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


Channel log from 12 December 2010   (all times are UTC)

00:04Pacifica has joined #ltsp
01:21Pacifica has quit IRC
01:36alkisg has quit IRC
01:41alkisg has joined #ltsp
02:12map7_ has joined #ltsp
02:17chupacabra has quit IRC
02:40johnny has left #ltsp
02:41johnny has joined #ltsp
02:48
<alkisg>
Hmmm when I login from an ltsp client, my ~/.profile isn't sourced...
02:48johnny has left #ltsp
02:48johnny has joined #ltsp
02:51johnny has left #ltsp
02:52Gremble has joined #ltsp
02:52alkisg has quit IRC
02:57alkisg has joined #ltsp
03:26alkisg has quit IRC
03:31alkisg has joined #ltsp
03:32Trixboxer has joined #ltsp
03:35gorkhaan has quit IRC
03:40alkisg has quit IRC
03:40bobby_C has joined #ltsp
03:42alkisg has joined #ltsp
03:45bobby_C has quit IRC
03:45gorkhaan has joined #ltsp
04:04bobby_C has joined #ltsp
04:15MorningSon has quit IRC
04:21Gremble has quit IRC
04:27bobby_C has quit IRC
04:48alkisg has quit IRC
04:53alkisg has joined #ltsp
05:02alkisg has quit IRC
05:06alkisg has joined #ltsp
06:09indradg_ has quit IRC
06:09gorkhaan has quit IRC
06:27map7_ has quit IRC
07:07
<knipwim>
Is the lp:~ltsp-docwriters/ltsp/ltsp-docs-trunk going to be tagged with 5.2.5 as well?
07:45indradg_ has joined #ltsp
07:57F-GT has quit IRC
08:14F-GT has joined #ltsp
08:16
<alkisg>
Has anyone done extensive tests with nbd swapping? It seems to me that it's too unstable and it frequently crashes the clients, where if local swapping is used they never crash.
08:45bobby_C has joined #ltsp
08:46gorkhaan has joined #ltsp
09:27alkisg has quit IRC
09:34alkisg has joined #ltsp
09:38alkisg has quit IRC
09:42komunista has joined #ltsp
09:48alkisg has joined #ltsp
09:51alkisg has quit IRC
09:59alkisg has joined #ltsp
10:18
<alkisg>
Has anyone tried MOZ_DISABLE_IMAGE_OPTIMIZE=1 ./firefox in Lucid? I can't get it to work, firefox still caches pixmaps with it...
10:28alkisg has quit IRC
10:30bobby_C has quit IRC
10:35alkisg has joined #ltsp
10:52alkisg has quit IRC
10:56Trixboxer has quit IRC
10:57alkisg has joined #ltsp
11:04
<alkisg>
Hmmm the default 64mb swap seems too low for firefox pixmaps, maybe nbd swap is ok after all - trying 512 mb now..
11:05mistik1 has quit IRC
11:06
<alkisg>
Yup, firefox uses the 100mb of it by just browsing google.com/images...
11:07
(firefox running on the server, not on the client)
11:07mistik1 has joined #ltsp
11:13mistik1 has quit IRC
11:13mistik1 has joined #ltsp
11:14
<komunista>
alkisg: maybe this help you: http://slavino.sk/linux/ltsp/173-optimalizacia-prehliadacov
11:15
in slovak only
11:15
<alkisg>
np, google translate in work... :)
11:15
<komunista>
my notes about FF settings
11:15
<alkisg>
At least nbd swap seems to be stable, I was worried with all the constant crashes, but it appears that the default 64mb of swap is too low
11:16
<komunista>
alkisg: defaults 64 MB can be enough, if client has a lot of the own RAM
11:17
<alkisg>
If it has a lot of RAM then no nbd swap is needed :)
11:17
But by running firefox on the server I got 200mb of ram usage on the client (pixmaps)
11:17
That's really too much :(
11:17
<komunista>
still it create one swap file per client?
11:18
<alkisg>
There's a new "NBD_RAM_THRESHOlD" variable in lts.conf, if the client has more ram than that then it doesn't create one
11:18
(by default, NBD_SWAP=false, so no swap by default)
11:18
<komunista>
alkisg: I am suing clients with 256 MB RAM (-64 MB for VGA)
11:18
*using
11:18
and only FF (iceweasel) is installed
11:18
<alkisg>
komunista: right, then if you open 2 tabs on google.com/images, they'll probably hang...
11:19
Have you been seeing any crashes once in a while? E.g. once per week?
11:19
<komunista>
and I sometime check the used ram, but never grows more than 100 - 110 MB used
11:19
I have crashes, but without FF and only on two clients...
11:19
IMHO HW realted...
11:19
<alkisg>
Openoffice uses pixmaps too
11:19
OK
11:20chupacabra has joined #ltsp
11:21mistik1 has quit IRC
11:21mistik1 has joined #ltsp
11:25
<komunista>
alkisg: my pupils uses OOo too(writer, calc and impress), without any problems (only impress animations are slow)
11:26
<alkisg>
komunista: mine too - I'm just saying that if pixmaps exceed the available ram, then the client slows down alot,
11:26
and if they exceed the swap too, then the client hangs
11:27
<komunista>
I am using the RAM=80 option in lts.conf - I forgot exact name :-)
11:27
<alkisg>
ramperc
11:27
that usually prevents the client from crashing, but the apps do crash
11:28
<komunista>
uff, now I see - I don't use that (I switched from ubuntu to debian) ;-)
11:29* komunista is adding it :-)
11:29
<komunista>
!lts.conf
11:30
<ltspbot>
komunista: "lts.conf" :: http://manpages.ubuntu.com/lts.conf
11:33
<alkisg>
Impress is using another 40M of RAM on the client... X should check the available RAM before using pixmaps :(
11:33
So anyway the bottom line is that nbd-swap is OK and 64MB of swap is not enough for my clients :)
11:35
(so xorg reached 300mb of ram usage on the client with firefox + impress open)
11:40
<komunista>
alkisg: if I do not forgot, then in Wednesday I can see amounth of the used RAM for OOo and FF together on clients
11:42
but I have plan - setting the cacti (with SNMP proxing via ltsp server) to graphing the clients
11:42
only one problem, I have this plan some weeks, but have no time for playing with this :-(
11:58
<alkisg>
Yes it takes a lot of time as it depends on the sites you visit with firefox as well
11:59
E.g. there was a site which used 500 mb ram on the thin client (while firefox was still installed on the server)
12:04
<Lumiere>
3:1 it used flash too
12:09komunista has quit IRC
12:17alkisg has quit IRC
12:20alkisg has joined #ltsp
12:23komunista has joined #ltsp
12:37komunista has quit IRC
12:42komunista has joined #ltsp
13:34komunista has quit IRC
13:40komunista has joined #ltsp
14:04bobby_C has joined #ltsp
14:07komunista has quit IRC
14:43vagrantc has joined #ltsp
14:44
<alkisg>
Setting the ssh /proc/pid/oom_score to 15 gives much healthier crashes - Xorg exits with "resource temporarily unavailable" and no ghost processes are left for this user. Now to see where to put it so that its childs don't inherit the score...
14:45gorkhaan has quit IRC
14:47
<alkisg>
Ah it's the last one, no childs, nice...
14:52* vagrantc wonders what adventures alkisg is up to today
14:53
<alkisg>
Ah vagrantc congrats for the debian ltsp, it's quite good! Everything worked out of the box with a 128mb ram server and a 64mb ram client! :)
14:53
No need to add greek in ldm, already there, no need to remove nbd_proxy or compcache, everything fine on the first try :)
14:54
<vagrantc>
glad to hear!
14:54* alkisg spend the day thinking he had some driver problems with openchrome but it turned out to be the classic pixmaps problem, and nbd_swap=64mb was too little...
14:55
<alkisg>
With 512mb swap everything fine again
14:55
<vagrantc>
so using NBD swap?
14:55
i haven't really tested that much at all
14:55
<alkisg>
I thought it had problems, but I've done extensive tests the last 2 hours, it's rock solid
14:56
Of course when the swap memory doesn't suffice then chaos emerges, that's why I'm looking at oom_adj now...
14:56
Setting the ssh oom_adj to 15 makes crashes much "healthier"
14:57
<vmlintu>
512mb of nbd swap sounds a lot
14:57
<alkisg>
...so maybe we should add a "-o localcommand <change oom>" to the ssh already long line
14:57
Surfing with firefox on google.com/images and have 1 openoffice open resulted in 300 mb swap being used
14:57
(testing on an 128mb client)
14:58
(with no localapps of course)
15:03
<vagrantc>
egads
15:06
<alkisg>
vagrantc: how does this look? ssh localhost -o 'PermitLocalCommand=yes' -o 'LocalCommand=echo 15 > /proc/$PPID/oom_adj'
15:06
(need to be root for echo > oom_adj to work)
15:06
Can I add something like this to X95-run-x-session?
15:07
(with proper testing of course :))
15:08
<vagrantc>
do we use oom_adj anywhwere else?
15:08
<alkisg>
Nope
15:08
`man proc` says it's available since Linux 2.6.11
15:09
<vagrantc>
seems like that should be done with nbd-client root or swap
15:10
<alkisg>
I think the most important thing is killing the ssh session first
15:10
That way X closes gracefully, and the user processes exit
15:10
I don't think anything else would need to be killed or protected
15:10
<vagrantc>
wonder what's the difference between oom_adj and oom_score
15:10
<alkisg>
The first is setting, the other is reading
15:11
<vagrantc>
ah
15:11* alkisg puts up a web page to consistently crash thin clients... :D
15:11
<alkisg>
(ok for testing :D)
15:41
That line generates a list of images and an index.htm in the current folder which does crash firefox for me:
15:41
echo '<html><body>' > index.htm; i=100; while [ $i -lt 300 ]; do echo $i; echo "<img src='800x$i.jpg'><br>" >> index.htm; convert -size 800x$i radial-gradient:yellow 800x$i.jpg; i=$(($i+1)); done; echo '</body></html>' >> index.htm
15:41
With oom_adj the crash is clean, no ghost processes and the TC is stable after X restarts
15:42
<vagrantc>
sounds promising
15:43
<alkisg>
Sending a mail to the devs list to sum it up, will commit if noone objects
16:05gorkhaan has joined #ltsp
16:09
<alkisg>
Hmmm maybe we should give ssh a lower score than 15, and give 15 to localapps instead
16:09* alkisg isn't using localapps and forgets about them :)
16:17
<vagrantc>
i.e. localapps should be killed before killing the ssh connection?
16:18bobby_C has quit IRC
16:18
<alkisg>
Right
16:18
So if a user runs a local firefox, then his session won't die
16:19
Just the local firefox should usually be enough
16:19
<vagrantc>
seems reasonable
16:19
though isn't that the case as-is?
16:19
<alkisg>
No
16:19
X gets a very high score and gets killed first
16:19
<vagrantc>
er, with your oom_adj
16:20
<alkisg>
If we only set the ssh priority, then that would be killed first
16:20* vagrantc gets confused by high/low comparison, having no idea which direction they go
16:20
<alkisg>
oom_adj = 15 is the highest and it means "kill this one first"
16:20
<vagrantc>
so set them to 15 on localapps?
16:20
<alkisg>
So e.g. localapps could go to 14, and ssh to 12
16:21
<vagrantc>
ah, you *want* the ssh session to die more easily, as that leads to a clean session shutdown?
16:21
but localapps could go first
16:22
<alkisg>
Right. So, localapps die first, then if that's not enough ssh dies, and with it X exits too.
16:22
<vagrantc>
sounds reasonable
16:22
i think the $PPID variable is a bashism
16:23
though i may be thinking of something else
16:23
<alkisg>
No it isn't but it didn't work when I put it in the script
16:23
It's defined by posix if I recall well
16:23
Maybe I wrote something wrong in the x95 script, will look at it another day, it's getting late here :)
16:24
<vagrantc>
alkisg: you might be getting the shell's PPID in that case, for example
16:25
rather than ssh's PPID
16:25
er, rather than ssh being the parent
16:26Mobe_ has joined #ltsp
16:26
<alkisg>
vagrantc: this works correctly though: ssh localhost -o 'PermitLocalCommand=yes' -o 'LocalCommand=echo $PPID>/tmp/the-pid' ps -ef|grep 'ssh localhost'; cat /tmp/the-pid
16:27
I.e. if I test it from a shell, it seems to work
16:27
Maybe I did something wrong when I tried to put it in the x95 script
16:30Mobe has quit IRC
16:36alkisg has quit IRC
16:39Mobe__ has joined #ltsp
16:39pmatulis_ has joined #ltsp
16:40elias_a_ has joined #ltsp
16:41klausade has quit IRC
16:41Mobe_ has quit IRC
16:41pmatulis has quit IRC
16:41pthswork has quit IRC
16:41elias_a has quit IRC
16:41pthswork has joined #ltsp
16:41wwx has quit IRC
16:41wwx has joined #ltsp
16:42klausade has joined #ltsp
17:29gorkhaan has quit IRC
17:49johnny has joined #ltsp
19:24F-GT has quit IRC
19:40F-GT has joined #ltsp
20:22rjune has joined #ltsp
20:56johnny has left #ltsp
21:17johnny has joined #ltsp
21:30daya has joined #ltsp
22:06johnny has left #ltsp
22:06johnny has joined #ltsp
22:19johnny has left #ltsp
22:19johnny has joined #ltsp
23:06alkisg has joined #ltsp
23:16cyberorg has joined #ltsp
23:40indradg_ has quit IRC