, with dbus, upstart etc
05:12
If lxc allows chroots to be bootable with xorg and all, like VMs, it may make them easy again
05:12
<cyberorg>
wonder what machinectl would bring
05:14
<work_alkisg>
The main point is not to reinvent the wheel; users want to maintain a desktop machine, they should be able to do it in the traditional desktop way, install a desktop cd, use software center to install packages, have popup update messages etc etc
05:15
ltsp-pnp is good in that they not only maintain a desktop instead of a chroot, they only maintain _one_ desktop, because the server is reused as a template for the clients
05:15
So even if lxc etc are optimized, ltsp-pnp will still be better as long as it works
05:16
(and fortunately it appears more stable than chroots so far)
05:16
<cyberorg>
yup, not sure anything else will give that advantage
06:27vmlintu has joined IRC (vmlintu!~vmlintu@a91-152-200-13.elisa-laajakaista.fi)
06:44ricotz has joined IRC (ricotz!~rico@ubuntu/member/ricotz)
07:45F-GT has left IRC (F-GT!~phantom@ppp121-44-57-79.lns20.syd4.internode.on.net, Read error: Connection reset by peer)
07:46epoptes_user7 has joined IRC (epoptes_user7!58b521fb@gateway/web/freenode/ip.88.181.33.251)
07:49epoptes_user7 has left IRC (epoptes_user7!58b521fb@gateway/web/freenode/ip.88.181.33.251, Client Quit)
10:32work_alkisg is now known as alkisg
10:36fnurl has joined IRC (fnurl!650e83dc@gateway/web/freenode/ip.101.14.131.220)
10:41komunista has joined IRC (komunista!~slavko@87.244.209.121)
10:42komunista has left IRC (komunista!~slavko@87.244.209.121)
10:50fnurl has left IRC (fnurl!650e83dc@gateway/web/freenode/ip.101.14.131.220, Quit: Page closed)
11:16
<alkisg>
OK I got the 15.10 issue, http://cgit.freedesktop.org/xorg/xserver/commit/?id=cc59be38b7eff52a1d003b390f2994c73ee0b3e9
11:29adrianorg has left IRC (adrianorg!~adrianorg@179.182.75.70, Ping timeout: 260 seconds)
11:31adrianorg has joined IRC (adrianorg!~adrianorg@177.204.77.140.dynamic.adsl.gvt.net.br)
11:38riddle has left IRC (riddle!riddle@us.yunix.net, Quit: Lost terminal)
11:39riddle has joined IRC (riddle!riddle@us.yunix.net)
15:24khildin has joined IRC (khildin!~khildin@ip-62-235-42-27.dial.scarlet.be)
15:48ricotz has left IRC (ricotz!~rico@ubuntu/member/ricotz, Quit: Ex-Chat)
15:48khildin has left IRC (khildin!~khildin@ip-62-235-42-27.dial.scarlet.be, Ping timeout: 265 seconds)
16:13cor_geeks has joined IRC (cor_geeks!~cor@cpe-76-92-215-174.kc.res.rr.com)
16:14cor_geeks is now known as eadthem_cor
17:18mikkel has joined IRC (mikkel!~mikkel@mail.dlvs.dk)
17:24khildin has joined IRC (khildin!~khildin@ip-62-235-42-27.dial.scarlet.be)
17:41
<alkisg>
!arcfour
17:41
<ltsp>
arcfour: http://en.wikipedia.org/wiki/RC4 is an SSH cipher which is more than 2 times faster than the default aes128-ctr. To enable it, set LDM_SSHOPTIONS="-o Ciphers=arcfour128,chacha20-poly1305@openssh.com,aes128-ctr".
18:41andygraybeal has left IRC (andygraybeal!~andy@40.133.169.240, Read error: Connection reset by peer)
18:42andygraybeal has joined IRC (andygraybeal!~andy@40.133.169.240)
19:12khildin has left IRC (khildin!~khildin@ip-62-235-42-27.dial.scarlet.be, Remote host closed the connection)
19:40
<alkisg>
!forget arcfour
19:40
<ltsp>
The operation succeeded.
19:41
<muppis>
Wasn't usable after all?
19:41
<alkisg>
!learn arcfour as `http://en.wikipedia.org/wiki/RC4 is an SSH cipher which is more than 2 times faster than the default aes128-ctr. To enable it, set LDM_SSHOPTIONS="-o Ciphers=arcfour128". It's deprecated since Ubuntu 15.04.
19:41
<ltsp>
Error: No closing quotation
19:42mikkel has left IRC (mikkel!~mikkel@mail.dlvs.dk, Quit: Leaving)
19:42
<alkisg>
!learn arcfour as `http://en.wikipedia.org/wiki/RC4 is an SSH cipher which is more than 2 times faster than the default aes128-ctr. To enable it, set LDM_SSHOPTIONS="-o Ciphers=arcfour128". It's deprecated since Ubuntu 15.04 and it needs to be manually enabled in sshd_config.`
19:42
<ltsp>
The operation succeeded.
19:42
<alkisg>
!arcfour
19:42
<ltsp>
arcfour: http://en.wikipedia.org/wiki/RC4 is an SSH cipher which is more than 2 times faster than the default aes128-ctr. To enable it, set LDM_SSHOPTIONS="-o Ciphers=arcfour128". It's deprecated since Ubuntu 15.04 and it needs to be manually enabled in sshd_config.
19:43* alkisg ended up doing a lot of version checks lately... haven't done so since trying to make some websites compatible with IE4 :(
19:55alkisg is now known as work_alkisg
20:05vmlintu has left IRC (vmlintu!~vmlintu@a91-152-200-13.elisa-laajakaista.fi, Ping timeout: 264 seconds)
21:15vagrantc has joined IRC (vagrantc!~vagrant@unaffiliated/vagrantc)
22:38vagrantc has left IRC (vagrantc!~vagrant@unaffiliated/vagrantc, Ping timeout: 252 seconds)
23:11adrianorg has left IRC (adrianorg!~adrianorg@177.204.77.140.dynamic.adsl.gvt.net.br, Ping timeout: 272 seconds)
23:11adrianorg has joined IRC (adrianorg!~adrianorg@177.204.77.140.dynamic.adsl.gvt.net.br)
23:13gbaman has joined IRC (gbaman!~gbaman@host81-139-206-234.in-addr.btopenworld.com)
23:16eadthem_cor has left IRC (eadthem_cor!~cor@cpe-76-92-215-174.kc.res.rr.com, Quit: Leaving)