1103414982 J * rpetre ~rpetre@82.76.1.167 1103414987 M * rpetre hello 1103415011 M * rpetre Bertl: are you around? 1103415016 M * Bertl yep! 1103415022 M * rpetre remember me? 1103415031 M * Bertl yep! 1103415048 M * rpetre i made a new kernel, 2.4.28 with 1.29 1103415075 M * Bertl excellent! 1103415082 M * rpetre tested it on another machine, works fine, on that machine, exactly the same problem 1103415094 M * rpetre with the same vserver, copied exactly 1103415128 M * Bertl please refresh my memory regarding the issue ... 1103415147 M * rpetre i had a process that doesn't die 1103415156 M * rpetre and doesn't respond on any signals 1103415161 M * Bertl the straming daemon? 1103415168 M * rpetre though it's still in the 'R' state 1103415170 M * rpetre yes 1103415192 M * Bertl okay, and the 'other' machine has that daemon running too? 1103415211 M * rpetre ok, let me rephrase: 1103415250 M * rpetre i copied the vserver over, ran it with the new kernel, the streaming daemon works as expected 1103415300 M * rpetre then i copied the kernel on the troublemaking machine, rebooted, tried again the server, the same behaviour, it hangs when killed 1103415319 M * Bertl but you can sucessfully kill it on the other machine? 1103415334 M * rpetre "hangs" == "stays in the R state with all available CPU, not responding to signals" 1103415358 M * rpetre on the original machine no, on my test machine yes 1103415386 M * rpetre (sorry if i'm not too coherent, it's kinda late here) 1103415412 M * rpetre any other tricks i can try? 1103415426 M * Bertl well, that is interesting ... because it actually means that the _machine_ is to blame (if I got that right) 1103415439 M * rpetre apparently yes 1103415447 M * Bertl which means either the userspace setup is different or the hardware has some effect on that 1103415457 M * rpetre though i haven't noticed anything with it 1103415471 M * rpetre and runs another ~10 vservers 1103415477 M * Bertl I would not suspect the harware in the first place 1103415490 M * Bertl so you could try to copy over the entire vserver 1103415495 M * rpetre i did 1103415500 M * rpetre that's the funny part 1103415507 M * Bertl hmm, kernel and vserver was copied? 1103415510 M * rpetre yup 1103415513 M * rpetre config also 1103415517 M * rpetre just in case 1103415532 M * Bertl hum hum ... that sounds weird ... 1103415580 M * rpetre the only different thing would be the herdware (though not by much) and the soundcard module 1103415627 M * rpetre i'll try changing the soundcard , but i doubt that it would make some difference 1103415655 M * rpetre maybe you have some bell ringing, cause i'm totally clueless 1103415673 M * Bertl well, maybe some 'kernel tuning' which is done on one machine, but not on the other? 1103415682 M * rpetre nope 1103415684 M * rpetre same config 1103415700 M * rpetre what kind of tuning? 1103415711 M * Bertl something like changes in /proc/sys/* 1103415729 M * rpetre neah, nothing like that 1103415865 M * rpetre even if anything is to blame to the application (which may very well be the case), i guess the kernel shouldn't worry about it 1103415905 M * Bertl yes, that is right ... 1103415920 M * Bertl can you test with that machine? (the one which is failing) 1103415937 M * rpetre what test? 1103415938 M * Bertl I mean like changing some code in the kernel, reboot, try again ... 1103415961 M * rpetre umm, not too heavily, but it's acceptable at night i guess 1103415986 M * Bertl maybe you could move the other vservers to a backup machine for some time? 1103416015 M * rpetre nope 1103416146 M * rpetre tell me what would you do to the machine, i'll worry about the availability 1103416177 M * Bertl okay, the idea would be to add certain debug statements to the kernel (in interesting places like the signaling stuff) 1103416202 M * Bertl then look what they report, after that, add a few more, recompile the kernel, reboot, test again 1103416222 M * rpetre as long as you promise not to panic my kernel so i can't login, it;s acceptable :) 1103416264 M * Bertl kernel panic should not be an issue, you can add panic=30 then the machine will reboot 30s after a panic 1103416291 M * rpetre ok then 1103416301 M * rpetre i use lilo -R anyways 1103416362 M * Bertl okay, get the patched kernel source (2.4.28-vs1.29) onto that machine ... compile it with your settings (.config) 1103416386 M * rpetre i used ccache on the compiling machine 1103416397 M * rpetre i don't have devel tools there 1103416419 M * Bertl okay, that is your part, I don't care ... 1103416441 M * Bertl you should be able to add a patch or make a change to the code and retry, how you do that, is up to you ;) 1103416472 M * rpetre obviously, sorry about that 1103416550 M * Bertl np, let me prepare a basic debug patch ... 1103416560 M * rpetre thanks 1103418405 M * Bertl we are talking about x86 machines? 1103418419 M * rpetre tup 1103418422 M * rpetre yup 1103418426 M * Bertl okay, just checking ;) 1103419124 M * Bertl okay, first test kernel compiling here .. will take a little ... 1103420483 M * Bertl okay, first patch ready and tested 1103420504 M * Bertl rpetre: http://vserver.13thfloor.at/Experimental/delta-2.4.28-vs1.29-db01.diff 1103420516 M * no_maam Bertl: hi 1103420517 M * rpetre ok 1103420522 M * Bertl hey no_maam! 1103420532 M * Bertl has been quite some time! 1103420541 M * no_maam Bertl: what kernel would you currently reconnoment for a system which should work 1103420554 M * Bertl 2.4.28 or 2.6.10-rc3 1103420557 M * no_maam just made a 2.6.10-rc1 with a vserver-patch crash during vserver startup 1103420651 M * rpetre Bertl: as far as my skills go, i presume that prints stuff into kern.log when signaling a process, right? 1103420669 M * Bertl yes, well, it prints it everywhere ... 1103420684 M * rpetre :) 1103420707 M * Bertl should show up on console and /var/log/messages 1103420712 M * rpetre what am i supposed to do? sorry, but i think i'll crash, i haven't slept last night too 1103420730 M * Bertl no problem, we can continue tomorrow if you like that 1103420736 M * no_maam Bertl: which vserver patch? 1103420743 M * Bertl 1.9.3.11 1103420750 M * no_maam ok 1103420753 M * Bertl or vs1.29 for 2.4 1103420767 M * rpetre Bertl: i'll try to take this myself as far as possible, just tell me what i can do 1103420780 M * Bertl # sleep 100 & 1103420780 M * Bertl # killall sleep 1103420780 M * Bertl !!! sys_kill 15 -> [60] 1103420780 M * Bertl !!! do_signal 15 to 93e10000[60] 1103420780 M * Bertl [1] + Terminated sleep 100 1103420797 M * Bertl (this is how a normal signal should look like) 1103420805 M * Bertl the !!! messages are the debug output 1103420838 M * Bertl after you added the patch, compiled and installed the kernel, you should start that process and try the following: 1103420848 M * Bertl a) kill the process from inside the context 1103420864 M * Bertl b) kill the process from outside the context (host) 1103420871 M * Bertl and with the following signals: 1103420917 M * Bertl 15 (TERM), 4 (ILL) , 9 (KILL) 1103420922 M * Bertl in that order ... 1103420947 M * rpetre aha 1103420949 M * Bertl collect the debug output and send it to me (after that we decide what to do next) 1103420955 M * rpetre so 6 reboots, right? 1103420968 M * Bertl nope, if it isn't killable, then just one 1103420974 M * rpetre okay 1103420998 M * rpetre we'll talk tomorrow then 1103421000 M * Bertl you have to identify the debug output (if other apps get signals) 1103421008 M * rpetre thanks a lot 1103421018 M * rpetre yeah, i can handle that 1103421019 M * Bertl you're welcome! have a good night! 1103421060 M * rpetre ... or what's left of it.. same to you, get some sleep, it's night in AT too ;) 1103421099 M * no_maam is it leagal to have devfs mounted in a vserver? 1103421133 M * Bertl legal yes ;) wise, no, because vserver-root can wipe out your machine 1103422686 Q * sladen Ping timeout: 480 seconds 1103422756 J * sladen paul@starsky.19inch.net 1103426772 Q * Thorsten Quit: Leaving 1103427495 Q * Doener|gone Ping timeout: 480 seconds 1103427788 J * Doener|gone doener@193.24.208.125 1103427965 J * DuckMaster ~Duck@dyn-83-155-1-92.ppp.tiscali.fr 1103427984 M * Bertl welcome DuckMaster! 1103428396 Q * DuckKing Ping timeout: 480 seconds 1103436600 M * DuckMaster lol 1103436612 M * DuckMaster Time to sleep for me :) 1103436619 M * Bertl night then! 1103436627 M * DuckMaster Server still working like a charm !) 1103436645 M * DuckMaster Nice uptime also :) 1103436649 M * Bertl that's the way it's supposed to be ... 1103436651 M * DuckMaster Congratulation 8) 1103436677 M * DuckMaster See 1103436685 M * DuckMaster Too much bier tonight ;) 1103436703 M * Bertl off to bed then! 1103436720 M * DuckMaster lol 1103436730 M * DuckMaster Ready to go !) 1103437052 M * Bertl night everyone ... I'm off to bed now too ... 1103437065 N * Bertl Bertl_zZ 1103437770 J * lilo ~lilo@lilo.usercloak.oftc.net 1103440102 Q * lilo Quit: 1103440288 J * lilo ~lilo@lilo.usercloak.oftc.net 1103455347 Q * DuckMaster Quit: Client exiting 1103458875 Q * Doener|gone Ping timeout: 480 seconds 1103459616 J * Doener|gone doener@193.24.208.125 1103459672 N * Doener|gone Doener 1103460791 Q * no_maam Remote host closed the connection 1103460797 J * no_maam ~erik@datenzone.de 1103461562 J * Thorsten ~Thorsten@dsl-084-057-069-231.arcor-ip.net 1103462649 J * mboman ~michael@cm48.sigma230.maxonline.com.sg 1103462863 Q * mboman Quit: 1103462883 J * mboman ~michael@cm48.sigma230.maxonline.com.sg 1103465624 N * Bertl_zZ Bertl 1103465639 M * Bertl morning folks! 1103466327 M * mboman Hi Bertl 1103466380 M * Bertl hey michael! 1103466433 M * mboman How's it going? 1103466498 M * Bertl well, most stuff is fine ... the mailing lists cause grief again ... 1103466560 M * mboman MLM problems? 1103466604 M * Bertl hopefully martin will fix it soon :/ 1103466616 A * Thorsten broke it ;-) 1103466672 M * Bertl maybe folks should send a short mail to martin to show that the ml is important ;) 1103469418 M * eyck all hail python! apps written in python are inherently bug-free! 1103469486 M * Bertl hmmm .. are you planning to rewrite the linux kernel ;) 1103469511 M * eyck I don't do python. 1103469553 M * Bertl what do you do? 1103469569 M * no_maam perhaps use java 1103469594 M * no_maam or even better c# 1103469635 M * eyck perl. 1103469656 M * Bertl is anybody interested in writing a small script (perl, awk, bash, sed) to help improve linux-vserver in kernel docu? 1103469673 M * eyck what kind of script? 1103469687 M * Bertl basically rearranging strings ... 1103469694 M * eyck hmm, ok. 1103469763 M * Bertl okay, let me explain what and how ... 1103469793 M * Bertl http://vserver.13thfloor.at/Experimental/debug.txt 1103469808 M * Bertl this is an example how the 'debug' system is documented inside the kernel 1103469818 M * Bertl (the linux-vserver debug system, that is) 1103469851 M * Bertl basically there are debug entries reachable via /proc/sys/vserver/debug_* 1103469876 M * Bertl those values are integers, and every bit is used for a bunch of messages 1103469908 M * eyck I've got nothing in proc/sys/vserver/ ;) 1103469921 M * Bertl now naturely those drift apart from the actual debug statements in the kernel (as I'm a lazy bastard and do not update them) 1103469979 M * Bertl to simplify this, I added a smart macro section, which 'compiles' the information into a separate kernel section 1103469990 M * eyck ok.. 1103470042 M * Bertl http://vserver.13thfloor.at/Experimental/vxdebug.txt 1103470053 M * Bertl this is such an output (the section contents) 1103470059 M * Bertl and it has the following format: 1103470084 M * Bertl fields are separated by CTRL-A, entries by CTRL-B 1103470102 M * Bertl (this is because newlines and similar can appear in the output too) 1103470135 M * Bertl the columns are from left to right: 1103470146 M * eyck ok... 1103470163 M * Bertl location (file,line), type, condition, format, arguments 1103470202 M * Bertl to explain the condition: all debug entries look like this 1103470261 M * Bertl vxdprintk(, , args ...) 1103470284 M * Bertl where condition can be an arbitrary expression 1103470315 M * Bertl so what has to be considered: 1103470331 M * Bertl a) the file might contain various debug outputs more than once 1103470343 M * Bertl (i.e. they have to be unified somehow) 1103470379 M * Bertl b) there are categorizeable debug messages which use a condition in the form of VXD_CBIT(ngnet, 1) 1103470404 M * Bertl (or more precisely VXD_CBIT(, ) 1103470435 M * Bertl and there are the non categorizeable debug outputs (which use a different condition) 1103470506 M * Bertl so I would like to 'auto' generate the debug.txt (plus a list of non categorized debug outputs) in two forms 1103470546 M * Bertl one should look like the one now, the other should include the location and arguments to verify that the debug output is correct 1103470610 M * Bertl any questions? 1103470975 Q * mboman Quit: One day I'll get that peer and reset HIS connection! 1103472483 Q * logger Ping timeout: 480 seconds 1103473368 M * Thorsten How can I limit the CPU power a vserver can achieve? /etc/vservers/vserver-name/rlimits/resource.hard? cpu=??? 1103473391 M * Bertl nope, you configure the scheduler parameters for that 1103473419 M * Bertl if you want to do a hard limit, you need to enable the hard scheduler too ... 1103473463 J * ni ~schodet@gw-labos-infos.efrei.fr 1103473472 M * Bertl welcome ni! 1103473476 M * ni hi 1103473498 M * Bertl anything wrong with your vds? 1103473508 M * ni yes, no file access :) 1103473528 M * Bertl means lycos is sleeping? 1103473543 M * Thorsten What is the difference between hard/?soft? limit with CPU power? 1103473584 N * are|afk _are_ 1103473602 M * ni Bertl: I suppose there is something wrong since the last change 1103473607 M * Bertl Thorsten: the 'soft' stuff basically adjusts the priorities ... so that 'gold' vservers will get 'more' cpu than 'silver' servers ... 1103473631 M * Bertl if the 'gold' doesn't use any cpu, the 'silver' gets it all 1103473661 M * Bertl with the 'hard' scheduler, a server will get no cpu when the token bucket is depleted 1103473671 M * Bertl ni: since what change? 1103473674 M * Thorsten Ah i.c., so I guess I want hard 1103473698 M * Bertl wb _are_! 1103473730 M * ni Bertl: There was a hard problem some days ago with a filer 1103473769 M * Bertl ah, yes, I read about that (something broke) 1103473790 M * ni Bertl: but now, file access is freezed for the last 30 minutes 1103473833 M * Bertl unfortunately rs is on vaccation, but brucey might be in charge ... 1103473895 M * ni I will post a message on the phpBB then 1103473960 M * Bertl aside from this detail, how is your overall impression? 1103473997 J * logger ~rs@vds.pas-mal.com 1103474009 M * ni It is running fine for one month 1103474019 M * ni without any problem 1103474030 M * Bertl any setup/config issues you encountered? 1103474054 M * ni Nope, I have basics need :) 1103474055 M * Thorsten Bertl, I created a file /etc/vservers/vserver-name/schedule but I can't find the syntax for it? 1103474074 M * ni All you need is Debian :) 1103474084 M * Bertl ni: thanks for the feedback .. you might want to add yourself to the list of happy linux-vserver users ;) 1103474100 M * ni Bertl: where is this list ? 1103474125 M * ni Bertl: But when a problem arise, it is always NFS related 1103474132 M * Bertl http://linux-vserver.org/VServer+Users 1103474163 M * ni Bertl: Yahoo ! fixed ! 1103474173 M * ni it runs again 1103474181 M * Bertl (it's not heavily used, because usually folks do not know it or are providers themselves (they have a separate page)) 1103474206 M * Bertl ni: great! 1103474227 M * Bertl (but it should probably be Lycos! not Yahoo! ;) 1103474253 M * ni I do not know how they can use phpBB for user support 1103474255 M * ni Bertl: :) 1103474961 M * Thorsten sorry, but I still can't find any documentation for this schedule file. Does anyone have an example how to hardlimit the cpu? 1103475376 M * Bertl check util-vserver-0.30.196/doc/configuration.xml 1103475389 M * Bertl you find 1103475423 M * Bertl and the options fill-rate, interval, tokens, tokens-min, tokens-max 1103475480 M * Bertl check http://linux-vserver.org/Linux-VServer-Paper-06 06.3. for details 1103476361 M * Thorsten Thanks Bertl, and the syntax is then "cpu fill-rate interval tokens ..."? 1103476526 M * Bertl as this is a 'hash' I would use key/value pairs ... 1103476593 M * Thorsten cpu fill-rate=1 interval=5 ...? 1103476647 M * Bertl check the results in /proc/virtual//sched 1103476672 M * Bertl don't know what is parsed, you would have to check with the source code ... 1103476697 M * Thorsten kk, thanks 1103478588 M * eyck back 1103478595 M * Bertl wb eyck! 1103478625 M * eyck sorry, only now I'll start reading what you wrote, I had to run. 1103478642 M * Bertl np 1103479851 M * Bertl okay, back later ... 1103479861 N * Bertl Bertl_oO 1103481695 Q * TheSeer Remote host closed the connection 1103481857 J * TheSeer ~theseer@border.office.salesemotion.net 1103483482 Q * Loki|muh Remote host closed the connection 1103483592 J * Loki|muh loki@satanix.de 1103483942 Q * Loki|muh Remote host closed the connection 1103484175 J * Loki|muh ~loki@satanix.de 1103484662 Q * Loki|muh Ping timeout: 480 seconds 1103484727 J * gaber gaber@linuxpl.net 1103484744 M * gaber hola boyz and girlz 1103484745 M * gaber :) 1103485176 J * Loki|muh ~loki@satanix.de 1103485342 Q * Loki|muh Remote host closed the connection 1103486143 J * Loki|muh loki@satanix.de 1103486204 Q * Thorsten Quit: Leaving 1103486349 J * Thorsten ~Thorsten@dsl-084-057-069-231.arcor-ip.net 1103489953 T * * http://linux-vserver.org/ | latest stable 1.29, devel 1.3.9, 1.9.3, ng8.5 1103489953 T * Bertl - 1103491021 Q * complexho Read error: Connection reset by peer 1103491320 M * no_maam hi 1103491365 Q * sannes Read error: Connection reset by peer 1103492262 M * no_maam I think I got a little problem 1103492458 M * no_maam somehow I still got problems with /dev/pts 1103492743 J * duckx ~Duck@dyn-83-155-1-92.ppp.tiscali.fr 1103493292 N * Bertl_oO Bertl 1103493317 M * Bertl hey no_maam! what issues? 1103493326 M * Bertl hola gaber! 1103493647 M * Thorsten Hi Bertl 1103493658 M * rpetre Bertl: can i have some of you time again? 1103493668 M * rpetre s/you/your/ 1103493679 M * Bertl hey Thorsten! please keep sending those patches! I really like them! 1103493691 M * Thorsten :-) 1103493705 M * Bertl hey rpetre! yes, let's continue where we left off yesterday ... 1103493728 M * rpetre well, i made the new kernel, and i made the "killings" 1103493747 M * Bertl Thorsten: and don't let the 'silence' disappoint you ... it takes some time ... and the solutions will be different, but I guess over time userspace will improve drastically ... 1103493772 M * Bertl rpetre: excellent, let's see the results then ... 1103493807 M * rpetre the funny thing is, i get nothing in /var/log/messages when using vkill from outside the vserver 1103493842 M * Bertl okay, that is possible ... did you try with 'kill' from outside? 1103493854 M * Bertl (you need to know the pid) 1103493873 M * rpetre i thought it doesn't work, hold on a sec 1103493919 M * rpetre "no such process" 1103493987 M * Bertl try chcontext --ctx 1 kill - 1103494059 M * Bertl it's okay if you get no such process ... but keep the log 1103494090 M * rpetre http://www.pastebin.com/131342 - these are from within the vserver 1103494113 M * rpetre i used grep to recover them, it's relly noisy in the log now :) 1103494180 M * Bertl hmm, okay, you have to explain the details ... what are those ips? 1103494205 M * rpetre wait, i have to translate them 1103494303 M * Thorsten Bertl, was I wrong to assume that if cat /proc/virtual/ID/sched shows: Token: 0 the vserer will have to stop? 1103494333 M * rpetre are you sure they're ips? they don't seem familiar, besides, they're too round 1103494378 M * Bertl forget it I am talking nonsense, I meant pids and I found your text above the pasting ... 1103494404 J * complexho ~yeah@217.22.85.161 1103494433 M * Bertl Thorsten: when tokens reach zero, and another token would be required (in a process of that context) the context will be 'put on hold' until the minimum number of tokens is available again ... 1103494447 M * Bertl welcome complexho! 1103494451 M * no_maam hi 1103494456 M * complexho hey Bertl 1103494475 M * no_maam Bertl: is there a special reason for mounting devpts with gid=5 so that only users from the tty group are allowed to launch screen and so 1103494479 M * complexho got my box up and running now :) had a few remote kernel booting issues 1103494488 M * gaber Bertl: hello. 1103494494 M * Bertl rpetre: okay, so do you know how the processes you list where related? might it be that there was some parent child relationship between them? 1103494506 M * gaber Bertl: I was walking with my dog :) 1103494519 M * no_maam Bertl: at least my vserver binary does that, which causes some trouble 1103494543 M * rpetre Bertl: 3904 was the first one, i guess the other are kids, but i don't know for sure 1103494548 M * Bertl no_maam: well, it really depends on your config ... typically it's a security measure of most distros 1103494557 M * no_maam Bertl: uhm 1103494574 M * no_maam Bertl: at least I was able to launch screen as an average user on every distro I had before 1103494578 M * rpetre Bertl: 3904 was the one i launched into background 1103494586 M * no_maam Bertl: if installed on a usual system 1103494590 M * Bertl rpetre: okay, so we basically killed the parent .. and now the 'kid' is dangling around ... 1103494600 M * rpetre smth like that 1103494609 M * rpetre nasty little brat ... 1103494616 M * Bertl no_maam: how do you test/try? 1103494626 M * no_maam Bertl: uh, what? 1103494665 M * Thorsten well Bertl, the token value here only decreases (whatever I enter for FillRate and Interval) and then stays constantly at 0. However the vserver still seem to work at full power. [Maybe I should come back on that topic later when there is less traffic ;-)] 1103494669 M * Bertl rpetre: okay, could you upload the output of all entries present in /proc/3908/ ? 1103494675 M * rpetre yes 1103494705 M * Bertl no_maam: you said you had issues with that setup, how do you test/get/see them? 1103494727 M * no_maam Bertl: I installed debian sid on an amd64, installed a vserver kernel, then installed gentoo 1103494732 M * no_maam Bertl: in a vserver 1103494741 M * no_maam Bertl: user erik is uid=1000, gid=1000, no other groups 1103494751 M * no_maam Bertl: and is not able to launch screen or xterm or so 1103494772 M * no_maam Bertl: after having added him to the tty group, problem is gone 1103494787 M * Bertl okay, what is in /etc/fstab? 1103494804 M * no_maam Bertl: nothing special on the gentoo 1103494820 M * no_maam Bertl: but vserver on the host-system mounts devpts with gid=5,mode=0620 1103494929 M * Bertl well, it does so on 'legacy' configs and it is a sane default, but you can easily override that for new style configs 1103494940 M * no_maam ok 1103494944 M * Bertl (legacy is this way, because it always was this way ;) 1103494996 N * Thorsten Thorsten|afk 1103495031 M * Bertl Thorsten|afk: you probably just see it sitting around 0 1103495064 M * Bertl those changes happen quite fast so you will not be able to detect it ... or your rate/interval is weird 1103495084 M * Bertl (unless you enable the hard scheduling flag and the hard scheduler, you will not get hard scheduling) 1103495142 N * Thorsten|afk Thorsten 1103495171 M * Thorsten Yes Bertl, why? And why doesn't that stop the processes? 1103495202 M * Thorsten Ah, I didn't see the last 2 answers 1103495239 M * Bertl Thorsten: so you have to 'enable' it at compile time, then 'configure' it for each vserver (as flag) 1103495322 M * Thorsten util-vserver-0.30.196/# ./configure --help | grep -ie hard 1103495331 M * Thorsten ? 1103495338 M * rpetre Bertl: http://www.pastebin.com/131349 1103495357 M * Bertl tx 1103495368 M * Bertl Thorsten: kernel compile time ... 1103495375 M * Thorsten ah 1103495415 M * Thorsten gzip -dc /proc/config.gz | grep -ie hard 1103495415 M * Thorsten CONFIG_VSERVER_HARDCPU=y 1103495415 M * Thorsten # CONFIG_VSERVER_HARDCPU_IDLE is not set 1103495464 M * Bertl okay, so it is enabled, but not for idle processes 1103495474 M * Bertl (which will be fine for your purpose) 1103495548 M * Loki|muh how can idle processes consume cpu? 1103495554 M * complexho Bertl: Would you mind if i asked you a hopefully quick question? 1103495587 M * Bertl np, go ahead ... 1103495605 M * Thorsten Mmm, configuration.html doesn't say how to enable hard limit for a vserver, does it? 1103495629 M * Bertl Loki|muh: no they can't but when no process is running the idle process is active and this _can_ take a longer than expected amount of time (without that option) 1103495646 M * Loki|muh okay, thx 1103495657 M * Bertl rpetre: okay, so it looks to me like your init process is not reaping that child (which is his now) 1103495696 M * Bertl rpetre: could it be that your init (on the host) is different on both machines or that the config is different? 1103495714 M * complexho Well I am up and running with vs-1.9.3 for the first time, with util-vserver-0.30.196. What I want to do is get a quick vps up and running to get familiar but the newvserver tool seems to have disappeared... (I am used to util-vserver-0.29) Any quick pointers? 1103495722 M * rpetre should be the same, i'll check 1103495762 M * Thorsten complexho, you can use your old installation, can't you? 1103495781 M * complexho you mean just copy the newvserver over...? Will that work ok? 1103495801 M * Bertl complexho: http://linux-vserver.org/alpha+util-vserver (look at the 'build' parts) 1103495813 M * complexho thanks will go now :) 1103495814 M * Thorsten I copied a standard linux installation and that worked 1103495815 M * rpetre Bertl: i have the same init on both machines 1103495941 N * Thorsten Thorsten|afk 1103496028 M * Bertl rpetre: on the machine which works (where the kill works) could you send a kill signal to the parent only (the first pid) and see what happens? 1103496223 M * rpetre Bertl: everybody dies 1103496253 M * Bertl okay, I'd suggest to try the following on the not working machine: 1103496275 M * Bertl a) start a new vserver with that process/daemon (if necessary reboot the machine) 1103496289 M * Bertl b) kill only the children (not the parent) 1103496302 M * Bertl c) if just the parent remains, kill it too ... 1103496311 M * rpetre okay, i understand 1103496329 M * rpetre should i use the debugging kernel? 1103496364 M * Bertl yes 1103496379 M * rpetre okay, i'll be back with the results 1103497307 M * rpetre Bertl: the kids don't want to die, as well 1103497417 M * rpetre and the fun thing is, in pstree i see only 2 processes insted of four, both children to init 1103497447 M * Bertl hmm ... 1103497483 M * rpetre i should have a parent and three kids 1103497555 M * Bertl hmm, I tend to say that this process somehow 'detects' a parameter/feature/etc of that machine and decides to change its behaviour depending on that 1103497588 M * rpetre probably 1103497590 M * Bertl and you probably would see the same issues on the other machine, if the children would get reparented to init 1103497622 M * rpetre i was thinkig to change oss for alsa, maybe something happens 1103497638 M * Bertl would probably be worth a try ... 1103497658 M * Bertl what init process do you use? 1103497665 M * rpetre ? 1103497681 M * rpetre i have debian stable on both the master and the vserver 1103497702 M * rpetre and util-vserver from backports.org 1103497775 M * Bertl okay .. would it be possible to start the daemon on the host instead of the vserver? (just for a test) 1103497778 M * rpetre the sysvinit package is version 2.86-5 1103497863 M * rpetre i don't think so, it would require upgrading a lot of packages to testing, as glibc changed 1103497884 M * Bertl hmm .. okay, was just an idea ... 1103497977 M * rpetre if it would really be required, i think i might be able to recompile the thing 1103498016 M * rpetre but that triggers a lot of hairy stuff, and i'm not ready to do that yet 1103498284 M * Bertl no need to, I just thought it might provide some insight if it is vserver related at all ... 1103498322 M * Bertl because to me it looks like that init gets the child and doesn't care about it ... 1103498530 J * sannes ~ace@home.skarby.no 1103498554 M * Bertl wb sannes! 1103498566 N * Thorsten|afk Thorsten 1103498568 M * Thorsten re 1103498665 M * Thorsten OK, Bertl if have the cpu hard limit option in my kernel and I have a schedule which seems to work /proc/virtual/ID/schedule changes, what to have have to do next? 1103498784 M * Bertl you have to enable the flag for that context, that should be sufficient 1103498820 M * Thorsten (Maybe silly question) How do I do that? 1103498879 M * Thorsten flag -> the file flag in /etc? 1103498902 M * Bertl either use the vattribute (to change it at runtime) or configure it in the vserver config 1103498903 M * Thorsten ah and sched_hard is the keyword 1103499349 M * Thorsten OK, I have this /vservers/X/schedule: 1103499350 M * Thorsten 5 1103499350 M * Thorsten 1 1103499350 M * Thorsten 50000 1103499350 M * Thorsten 100 1103499351 M * Thorsten 10000 1103499352 M * Thorsten 0 1103499362 M * Thorsten /proc/virtual/49169/sched: 1103499366 M * Thorsten Token: 9912 1103499366 M * Thorsten FillRate: 5 1103499366 M * Thorsten Interval: 1 1103499366 M * Thorsten TokensMin: 100 1103499366 M * Thorsten TokensMax: 10000 1103499368 M * Thorsten PrioBias: 0 1103499372 M * Thorsten cpu 0: 47 41 0 1103499394 M * Bertl okay, you read the docu? 1103499399 M * Bertl (I assume so ;) 1103499413 M * Bertl the interval is the time between refills 1103499420 M * Thorsten in ticks 1103499422 M * Thorsten ? 1103499434 M * Bertl and the fill rate is the amount of tokens added every interval 1103499438 M * Bertl yes, in ticks 1103499445 M * Thorsten OK so did I understand 1103499453 M * Thorsten cat /proc/virtual/49169/sched 1103499453 M * Thorsten Token: 9912 1103499456 M * Bertl and you consume one token each tick 1103499461 M * Thorsten Unchanged# 1103499464 M * Thorsten OK 1103499502 M * Thorsten I should mention that I also tried very high values for FillRate 1103499506 M * Bertl so I guess even if you try very hard you won't be able to consume 5 tokens with less than 5 simultaneously running processes 1103499536 M * Bertl where simultaneously means _at_once_ not processes in running state 1103499545 M * Thorsten so the number should increase, shouldn't it? 1103499567 M * Bertl it is clamped to the maximum ... 1103499597 M * Bertl it looks a little funny that you do not see 10k there ... 1103499626 M * Thorsten I though aslong as I'm below 10.000 it should increase? 1103499634 M * Bertl but otoh, a load of 47 suggests that at least some processes are consuming tokens ... 1103499678 M * Bertl let's try some better values for a test ... 1103499682 M * Thorsten OK 1103499694 M * Bertl set the interval to 100, the rate to 20 1103499706 M * Bertl (this will give you 20 percent cpu by default) 1103499725 M * Bertl then let's decrease the max to hmm, 500 1103499757 M * Bertl which means you can burst for about half a second 1103499771 M * Bertl and let's set the minimum to 0 for now 1103499809 M * Bertl (this disables the hysteresis) 1103499896 M * Bertl then let the context run without any specific activity ... 1103499913 M * Bertl the token setting should reach the 500 very soon 1103499971 M * Bertl starting a cpu bound process (e.g. cpuhog) inside the context should result in short burst (100% cpu) and a token decrease until 0 is reached 1103499988 M * Thorsten I have: 1103499989 M * Bertl after that the context will consume 20% in average 1103499991 M * Thorsten cat /usr/local/etc/vservers/testing/schedule 1103499991 M * Thorsten 20 1103499991 M * Thorsten 0 1103499991 M * Thorsten 100 1103499991 M * Thorsten 100 1103499993 M * Thorsten 500 1103499994 M * Thorsten 0 1103500029 M * Thorsten cat /proc/virtual/49174/sched 1103500029 M * Thorsten Token: 15 1103500029 M * Thorsten FillRate: 20 1103500029 M * Thorsten Interval: 1000 1103500029 M * Thorsten TokensMin: 100 1103500031 M * Thorsten TokensMax: 500 1103500034 M * Thorsten PrioBias: 0 1103500036 M * Thorsten cpu 0: 46 38 0 1103500042 M * mugwump 2%! 1103500045 M * Thorsten There is something wrong, why is intervall 1000? 1103500075 M * Bertl no idea, in case of doubt use vsched ... 1103500091 M * Thorsten Ah found it 1103500101 M * Bertl typo? 1103500111 M * Bertl hey Sam! 1103500118 M * Thorsten Yes, I changed Intervall to 0 instead Min ;-) 1103500128 M * mugwump Hi there Herbert 1103500352 M * Thorsten Ok, I now have: 1103500353 M * Thorsten cat /proc/virtual/49179/sched 1103500354 M * Thorsten Token: 419 1103500354 M * Thorsten FillRate: 20 1103500354 M * Thorsten Interval: 100 1103500354 M * Thorsten TokensMin: 0 1103500355 M * Thorsten TokensMax: 500 1103500357 M * Thorsten PrioBias: 0 1103500359 M * Thorsten cpu 0: 43 38 0 1103500370 M * Thorsten everything as you wished, is it? 1103500398 M * Bertl yes, except for the token value itself ;( 1103500410 M * Bertl but that is not under your control ... 1103500427 M * Thorsten Token: 407 1103500427 M * Thorsten and it seems to stay there 1103500452 M * Bertl sec 1103500477 M * Bertl what kernel/patches? 1103500514 M * Thorsten I entered the verser, now it is at 396, after a ls at 394. patch-2.6.9-vs1.9.3.diff 1103500586 M * Thorsten I read that there is some proc issue (run some procsecurity script). Do I have to resolve that before I do cpu limits? 1103500639 M * Bertl could I get another snapshot of your /proc/virtual/*/shed ? 1103500664 M * Thorsten cat /proc/virtual/49179/sched 1103500664 M * Thorsten Token: 387 1103500664 M * Thorsten FillRate: 20 1103500664 M * Thorsten Interval: 100 1103500664 M * Thorsten TokensMin: 0 1103500665 M * Thorsten TokensMax: 500 1103500667 M * Thorsten PrioBias: 0 1103500669 M * Thorsten cpu 0: 59 51 0 1103500702 M * Bertl this is x86_64? or x86? 1103500710 M * Thorsten x86 1103500758 M * Thorsten model name : AMD Athlon(TM) XP 2600+, Debian Sarge, vanialla 2.6.9 Kernel, nivida nonfree driver 1103500761 M * Bertl okay, looks like your tokens are not refilled ... 1103500765 M * Thorsten yup 1103500773 M * Bertl the cpu values are: 1103500782 M * Bertl user/sys/hold ticks